diff --git a/run.log b/run.log new file mode 100644 index 0000000000000000000000000000000000000000..fd22d41e4f08c22be8c43bc1387965507394c2b0 --- /dev/null +++ b/run.log @@ -0,0 +1,63506 @@ +[2026-03-25 22:00:55,425][mllm.models.large_language_model_local][INFO] - Initializing adapter 'agent_adapter': no initial weights provided or found; starting from scratch. +[2026-03-25 22:00:56,154][mllm.models.adapter_training_wrapper][INFO] - Adapter 'agent_adapter': initialized with fresh weights (no initial weights found). +[2026-03-25 22:00:56,161][mllm.models.large_language_model_local][INFO] - Initializing adapter 'critic_adapter': no initial weights provided or found; starting from scratch. +[2026-03-25 22:00:56,859][mllm.models.adapter_training_wrapper][INFO] - Adapter 'critic_adapter': initialized with fresh weights (no initial weights found). +[2026-03-25 22:03:06,266][__main__][INFO] - Starting iteration 0. +[2026-03-25 22:03:06,270][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 0 and human policies 1. +[2026-03-25 22:03:06,270][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:03:10,497][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls + +Given the values, you have a much higher value for hats compared to Bob, making hats the most desirable item for you. Books and balls have significantly lower values for you compared to their values for Bob, so it's best to focus solely on hats to maximize your points. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:03:19,638][__main__][INFO] - Number of regex retries in iteration 0: 1 +[2026-03-25 22:03:19,640][__main__][INFO] - agents played in iteration 0 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:03:30,337][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 37.45%, Block Peak % of device VRAM: 18.68%, ΔTime: 00:00:00 +[2026-03-25 22:03:30,360][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 37.45%, Block Peak % of device VRAM: 18.68%, ΔTime: 00:00:00 +[2026-03-25 22:03:30,382][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 37.45%, Block Peak % of device VRAM: 18.68%, ΔTime: 00:00:00 +[2026-03-25 22:03:30,405][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 37.45%, Block Peak % of device VRAM: 18.68%, ΔTime: 00:00:00 +[2026-03-25 22:03:30,405][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:03:30,406][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:03:30,990][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:03:32,552][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:03:33,056][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:03:33,562][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:03:34,059][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:03:34,555][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:03:35,052][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:03:35,548][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:03:36,049][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:03:36,548][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:03:37,050][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:03:37,545][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:03:38,043][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:03:38,535][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:03:39,029][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:03:39,523][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:03:40,017][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:03:40,512][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:03:41,008][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:03:41,501][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:03:41,996][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:03:42,490][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:03:42,983][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:03:43,476][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:03:43,971][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:03:44,467][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:03:44,966][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:03:45,461][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:03:45,960][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:03:46,455][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:03:46,953][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:03:47,447][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:03:47,944][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:03:48,437][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:03:48,932][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:03:49,426][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:03:49,921][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:03:50,416][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:03:50,910][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:03:51,404][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:03:51,898][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:03:52,392][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:03:52,888][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:03:53,382][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:03:53,877][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:03:54,372][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:03:54,873][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:03:55,374][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:03:55,873][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:03:56,367][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:03:56,860][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:03:57,353][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:03:57,846][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:03:58,338][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:03:58,831][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:03:59,323][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:03:59,817][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:04:00,311][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:04:00,805][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:04:01,300][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:04:01,793][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:04:02,286][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:04:02,781][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:04:03,274][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:04:03,768][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9576 tokens. +[2026-03-25 22:04:04,416][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 15.93%, Current % of VRAM taken: 53.38%, Block Peak % of device VRAM: 60.89%, ΔTime: 00:00:33 +[2026-03-25 22:04:05,055][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:04:05,057][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:04:05,059][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:04:05,722][__main__][INFO] - Iteration 1 took 59s (22.49% Gen, 76.39% Train). Generation: 13s, Training: 45s. Estimated remaining time: 49h 29m 22s. Estimated total time: 49h 32m 39s. Time estimates for 10 more iterations: 9m 54s, 100 more iterations: 1h 39m 5s, 500 more iterations: 8h 15m 26s. +[2026-03-25 22:04:05,724][__main__][INFO] - Starting iteration 1. +[2026-03-25 22:04:06,126][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 0 and human policies 1. +[2026-03-25 22:04:06,127][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:04:18,699][__main__][INFO] - Number of regex retries in iteration 1: 0 +[2026-03-25 22:04:18,700][__main__][INFO] - agents played in iteration 1 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:04:19,491][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 46.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:04:19,514][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 46.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:04:19,537][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 46.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:04:19,559][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 46.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:04:19,560][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:04:19,560][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:04:20,152][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:04:20,604][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:04:21,101][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:04:21,594][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:04:22,087][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:04:22,579][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:04:23,073][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:04:23,568][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:04:24,071][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:04:24,572][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:04:25,074][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:04:25,574][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:04:26,070][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:04:26,565][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:04:27,062][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:04:27,556][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:04:28,049][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:04:28,542][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:04:29,034][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:04:29,529][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:04:30,022][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:04:30,516][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:04:31,013][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:04:31,509][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:04:32,002][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:04:32,522][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:04:33,016][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:04:33,510][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:04:34,005][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:04:34,498][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:04:34,992][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:04:35,487][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:04:35,983][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:04:36,478][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:04:36,972][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:04:37,470][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:04:37,968][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:04:38,462][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:04:38,956][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:04:39,451][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:04:39,944][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:04:40,436][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:04:40,930][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:04:41,423][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:04:41,916][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:04:42,409][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:04:42,903][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:04:43,399][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:04:43,891][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:04:44,385][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:04:44,879][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:04:45,372][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:04:45,865][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:04:46,364][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:04:46,865][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:04:47,361][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:04:47,854][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:04:48,348][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:04:48,841][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:04:49,333][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:04:49,827][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:04:50,323][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:04:50,815][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:04:51,309][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:04:51,804][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9634 tokens. +[2026-03-25 22:04:52,477][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.95%, Current % of VRAM taken: 60.38%, Block Peak % of device VRAM: 61.57%, ΔTime: 00:00:32 +[2026-03-25 22:04:53,600][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:04:53,602][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:04:53,604][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:04:54,262][__main__][INFO] - Iteration 2 took 48s (26.12% Gen, 72.51% Train). Generation: 12s, Training: 34s. Estimated remaining time: 40h 2m 44s. Estimated total time: 40h 6m 50s. Time estimates for 10 more iterations: 8m 1s, 100 more iterations: 1h 20m 13s, 500 more iterations: 6h 41m 8s. +[2026-03-25 22:04:54,265][__main__][INFO] - Starting iteration 2. +[2026-03-25 22:04:54,666][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 0 and human policies 1. +[2026-03-25 22:04:54,666][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:05:00,346][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls + +This proposal takes advantage of the high value you have for hats and the low value Bob has for hats, assuming that he will not propose to take all of them. Since Bob values books more, he is likely to propose to take more books and balls. By proposing to take all the hats, I maximize my share of items with the highest value to me while minimizing the risk of splitting items with lower value to me. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:05:02,861][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls + +Given the per-item values, hats are most valuable for you, followed by books, and balls are least valuable. Since Bob values books more than you do and hats less than you do, taking all the hats offers you the highest immediate value. With 10 hats and your per-item value of 10 for hats, this gives you a total of 100 points from hats alone. This proposal puts all the risk on allocation proportional to the quantity if the total sum exceeds the item quantity, but since you are taking all the hats, this risk is minimal. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:05:08,463][__main__][INFO] - Number of regex retries in iteration 2: 2 +[2026-03-25 22:05:08,464][__main__][INFO] - agents played in iteration 2 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:05:09,338][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:05:09,361][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:05:09,385][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:05:09,408][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:05:09,408][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:05:09,409][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:05:10,044][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:05:10,499][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:05:11,000][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:05:11,497][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:05:11,993][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:05:12,498][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:05:13,000][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:05:13,498][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:05:13,993][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:05:14,490][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:05:14,986][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:05:15,479][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:05:15,972][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:05:16,468][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:05:16,963][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:05:17,457][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:05:17,951][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:05:18,446][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:05:18,940][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:05:19,437][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:05:19,933][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:05:20,438][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:05:20,935][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:05:21,430][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:05:21,925][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:05:22,418][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:05:22,911][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:05:23,409][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:05:23,904][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:05:24,399][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:05:24,894][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:05:25,388][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:05:25,882][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:05:26,376][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:05:26,871][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:05:27,365][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:05:27,862][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:05:28,356][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:05:28,849][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:05:29,343][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:05:29,838][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:05:30,332][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:05:30,825][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:05:31,318][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:05:31,812][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:05:32,308][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:05:32,801][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:05:33,296][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:05:33,790][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:05:34,284][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:05:34,780][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:05:35,275][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:05:35,768][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:05:36,264][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:05:36,760][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:05:37,262][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:05:37,759][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:05:38,255][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:05:38,748][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:05:39,242][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:05:39,736][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:05:40,230][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:05:40,725][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:05:41,218][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:05:41,712][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9657 tokens. +[2026-03-25 22:05:42,397][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.92%, Current % of VRAM taken: 60.35%, Block Peak % of device VRAM: 61.65%, ΔTime: 00:00:32 +[2026-03-25 22:05:43,113][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:05:43,115][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:05:43,117][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:05:43,753][__main__][INFO] - Iteration 3 took 49s (28.11% Gen, 70.59% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 49m 29s. Estimated total time: 40h 54m 24s. Time estimates for 10 more iterations: 8m 10s, 100 more iterations: 1h 21m 48s, 500 more iterations: 6h 49m 4s. +[2026-03-25 22:05:43,755][__main__][INFO] - Starting iteration 3. +[2026-03-25 22:05:44,157][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 0 and human policies 1. +[2026-03-25 22:05:44,158][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:05:46,949][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:05:57,326][__main__][INFO] - Number of regex retries in iteration 3: 1 +[2026-03-25 22:05:57,327][__main__][INFO] - agents played in iteration 3 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:05:58,200][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:05:58,223][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:05:58,246][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:05:58,269][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:05:58,269][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:05:58,270][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:05:58,880][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:05:59,331][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:05:59,829][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:06:00,322][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:06:00,819][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:06:01,318][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:06:01,816][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:06:02,312][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:06:02,809][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:06:03,302][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:06:03,799][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:06:04,295][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:06:04,795][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:06:05,289][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:06:05,786][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:06:06,284][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:06:06,781][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:06:07,294][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:06:07,788][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:06:08,284][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:06:08,779][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:06:09,276][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:06:09,770][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:06:10,265][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:06:10,759][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:06:11,254][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:06:11,748][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:06:12,242][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:06:12,736][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:06:13,229][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:06:13,721][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:06:14,214][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:06:14,706][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:06:15,200][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:06:15,697][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:06:16,194][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:06:16,689][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:06:17,184][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:06:17,678][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:06:18,173][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:06:18,669][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:06:19,164][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:06:19,660][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:06:20,156][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:06:20,652][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:06:21,148][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:06:21,643][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:06:22,137][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:06:22,631][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:06:23,126][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:06:23,620][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:06:24,113][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:06:24,608][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:06:25,102][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:06:25,597][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:06:26,091][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:06:26,586][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:06:27,079][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:06:27,575][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:06:28,071][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:06:28,571][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:06:29,070][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:06:29,567][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:06:30,062][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:06:30,556][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9662 tokens. +[2026-03-25 22:06:31,235][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 10.91%, Current % of VRAM taken: 57.33%, Block Peak % of device VRAM: 61.56%, ΔTime: 00:00:32 +[2026-03-25 22:06:31,975][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:06:31,977][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:06:31,978][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:06:32,645][__main__][INFO] - Iteration 4 took 48s (27.16% Gen, 71.46% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 18m 43s. Estimated total time: 40h 24m 27s. Time estimates for 10 more iterations: 8m 4s, 100 more iterations: 1h 20m 48s, 500 more iterations: 6h 44m 4s. +[2026-03-25 22:06:32,648][__main__][INFO] - Starting iteration 4. +[2026-03-25 22:06:33,048][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 0 and human policies 1. +[2026-03-25 22:06:33,049][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:06:33,707][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:06:43,816][mllm.models.large_language_model_local][WARNING] - Response Proposal: 5 hats, 0 books, 5 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:06:45,915][__main__][INFO] - Number of regex retries in iteration 4: 2 +[2026-03-25 22:06:45,916][__main__][INFO] - agents played in iteration 4 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:06:46,773][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 50.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:06:46,796][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 50.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:06:46,819][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 50.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:06:46,842][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 50.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:06:46,842][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:06:46,843][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:06:47,482][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:06:47,935][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:06:48,435][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:06:48,931][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:06:49,436][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:06:49,936][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:06:50,436][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:06:50,935][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:06:51,432][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:06:51,927][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:06:52,420][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:06:52,914][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:06:53,409][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:06:53,904][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:06:54,400][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:06:54,894][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:06:55,388][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:06:55,885][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:06:56,380][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:06:56,876][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:06:57,371][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:06:57,865][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:06:58,360][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:06:58,854][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:06:59,348][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:06:59,841][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:07:00,341][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:07:00,833][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:07:01,337][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:07:01,835][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:07:02,331][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:07:02,830][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:07:03,323][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:07:03,819][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:07:04,315][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:07:04,810][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:07:05,308][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:07:05,803][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:07:06,299][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:07:06,793][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:07:07,289][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:07:07,782][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:07:08,275][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:07:08,771][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:07:09,266][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:07:09,760][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:07:10,254][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:07:10,750][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:07:11,246][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:07:11,740][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:07:12,235][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:07:12,730][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:07:13,224][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:07:13,718][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:07:14,215][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:07:14,711][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:07:15,204][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:07:15,698][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:07:16,194][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:07:16,689][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:07:17,183][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:07:17,678][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:07:18,176][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:07:18,671][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:07:19,167][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9673 tokens. +[2026-03-25 22:07:19,847][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.95%, Current % of VRAM taken: 60.38%, Block Peak % of device VRAM: 61.70%, ΔTime: 00:00:32 +[2026-03-25 22:07:20,585][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:07:20,587][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:07:20,589][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:07:21,229][__main__][INFO] - Iteration 5 took 48s (26.71% Gen, 71.96% Train). Generation: 12s, Training: 34s. Estimated remaining time: 40h 2m 30s. Estimated total time: 40h 9m 2s. Time estimates for 10 more iterations: 8m 1s, 100 more iterations: 1h 20m 18s, 500 more iterations: 6h 41m 30s. +[2026-03-25 22:07:21,231][__main__][INFO] - Starting iteration 5. +[2026-03-25 22:07:21,633][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 0 and human policies 1. +[2026-03-25 22:07:21,634][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:07:34,243][__main__][INFO] - Number of regex retries in iteration 5: 0 +[2026-03-25 22:07:34,244][__main__][INFO] - agents played in iteration 5 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:07:35,054][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:07:35,077][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:07:35,100][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:07:35,123][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:07:35,123][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:07:35,124][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:07:35,744][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:07:36,197][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:07:36,697][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:07:37,202][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:07:37,697][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:07:38,195][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:07:38,700][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:07:39,203][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:07:39,707][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:07:40,204][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:07:40,700][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:07:41,195][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:07:41,691][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:07:42,185][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:07:42,680][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:07:43,175][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:07:43,669][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:07:44,162][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:07:44,655][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:07:45,151][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:07:45,644][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:07:46,138][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:07:46,632][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:07:47,126][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:07:47,620][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:07:48,115][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:07:48,610][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:07:49,105][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:07:49,601][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:07:50,102][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:07:50,598][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:07:51,093][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:07:51,746][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:07:52,297][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:07:52,789][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:07:53,282][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:07:53,776][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:07:54,270][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:07:54,762][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:07:55,257][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:07:55,750][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:07:56,244][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:07:56,736][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:07:57,235][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:07:57,727][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:07:58,219][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:07:58,710][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:07:59,204][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:07:59,698][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:08:00,197][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:08:00,693][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:08:01,187][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:08:01,681][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:08:02,175][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:08:02,670][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:08:03,165][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:08:03,658][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:08:04,152][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:08:04,647][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:08:05,140][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:08:05,635][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:08:06,129][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:08:06,621][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:08:07,115][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:08:07,610][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9614 tokens. +[2026-03-25 22:08:08,309][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.95%, Current % of VRAM taken: 60.38%, Block Peak % of device VRAM: 61.84%, ΔTime: 00:00:32 +[2026-03-25 22:08:09,110][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:08:09,112][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:08:09,114][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:08:09,825][__main__][INFO] - Iteration 6 took 48s (26.16% Gen, 72.36% Train). Generation: 12s, Training: 34s. Estimated remaining time: 40h 2m 16s. Estimated total time: 40h 9m 37s. Time estimates for 10 more iterations: 8m 1s, 100 more iterations: 1h 20m 19s, 500 more iterations: 6h 41m 36s. +[2026-03-25 22:08:09,829][__main__][INFO] - Starting iteration 6. +[2026-03-25 22:08:10,271][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 0 and human policies 1. +[2026-03-25 22:08:10,272][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:08:10,988][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:08:22,712][__main__][INFO] - Number of regex retries in iteration 6: 1 +[2026-03-25 22:08:22,713][__main__][INFO] - agents played in iteration 6 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:08:23,534][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:08:23,557][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:08:23,580][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:08:23,603][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:08:23,603][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:08:23,604][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:08:24,214][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:08:24,666][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:08:25,167][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:08:25,663][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:08:26,157][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:08:26,656][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:08:27,161][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:08:27,661][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:08:28,159][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:08:28,658][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:08:29,155][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:08:29,652][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:08:30,150][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:08:30,645][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:08:31,140][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:08:31,636][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:08:32,133][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:08:32,631][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:08:33,127][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:08:33,624][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:08:34,120][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:08:34,618][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:08:35,112][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:08:35,610][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:08:36,105][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:08:36,600][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:08:37,094][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:08:37,589][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:08:38,083][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:08:38,578][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:08:39,074][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:08:39,573][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:08:40,071][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:08:40,565][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:08:41,058][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:08:41,553][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:08:42,049][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:08:42,545][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:08:43,041][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:08:43,535][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:08:44,028][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:08:44,522][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:08:45,016][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:08:45,511][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:08:46,007][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:08:46,504][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:08:47,000][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:08:47,495][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:08:47,993][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:08:48,493][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:08:48,988][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:08:49,483][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:08:49,977][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:08:50,474][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:08:50,970][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:08:51,466][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:08:51,960][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:08:52,454][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:08:52,948][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:08:53,442][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:08:53,937][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:08:54,431][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:08:54,925][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:08:55,419][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:08:55,912][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9612 tokens. +[2026-03-25 22:08:56,589][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.95%, Current % of VRAM taken: 60.37%, Block Peak % of device VRAM: 61.48%, ΔTime: 00:00:32 +[2026-03-25 22:09:04,045][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:09:04,047][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:09:04,050][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:09:05,434][__main__][INFO] - Iteration 7 took 55s (22.55% Gen, 74.93% Train). Generation: 12s, Training: 41s. Estimated remaining time: 45h 49m 52s. Estimated total time: 45h 58m 9s. Time estimates for 10 more iterations: 9m 11s, 100 more iterations: 1h 31m 56s, 500 more iterations: 7h 39m 41s. +[2026-03-25 22:09:05,436][__main__][INFO] - Starting iteration 7. +[2026-03-25 22:09:05,838][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 0 and human policies 1. +[2026-03-25 22:09:05,839][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:09:18,773][__main__][INFO] - Number of regex retries in iteration 7: 0 +[2026-03-25 22:09:18,774][__main__][INFO] - agents played in iteration 7 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:09:19,595][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:09:19,619][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:09:19,642][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:09:19,664][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:09:19,665][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:09:19,665][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:09:20,283][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:09:20,735][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:09:21,233][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:09:21,732][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:09:22,226][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:09:22,722][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:09:23,218][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:09:23,717][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:09:24,212][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:09:24,706][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:09:25,200][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:09:25,697][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:09:26,194][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:09:26,689][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:09:27,185][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:09:27,680][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:09:28,179][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:09:28,677][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:09:29,173][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:09:29,671][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:09:30,168][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:09:30,667][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:09:31,162][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:09:31,657][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:09:32,154][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:09:32,649][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:09:33,143][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:09:33,638][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:09:34,134][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:09:34,630][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:09:35,124][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:09:35,639][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:09:36,134][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:09:36,634][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:09:37,130][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:09:37,622][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:09:38,117][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:09:38,614][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:09:39,113][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:09:39,608][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:09:40,105][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:09:40,598][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:09:41,093][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:09:41,589][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:09:42,083][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:09:42,577][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:09:43,072][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:09:43,565][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:09:44,060][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:09:44,556][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:09:45,051][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:09:45,546][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:09:46,040][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:09:46,535][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:09:47,028][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:09:47,523][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:09:48,017][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:09:48,512][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:09:49,009][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:09:49,503][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:09:49,997][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:09:50,492][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:09:50,987][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:09:51,482][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:09:51,976][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9630 tokens. +[2026-03-25 22:09:52,654][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.99%, Current % of VRAM taken: 60.41%, Block Peak % of device VRAM: 61.65%, ΔTime: 00:00:32 +[2026-03-25 22:09:53,384][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:09:53,386][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:09:53,388][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:09:54,038][__main__][INFO] - Iteration 8 took 48s (26.83% Gen, 71.81% Train). Generation: 12s, Training: 34s. Estimated remaining time: 40h 0m 57s. Estimated total time: 40h 10m 2s. Time estimates for 10 more iterations: 8m 2s, 100 more iterations: 1h 20m 20s, 500 more iterations: 6h 41m 40s. +[2026-03-25 22:09:54,040][__main__][INFO] - Starting iteration 8. +[2026-03-25 22:09:54,444][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 0 and human policies 1. +[2026-03-25 22:09:54,445][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:10:07,149][__main__][INFO] - Number of regex retries in iteration 8: 0 +[2026-03-25 22:10:07,151][__main__][INFO] - agents played in iteration 8 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:10:07,965][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:10:07,988][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:10:08,012][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:10:08,035][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:10:08,035][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:10:08,036][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:10:08,719][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:10:09,173][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:10:09,672][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:10:10,170][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:10:10,668][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:10:11,163][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:10:11,657][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:10:12,153][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:10:12,650][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:10:13,152][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:10:13,650][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:10:14,148][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:10:14,645][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:10:15,142][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:10:15,640][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:10:16,140][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:10:16,637][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:10:17,132][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:10:17,628][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:10:18,154][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:10:18,657][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:10:19,155][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:10:19,652][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:10:20,153][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:10:20,651][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:10:21,148][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:10:21,643][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:10:22,137][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:10:22,634][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:10:23,130][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:10:23,626][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:10:24,121][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:10:24,616][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:10:25,113][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:10:25,608][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:10:26,104][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:10:26,600][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:10:27,096][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:10:27,593][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:10:28,089][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:10:28,584][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:10:29,082][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:10:29,579][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:10:30,075][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:10:30,572][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:10:31,069][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:10:31,567][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:10:32,065][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:10:32,563][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:10:33,062][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:10:33,558][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:10:34,053][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:10:34,550][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:10:35,044][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:10:35,538][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:10:36,034][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:10:36,530][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:10:37,027][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:10:37,526][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:10:38,021][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:10:38,518][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:10:39,014][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:10:39,509][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:10:40,005][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:10:40,501][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9657 tokens. +[2026-03-25 22:10:41,172][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.98%, Current % of VRAM taken: 60.41%, Block Peak % of device VRAM: 61.71%, ΔTime: 00:00:32 +[2026-03-25 22:10:41,910][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:10:41,912][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:10:41,914][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:10:42,555][__main__][INFO] - Iteration 9 took 48s (26.41% Gen, 72.26% Train). Generation: 12s, Training: 34s. Estimated remaining time: 39h 55m 39s. Estimated total time: 40h 5m 33s. Time estimates for 10 more iterations: 8m 1s, 100 more iterations: 1h 20m 11s, 500 more iterations: 6h 40m 55s. +[2026-03-25 22:10:42,557][__main__][INFO] - Starting iteration 9. +[2026-03-25 22:10:42,958][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 0 and human policies 1. +[2026-03-25 22:10:42,958][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:10:55,593][__main__][INFO] - Number of regex retries in iteration 9: 0 +[2026-03-25 22:10:55,593][__main__][INFO] - agents played in iteration 9 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:10:56,431][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:10:56,455][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:10:56,478][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:10:56,501][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:10:56,502][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:10:56,503][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:10:57,114][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:10:57,567][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:10:58,068][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:10:58,567][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:10:59,064][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:10:59,561][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:11:00,061][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:11:00,558][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:11:01,055][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:11:01,551][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:11:02,048][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:11:02,542][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:11:03,037][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:11:03,534][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:11:04,031][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:11:04,529][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:11:05,025][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:11:05,521][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:11:06,017][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:11:06,512][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:11:07,008][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:11:07,502][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:11:07,997][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:11:08,493][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:11:08,987][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:11:09,481][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:11:09,974][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:11:10,469][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:11:10,963][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:11:11,457][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:11:11,952][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:11:12,448][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:11:12,941][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:11:13,437][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:11:13,933][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:11:14,427][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:11:14,922][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:11:15,416][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:11:15,912][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:11:16,406][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:11:16,900][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:11:17,398][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:11:17,894][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:11:18,390][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:11:18,886][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:11:19,383][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:11:19,878][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:11:20,375][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:11:20,873][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:11:21,370][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:11:21,865][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:11:22,361][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:11:22,855][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:11:23,352][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:11:23,851][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:11:24,349][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:11:24,844][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:11:25,338][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:11:25,831][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:11:26,325][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:11:26,819][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:11:27,315][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:11:27,809][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:11:28,303][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:11:28,797][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9684 tokens. +[2026-03-25 22:11:29,486][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.95%, Current % of VRAM taken: 60.38%, Block Peak % of device VRAM: 61.69%, ΔTime: 00:00:32 +[2026-03-25 22:11:30,228][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:11:30,231][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:11:30,233][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:11:30,890][__main__][INFO] - Iteration 10 took 47s (26.36% Gen, 72.26% Train). Generation: 12s, Training: 34s. Estimated remaining time: 39h 45m 57s. Estimated total time: 39h 56m 40s. Time estimates for 10 more iterations: 7m 59s, 100 more iterations: 1h 19m 53s, 500 more iterations: 6h 39m 26s. +[2026-03-25 22:11:30,893][__main__][INFO] - Starting iteration 10. +[2026-03-25 22:11:31,294][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 0 and human policies 1. +[2026-03-25 22:11:31,295][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:11:43,580][__main__][INFO] - Number of regex retries in iteration 10: 0 +[2026-03-25 22:11:43,581][__main__][INFO] - agents played in iteration 10 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:11:44,422][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.41%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:11:44,446][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.41%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:11:44,470][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.41%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:11:44,493][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.41%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:11:44,493][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:11:44,494][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:11:45,119][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:11:45,571][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:11:46,069][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:11:46,563][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:11:47,057][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:11:47,550][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:11:48,047][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:11:48,544][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:11:49,040][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:11:49,535][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:11:50,034][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:11:50,531][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:11:51,029][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:11:51,547][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:11:52,044][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:11:52,540][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:11:53,035][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:11:53,528][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:11:54,021][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:11:54,514][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:11:55,009][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:11:55,503][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:11:55,995][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:11:56,490][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:11:56,984][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:11:57,478][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:11:57,973][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:11:58,468][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:11:58,964][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:11:59,461][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:11:59,954][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:12:00,448][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:12:00,941][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:12:01,437][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:12:01,931][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:12:02,425][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:12:02,918][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:12:03,412][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:12:03,908][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:12:04,402][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:12:04,897][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:12:05,390][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:12:05,883][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:12:06,377][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:12:06,871][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:12:07,364][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:12:07,857][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:12:08,352][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:12:08,847][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:12:09,341][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:12:09,835][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:12:10,330][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:12:10,827][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:12:11,322][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:12:11,816][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:12:12,310][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:12:12,805][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:12:13,299][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:12:13,794][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:12:14,287][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:12:14,780][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:12:15,274][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:12:15,768][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:12:16,264][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:12:16,756][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9632 tokens. +[2026-03-25 22:12:17,422][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.93%, Current % of VRAM taken: 60.35%, Block Peak % of device VRAM: 61.66%, ΔTime: 00:00:32 +[2026-03-25 22:12:18,145][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:12:18,148][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:12:18,149][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:12:19,381][__main__][INFO] - Iteration 11 took 48s (25.55% Gen, 71.89% Train). Generation: 12s, Training: 34s. Estimated remaining time: 39h 52m 51s. Estimated total time: 40h 4m 22s. Time estimates for 10 more iterations: 8m 0s, 100 more iterations: 1h 20m 8s, 500 more iterations: 6h 40m 43s. +[2026-03-25 22:12:19,383][__main__][INFO] - Starting iteration 11. +[2026-03-25 22:12:19,782][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 1 and human policies 1. +[2026-03-25 22:12:19,783][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:12:33,009][__main__][INFO] - Number of regex retries in iteration 11: 0 +[2026-03-25 22:12:33,010][__main__][INFO] - agents played in iteration 11 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:12:33,827][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:12:33,851][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:12:33,875][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:12:33,898][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:12:33,898][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:12:33,899][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:12:34,514][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:12:34,967][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:12:35,470][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:12:35,970][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:12:36,470][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:12:36,969][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:12:37,466][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:12:37,962][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:12:38,458][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:12:38,953][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:12:39,449][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:12:39,942][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:12:40,437][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:12:40,931][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:12:41,429][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:12:41,923][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:12:42,419][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:12:42,914][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:12:43,410][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:12:43,906][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:12:44,402][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:12:44,899][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:12:45,396][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:12:45,892][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:12:46,390][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:12:46,888][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:12:47,384][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:12:47,881][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:12:48,377][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:12:48,872][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:12:49,368][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:12:49,863][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:12:50,360][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:12:50,855][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:12:51,351][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:12:51,849][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:12:52,347][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:12:52,843][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:12:53,338][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:12:53,834][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:12:54,331][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:12:54,824][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:12:55,318][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:12:55,812][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:12:56,309][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:12:56,802][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:12:57,296][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:12:57,789][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:12:58,284][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:12:58,779][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:12:59,273][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:12:59,770][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:13:00,268][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:13:00,765][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:13:01,261][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:13:01,756][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:13:02,253][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:13:02,747][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:13:03,240][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:13:03,734][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:13:04,229][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:13:04,723][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:13:05,216][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:13:05,710][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:13:06,205][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9696 tokens. +[2026-03-25 22:13:06,864][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.94%, Current % of VRAM taken: 60.37%, Block Peak % of device VRAM: 61.64%, ΔTime: 00:00:32 +[2026-03-25 22:13:07,602][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:13:07,604][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:13:07,606][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:13:08,260][__main__][INFO] - Iteration 12 took 48s (27.28% Gen, 71.36% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 11m 35s. Estimated total time: 40h 23m 54s. Time estimates for 10 more iterations: 8m 4s, 100 more iterations: 1h 20m 47s, 500 more iterations: 6h 43m 59s. +[2026-03-25 22:13:08,262][__main__][INFO] - Starting iteration 12. +[2026-03-25 22:13:08,664][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 1 and human policies 1. +[2026-03-25 22:13:08,665][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:13:21,214][__main__][INFO] - Number of regex retries in iteration 12: 0 +[2026-03-25 22:13:21,215][__main__][INFO] - agents played in iteration 12 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:13:22,033][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:13:22,057][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:13:22,081][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:13:22,104][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:13:22,104][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:13:22,105][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:13:22,725][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:13:23,177][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:13:23,676][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:13:24,172][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:13:24,672][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:13:25,171][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:13:25,666][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:13:26,163][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:13:26,660][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:13:27,157][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:13:27,653][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:13:28,149][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:13:28,645][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:13:29,141][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:13:29,637][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:13:30,132][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:13:30,628][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:13:31,123][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:13:31,620][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:13:32,117][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:13:32,616][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:13:33,114][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:13:33,609][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:13:34,104][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:13:34,604][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:13:35,100][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:13:35,597][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:13:36,094][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:13:36,589][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:13:37,085][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:13:37,582][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:13:38,077][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:13:38,573][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:13:39,067][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:13:39,561][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:13:40,055][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:13:40,551][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:13:41,045][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:13:41,538][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:13:42,032][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:13:42,529][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:13:43,025][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:13:43,518][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:13:44,013][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:13:44,509][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:13:45,002][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:13:45,496][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:13:45,990][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:13:46,485][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:13:46,979][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:13:47,475][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:13:47,971][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:13:48,465][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:13:48,957][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:13:49,452][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:13:49,946][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:13:50,443][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:13:50,937][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:13:51,432][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:13:51,927][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:13:52,423][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:13:52,917][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:13:53,412][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:13:53,907][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:13:54,403][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9644 tokens. +[2026-03-25 22:13:55,086][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.98%, Current % of VRAM taken: 60.41%, Block Peak % of device VRAM: 61.64%, ΔTime: 00:00:32 +[2026-03-25 22:13:55,826][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:13:55,828][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:13:55,830][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:13:56,525][__main__][INFO] - Iteration 13 took 47s (26.22% Gen, 72.32% Train). Generation: 12s, Training: 34s. Estimated remaining time: 39h 39m 57s. Estimated total time: 39h 53m 5s. Time estimates for 10 more iterations: 7m 58s, 100 more iterations: 1h 19m 46s, 500 more iterations: 6h 38m 50s. +[2026-03-25 22:13:56,527][__main__][INFO] - Starting iteration 13. +[2026-03-25 22:13:56,930][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 1 and human policies 1. +[2026-03-25 22:13:56,930][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:14:02,761][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:14:10,423][__main__][INFO] - Number of regex retries in iteration 13: 1 +[2026-03-25 22:14:10,424][__main__][INFO] - agents played in iteration 13 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:14:11,252][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:14:11,276][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:14:11,299][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:14:11,322][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:14:11,323][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:14:11,323][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:14:11,954][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:14:12,408][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:14:12,912][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:14:13,410][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:14:13,908][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:14:14,404][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:14:14,899][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:14:15,398][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:14:15,894][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:14:16,390][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:14:16,887][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:14:17,381][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:14:17,877][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:14:18,373][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:14:18,869][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:14:19,368][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:14:19,863][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:14:20,358][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:14:20,851][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:14:21,345][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:14:21,839][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:14:22,335][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:14:22,830][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:14:23,323][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:14:23,816][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:14:24,311][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:14:24,806][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:14:25,300][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:14:25,797][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:14:26,291][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:14:26,785][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:14:27,281][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:14:27,776][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:14:28,268][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:14:28,760][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:14:29,252][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:14:29,743][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:14:30,235][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:14:30,727][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:14:31,220][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:14:31,713][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:14:32,207][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:14:32,701][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:14:33,194][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:14:33,689][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:14:34,184][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:14:34,679][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:14:35,175][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:14:35,673][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:14:36,171][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:14:36,666][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:14:37,161][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:14:37,654][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:14:38,149][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:14:38,642][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:14:39,136][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:14:39,632][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:14:40,125][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:14:40,620][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:14:41,115][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:14:41,611][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:14:42,106][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:14:42,600][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:14:43,096][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:14:43,594][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9619 tokens. +[2026-03-25 22:14:44,277][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.05%, Current % of VRAM taken: 60.48%, Block Peak % of device VRAM: 61.69%, ΔTime: 00:00:32 +[2026-03-25 22:14:45,019][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:14:45,021][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:14:45,023][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:14:45,671][__main__][INFO] - Iteration 14 took 48s (27.68% Gen, 70.98% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 23m 8s. Estimated total time: 40h 37m 5s. Time estimates for 10 more iterations: 8m 7s, 100 more iterations: 1h 21m 14s, 500 more iterations: 6h 46m 10s. +[2026-03-25 22:14:45,673][__main__][INFO] - Starting iteration 14. +[2026-03-25 22:14:46,076][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 1 and human policies 1. +[2026-03-25 22:14:46,077][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:14:58,837][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls + +Given the per-item values: +- You value hats at 10 and books at 1, while Bob values books at 10 and balls at 10. +- To maximize your points, it's crucial to secure as many hats as possible since they offer the highest value to you. + +Since the total item quantity is 30 and you are proposing to take all 10 hats, 0 books, and 0 balls, the total sum of your proposals is 10, which is less than or equal to the item quantity. Therefore, you would receive 10 hats, 0 books, and 0 balls if this proposal is accepted. Given your per-item values, this maximizes your points for this round. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:15:01,489][__main__][INFO] - Number of regex retries in iteration 14: 1 +[2026-03-25 22:15:01,490][__main__][INFO] - agents played in iteration 14 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:15:02,318][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:15:02,342][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:15:02,366][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:15:02,389][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:15:02,389][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:15:02,390][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:15:03,015][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:15:03,468][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:15:03,966][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:15:04,463][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:15:04,956][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:15:05,453][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:15:05,948][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:15:06,442][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:15:06,934][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:15:07,429][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:15:07,923][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:15:08,416][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:15:08,915][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:15:09,410][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:15:09,904][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:15:10,398][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:15:10,892][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:15:11,388][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:15:11,882][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:15:12,377][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:15:12,873][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:15:13,369][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:15:13,866][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:15:14,361][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:15:14,855][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:15:15,352][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:15:15,845][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:15:16,337][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:15:16,831][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:15:17,324][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:15:17,818][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:15:18,310][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:15:18,804][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:15:19,297][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:15:19,791][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:15:20,287][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:15:20,781][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:15:21,274][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:15:21,768][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:15:22,260][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:15:22,754][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:15:23,249][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:15:23,744][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:15:24,240][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:15:24,736][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:15:25,259][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:15:25,756][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:15:26,253][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:15:26,751][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:15:27,251][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:15:27,749][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:15:28,242][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:15:28,735][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:15:29,230][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:15:29,723][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:15:30,216][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:15:30,712][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:15:31,205][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:15:31,698][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:15:32,191][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:15:32,685][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:15:33,176][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:15:33,670][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:15:34,163][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:15:34,656][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9688 tokens. +[2026-03-25 22:15:35,314][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.98%, Current % of VRAM taken: 60.40%, Block Peak % of device VRAM: 61.79%, ΔTime: 00:00:32 +[2026-03-25 22:15:36,052][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:15:36,054][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:15:36,056][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:15:36,703][__main__][INFO] - Iteration 15 took 50s (30.44% Gen, 68.28% Train). Generation: 15s, Training: 34s. Estimated remaining time: 41h 56m 33s. Estimated total time: 42h 11m 21s. Time estimates for 10 more iterations: 8m 26s, 100 more iterations: 1h 24m 22s, 500 more iterations: 7h 1m 53s. +[2026-03-25 22:15:36,705][__main__][INFO] - Starting iteration 15. +[2026-03-25 22:15:37,104][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 1 and human policies 1. +[2026-03-25 22:15:37,104][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:15:43,232][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:15:50,997][__main__][INFO] - Number of regex retries in iteration 15: 1 +[2026-03-25 22:15:50,998][__main__][INFO] - agents played in iteration 15 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:15:51,814][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:15:51,838][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:15:51,862][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:15:51,885][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:15:51,886][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:15:51,886][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:15:52,531][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:15:52,986][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:15:53,488][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:15:53,985][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:15:54,482][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:15:54,977][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:15:55,472][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:15:55,966][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:15:56,461][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:15:56,957][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:15:57,451][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:15:57,945][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:15:58,442][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:15:58,937][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:15:59,433][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:15:59,927][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:16:00,422][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:16:00,916][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:16:01,412][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:16:01,906][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:16:02,401][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:16:02,896][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:16:03,392][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:16:03,890][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:16:04,386][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:16:04,880][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:16:05,375][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:16:05,870][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:16:06,363][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:16:06,855][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:16:07,351][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:16:07,846][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:16:08,340][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:16:08,837][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:16:09,335][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:16:09,830][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:16:10,326][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:16:10,822][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:16:11,320][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:16:11,816][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:16:12,312][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:16:12,807][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:16:13,302][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:16:13,797][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:16:14,293][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:16:14,788][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:16:15,285][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:16:15,797][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:16:16,293][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:16:16,793][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:16:17,288][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:16:17,784][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:16:18,284][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:16:18,781][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:16:19,279][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:16:19,776][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:16:20,274][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:16:20,770][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:16:21,266][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:16:21,760][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:16:22,254][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:16:22,749][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:16:23,242][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:16:23,738][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:16:24,232][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9657 tokens. +[2026-03-25 22:16:24,922][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.98%, Current % of VRAM taken: 60.40%, Block Peak % of device VRAM: 61.61%, ΔTime: 00:00:32 +[2026-03-25 22:16:25,667][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:16:25,670][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:16:25,671][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:16:26,320][__main__][INFO] - Iteration 16 took 49s (28.23% Gen, 70.45% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 45m 14s. Estimated total time: 41h 0m 51s. Time estimates for 10 more iterations: 8m 12s, 100 more iterations: 1h 22m 1s, 500 more iterations: 6h 50m 8s. +[2026-03-25 22:16:26,323][__main__][INFO] - Starting iteration 16. +[2026-03-25 22:16:26,721][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 1 and human policies 1. +[2026-03-25 22:16:26,722][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:16:27,732][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:16:40,201][__main__][INFO] - Number of regex retries in iteration 16: 1 +[2026-03-25 22:16:40,202][__main__][INFO] - agents played in iteration 16 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:16:41,063][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:16:41,086][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:16:41,110][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:16:41,133][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:16:41,133][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:16:41,134][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:16:41,757][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:16:42,210][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:16:42,709][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:16:43,210][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:16:43,708][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:16:44,202][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:16:44,697][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:16:45,193][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:16:45,687][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:16:46,181][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:16:46,679][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:16:47,175][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:16:47,671][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:16:48,166][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:16:48,660][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:16:49,157][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:16:49,673][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:16:50,173][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:16:50,669][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:16:51,163][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:16:51,658][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:16:52,153][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:16:52,648][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:16:53,141][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:16:53,636][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:16:54,131][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:16:54,627][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:16:55,123][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:16:55,617][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:16:56,113][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:16:56,607][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:16:57,101][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:16:57,598][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:16:58,093][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:16:58,589][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:16:59,084][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:16:59,578][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:17:00,074][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:17:00,570][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:17:01,065][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:17:01,559][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:17:02,054][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:17:02,548][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:17:03,041][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:17:03,536][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:17:04,030][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:17:04,525][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:17:05,021][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:17:05,517][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:17:06,015][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:17:06,513][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:17:07,011][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:17:07,506][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:17:08,000][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:17:08,495][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:17:08,990][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:17:09,485][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:17:09,979][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:17:10,474][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:17:10,970][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:17:11,466][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:17:11,960][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:17:12,454][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:17:12,950][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:17:13,443][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9564 tokens. +[2026-03-25 22:17:14,116][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 12.93%, Current % of VRAM taken: 59.35%, Block Peak % of device VRAM: 61.39%, ΔTime: 00:00:32 +[2026-03-25 22:17:14,868][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:17:14,871][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:17:14,872][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:17:15,524][__main__][INFO] - Iteration 17 took 48s (27.62% Gen, 71.04% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 23m 41s. Estimated total time: 40h 40m 8s. Time estimates for 10 more iterations: 8m 8s, 100 more iterations: 1h 21m 20s, 500 more iterations: 6h 46m 41s. +[2026-03-25 22:17:15,526][__main__][INFO] - Starting iteration 17. +[2026-03-25 22:17:15,926][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 1 and human policies 1. +[2026-03-25 22:17:15,927][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:17:29,622][__main__][INFO] - Number of regex retries in iteration 17: 0 +[2026-03-25 22:17:29,623][__main__][INFO] - agents played in iteration 17 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:17:30,455][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:17:30,479][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:17:30,503][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:17:30,526][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:17:30,526][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:17:30,527][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:17:31,150][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:17:31,606][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:17:32,113][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:17:32,615][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:17:33,117][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:17:33,615][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:17:34,114][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:17:34,608][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:17:35,103][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:17:35,602][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:17:36,105][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:17:36,609][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:17:37,105][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:17:37,599][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:17:38,094][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:17:38,590][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:17:39,086][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:17:39,584][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:17:40,081][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:17:40,577][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:17:41,074][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:17:41,571][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:17:42,066][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:17:42,561][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:17:43,058][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:17:43,554][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:17:44,050][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:17:44,545][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:17:45,040][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:17:45,535][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:17:46,030][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:17:46,524][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:17:47,018][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:17:47,514][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:17:48,020][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:17:48,519][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:17:49,015][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:17:49,512][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:17:50,011][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:17:50,509][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:17:51,006][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:17:51,501][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:17:51,995][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:17:52,490][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:17:52,988][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:17:53,484][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:17:53,978][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:17:54,471][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:17:54,968][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:17:55,462][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:17:55,956][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:17:56,449][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:17:56,942][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:17:57,437][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:17:57,930][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:17:58,425][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:17:58,918][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:17:59,413][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:17:59,909][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:18:00,403][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:18:00,899][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:18:01,394][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:18:01,888][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:18:02,385][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:18:02,880][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9616 tokens. +[2026-03-25 22:18:03,558][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.96%, Current % of VRAM taken: 60.38%, Block Peak % of device VRAM: 61.64%, ΔTime: 00:00:32 +[2026-03-25 22:18:04,314][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:18:04,317][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:18:04,318][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:18:04,970][__main__][INFO] - Iteration 18 took 49s (27.93% Gen, 70.74% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 35m 1s. Estimated total time: 40h 52m 17s. Time estimates for 10 more iterations: 8m 10s, 100 more iterations: 1h 21m 44s, 500 more iterations: 6h 48m 42s. +[2026-03-25 22:18:04,972][__main__][INFO] - Starting iteration 18. +[2026-03-25 22:18:05,375][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 1 and human policies 1. +[2026-03-25 22:18:05,376][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:18:18,635][__main__][INFO] - Number of regex retries in iteration 18: 0 +[2026-03-25 22:18:18,635][__main__][INFO] - agents played in iteration 18 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:18:19,447][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:18:19,470][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:18:19,494][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:18:19,517][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:18:19,518][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:18:19,518][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:18:20,132][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:18:20,591][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:18:21,100][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:18:21,604][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:18:22,101][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:18:22,600][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:18:23,097][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:18:23,593][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:18:24,090][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:18:24,586][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:18:25,083][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:18:25,578][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:18:26,075][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:18:26,572][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:18:27,073][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:18:27,577][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:18:28,082][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:18:28,584][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:18:29,079][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:18:29,576][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:18:30,073][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:18:30,570][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:18:31,065][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:18:31,562][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:18:32,057][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:18:32,553][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:18:33,050][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:18:33,546][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:18:34,042][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:18:34,539][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:18:35,035][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:18:35,530][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:18:36,025][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:18:36,521][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:18:37,015][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:18:37,511][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:18:38,008][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:18:38,504][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:18:38,999][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:18:39,495][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:18:39,991][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:18:40,485][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:18:40,979][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:18:41,473][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:18:41,967][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:18:42,460][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:18:42,954][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:18:43,448][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:18:43,941][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:18:44,435][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:18:44,931][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:18:45,427][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:18:45,922][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:18:46,417][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:18:46,912][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:18:47,406][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:18:47,902][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:18:48,397][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:18:48,893][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:18:49,388][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:18:49,883][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:18:50,380][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:18:50,875][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:18:51,370][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:18:51,866][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9633 tokens. +[2026-03-25 22:18:52,524][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.99%, Current % of VRAM taken: 60.41%, Block Peak % of device VRAM: 61.58%, ΔTime: 00:00:32 +[2026-03-25 22:18:53,302][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:18:53,304][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:18:53,306][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:18:53,954][__main__][INFO] - Iteration 19 took 48s (27.29% Gen, 71.37% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 10m 53s. Estimated total time: 40h 28m 58s. Time estimates for 10 more iterations: 8m 5s, 100 more iterations: 1h 20m 57s, 500 more iterations: 6h 44m 49s. +[2026-03-25 22:18:53,956][__main__][INFO] - Starting iteration 19. +[2026-03-25 22:18:54,356][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 1 and human policies 1. +[2026-03-25 22:18:54,356][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:18:55,332][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:19:07,819][__main__][INFO] - Number of regex retries in iteration 19: 1 +[2026-03-25 22:19:07,820][__main__][INFO] - agents played in iteration 19 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:19:08,619][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:19:08,643][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:19:08,666][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:19:08,689][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:19:08,690][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:19:08,690][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:19:09,294][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:19:09,748][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:19:10,257][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:19:10,760][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:19:11,261][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:19:11,759][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:19:12,257][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:19:12,759][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:19:13,256][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:19:13,756][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:19:14,265][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:19:14,763][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:19:15,266][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:19:15,767][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:19:16,268][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:19:16,772][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:19:17,269][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:19:17,772][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:19:18,275][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:19:18,780][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:19:19,280][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:19:19,774][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:19:20,273][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:19:20,770][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:19:21,265][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:19:21,761][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:19:22,259][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:19:22,754][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:19:23,249][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:19:23,744][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:19:24,240][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:19:24,736][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:19:25,236][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:19:25,733][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:19:26,229][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:19:26,723][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:19:27,217][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:19:27,712][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:19:28,207][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:19:28,702][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:19:29,197][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:19:29,695][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:19:30,192][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:19:30,688][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:19:31,182][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:19:31,679][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:19:32,175][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:19:32,672][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:19:33,169][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:19:33,663][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:19:34,159][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:19:34,655][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:19:35,152][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:19:35,645][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:19:36,141][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:19:36,636][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:19:37,129][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:19:37,624][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:19:38,120][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:19:38,618][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:19:39,114][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:19:39,612][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:19:40,107][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:19:40,604][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:19:41,100][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9707 tokens. +[2026-03-25 22:19:41,763][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.01%, Current % of VRAM taken: 60.44%, Block Peak % of device VRAM: 61.83%, ΔTime: 00:00:32 +[2026-03-25 22:19:42,519][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:19:42,521][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:19:42,523][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:19:43,172][__main__][INFO] - Iteration 20 took 48s (27.58% Gen, 71.09% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 21m 55s. Estimated total time: 40h 40m 50s. Time estimates for 10 more iterations: 8m 8s, 100 more iterations: 1h 21m 21s, 500 more iterations: 6h 46m 48s. +[2026-03-25 22:19:43,176][__main__][INFO] - Starting iteration 20. +[2026-03-25 22:19:43,578][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 1 and human policies 1. +[2026-03-25 22:19:43,579][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:19:46,305][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:19:56,993][__main__][INFO] - Number of regex retries in iteration 20: 1 +[2026-03-25 22:19:56,994][__main__][INFO] - agents played in iteration 20 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:19:57,775][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:19:57,798][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:19:57,822][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:19:57,845][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:19:57,846][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:19:57,846][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:19:58,484][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:19:58,938][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:19:59,444][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:19:59,950][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:20:00,453][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:20:00,962][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:20:01,464][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:20:01,964][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:20:02,467][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:20:02,963][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:20:03,462][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:20:03,959][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:20:04,456][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:20:04,953][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:20:05,451][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:20:05,947][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:20:06,443][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:20:06,939][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:20:07,436][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:20:07,932][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:20:08,435][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:20:08,939][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:20:09,440][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:20:09,935][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:20:10,434][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:20:10,930][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:20:11,425][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:20:11,920][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:20:12,416][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:20:12,913][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:20:13,411][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:20:13,907][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:20:14,403][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:20:14,897][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:20:15,394][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:20:15,891][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:20:16,387][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:20:16,884][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:20:17,380][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:20:17,876][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:20:18,373][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:20:18,873][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:20:19,368][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:20:19,862][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:20:20,360][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:20:20,854][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:20:21,350][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:20:21,844][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:20:22,338][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:20:22,833][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:20:23,329][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:20:23,823][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:20:24,318][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:20:24,813][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:20:25,311][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:20:25,806][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:20:26,300][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:20:26,796][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:20:27,292][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:20:27,789][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:20:28,285][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:20:28,780][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:20:29,275][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:20:29,772][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:20:30,267][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9650 tokens. +[2026-03-25 22:20:30,945][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.98%, Current % of VRAM taken: 60.40%, Block Peak % of device VRAM: 61.69%, ΔTime: 00:00:32 +[2026-03-25 22:20:31,711][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:20:31,713][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:20:31,715][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:20:32,973][__main__][INFO] - Iteration 21 took 49s (27.16% Gen, 70.29% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 50m 2s. Estimated total time: 41h 9m 46s. Time estimates for 10 more iterations: 8m 13s, 100 more iterations: 1h 22m 19s, 500 more iterations: 6h 51m 37s. +[2026-03-25 22:20:32,976][__main__][INFO] - Starting iteration 21. +[2026-03-25 22:20:33,375][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 2 and human policies 1. +[2026-03-25 22:20:33,376][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:20:46,248][__main__][INFO] - Number of regex retries in iteration 21: 0 +[2026-03-25 22:20:46,249][__main__][INFO] - agents played in iteration 21 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:20:47,030][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:20:47,053][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:20:47,077][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:20:47,100][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:20:47,101][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:20:47,101][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:20:47,739][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:20:48,195][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:20:48,696][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:20:49,199][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:20:49,705][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:20:50,210][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:20:50,708][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:20:51,207][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:20:51,706][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:20:52,202][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:20:52,699][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:20:53,197][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:20:53,696][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:20:54,192][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:20:54,689][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:20:55,188][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:20:55,684][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:20:56,181][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:20:56,679][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:20:57,176][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:20:57,672][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:20:58,168][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:20:58,663][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:20:59,165][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:20:59,665][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:21:00,170][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:21:00,664][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:21:01,160][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:21:01,655][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:21:02,152][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:21:02,648][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:21:03,142][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:21:03,637][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:21:04,132][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:21:04,625][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:21:05,119][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:21:05,616][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:21:06,112][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:21:06,607][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:21:07,102][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:21:07,596][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:21:08,092][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:21:08,590][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:21:09,089][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:21:09,587][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:21:10,084][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:21:10,582][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:21:11,079][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:21:11,576][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:21:12,072][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:21:12,569][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:21:13,064][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:21:13,560][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:21:14,054][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:21:14,549][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:21:15,042][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:21:15,537][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:21:16,033][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:21:16,528][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:21:17,022][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:21:17,518][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:21:18,013][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:21:18,509][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:21:19,004][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:21:19,500][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9676 tokens. +[2026-03-25 22:21:20,173][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.96%, Current % of VRAM taken: 60.38%, Block Peak % of device VRAM: 61.49%, ΔTime: 00:00:32 +[2026-03-25 22:21:20,927][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:21:20,929][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:21:20,931][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:21:21,573][__main__][INFO] - Iteration 22 took 48s (26.71% Gen, 71.96% Train). Generation: 12s, Training: 34s. Estimated remaining time: 39h 49m 21s. Estimated total time: 40h 9m 53s. Time estimates for 10 more iterations: 8m 1s, 100 more iterations: 1h 20m 19s, 500 more iterations: 6h 41m 38s. +[2026-03-25 22:21:21,575][__main__][INFO] - Starting iteration 22. +[2026-03-25 22:21:21,975][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 2 and human policies 1. +[2026-03-25 22:21:21,976][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:21:35,886][__main__][INFO] - Number of regex retries in iteration 22: 0 +[2026-03-25 22:21:35,887][__main__][INFO] - agents played in iteration 22 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:21:36,705][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:21:36,728][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:21:36,752][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:21:36,775][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:21:36,775][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:21:36,776][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:21:37,390][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:21:37,843][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:21:38,348][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:21:38,843][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:21:39,349][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:21:39,853][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:21:40,357][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:21:40,856][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:21:41,355][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:21:41,854][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:21:42,351][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:21:42,847][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:21:43,343][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:21:43,838][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:21:44,335][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:21:44,832][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:21:45,328][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:21:45,823][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:21:46,319][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:21:46,815][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:21:47,311][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:21:47,807][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:21:48,303][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:21:48,799][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:21:49,297][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:21:49,794][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:21:50,298][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:21:50,801][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:21:51,303][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:21:51,800][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:21:52,297][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:21:52,798][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:21:53,298][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:21:53,796][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:21:54,292][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:21:54,792][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:21:55,290][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:21:55,788][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:21:56,286][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:21:56,780][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:21:57,275][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:21:57,785][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:21:58,281][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:21:58,778][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:21:59,273][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:21:59,770][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:22:00,267][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:22:00,762][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:22:01,258][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:22:01,754][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:22:02,250][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:22:02,744][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:22:03,240][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:22:03,735][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:22:04,230][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:22:04,725][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:22:05,221][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:22:05,717][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:22:06,213][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:22:06,711][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:22:07,207][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:22:07,700][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:22:08,196][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:22:08,693][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:22:09,190][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9732 tokens. +[2026-03-25 22:22:09,857][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.00%, Current % of VRAM taken: 60.43%, Block Peak % of device VRAM: 61.63%, ΔTime: 00:00:32 +[2026-03-25 22:22:10,618][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:22:10,621][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:22:10,622][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:22:11,279][__main__][INFO] - Iteration 23 took 49s (28.21% Gen, 70.45% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 43m 48s. Estimated total time: 41h 5m 11s. Time estimates for 10 more iterations: 8m 13s, 100 more iterations: 1h 22m 10s, 500 more iterations: 6h 50m 51s. +[2026-03-25 22:22:11,281][__main__][INFO] - Starting iteration 23. +[2026-03-25 22:22:11,682][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 2 and human policies 1. +[2026-03-25 22:22:11,683][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:22:25,259][__main__][INFO] - Number of regex retries in iteration 23: 0 +[2026-03-25 22:22:25,259][__main__][INFO] - agents played in iteration 23 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:22:26,036][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:22:26,060][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:22:26,083][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:22:26,106][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:22:26,107][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:22:26,108][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:22:26,752][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:22:27,204][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:22:27,706][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:22:28,203][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:22:28,701][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:22:29,206][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:22:29,712][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:22:30,217][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:22:30,716][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:22:31,217][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:22:31,713][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:22:32,213][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:22:32,710][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:22:33,206][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:22:33,701][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:22:34,198][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:22:34,693][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:22:35,190][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:22:35,686][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:22:36,181][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:22:36,677][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:22:37,173][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:22:37,669][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:22:38,163][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:22:38,658][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:22:39,153][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:22:39,649][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:22:40,146][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:22:40,648][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:22:41,147][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:22:41,648][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:22:42,144][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:22:42,641][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:22:43,136][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:22:43,632][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:22:44,131][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:22:44,629][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:22:45,127][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:22:45,623][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:22:46,117][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:22:46,613][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:22:47,108][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:22:47,604][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:22:48,098][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:22:48,596][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:22:49,094][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:22:49,589][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:22:50,085][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:22:50,579][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:22:51,075][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:22:51,569][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:22:52,063][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:22:52,561][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:22:53,056][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:22:53,552][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:22:54,047][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:22:54,552][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:22:55,048][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:22:55,542][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:22:56,039][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:22:56,535][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:22:57,030][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:22:57,526][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:22:58,021][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:22:58,519][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9715 tokens. +[2026-03-25 22:22:59,221][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.96%, Current % of VRAM taken: 60.39%, Block Peak % of device VRAM: 61.76%, ΔTime: 00:00:32 +[2026-03-25 22:22:59,976][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:22:59,979][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:22:59,980][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:23:00,624][__main__][INFO] - Iteration 24 took 48s (27.74% Gen, 70.94% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 24m 55s. Estimated total time: 40h 47m 7s. Time estimates for 10 more iterations: 8m 9s, 100 more iterations: 1h 21m 34s, 500 more iterations: 6h 47m 51s. +[2026-03-25 22:23:00,626][__main__][INFO] - Starting iteration 24. +[2026-03-25 22:23:01,027][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 2 and human policies 1. +[2026-03-25 22:23:01,027][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:23:03,951][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:23:14,618][__main__][INFO] - Number of regex retries in iteration 24: 1 +[2026-03-25 22:23:14,619][__main__][INFO] - agents played in iteration 24 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:23:15,395][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:23:15,419][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:23:15,442][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:23:15,469][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:23:15,470][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:23:15,471][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:23:16,162][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:23:16,615][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:23:17,117][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:23:17,615][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:23:18,118][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:23:18,622][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:23:19,124][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:23:19,621][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:23:20,119][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:23:20,619][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:23:21,117][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:23:21,614][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:23:22,110][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:23:22,605][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:23:23,102][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:23:23,598][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:23:24,095][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:23:24,590][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:23:25,085][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:23:25,582][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:23:26,078][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:23:26,574][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:23:27,069][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:23:27,563][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:23:28,057][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:23:28,551][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:23:29,045][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:23:29,539][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:23:30,034][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:23:30,533][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:23:31,034][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:23:31,532][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:23:32,030][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:23:32,525][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:23:33,027][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:23:33,524][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:23:34,020][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:23:34,517][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:23:35,012][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:23:35,507][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:23:36,005][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:23:36,502][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:23:36,999][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:23:37,498][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:23:37,994][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:23:38,493][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:23:38,987][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:23:39,483][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:23:39,979][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:23:40,476][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:23:40,976][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:23:41,472][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:23:41,968][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:23:42,463][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:23:42,958][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:23:43,454][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:23:43,949][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:23:44,445][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:23:44,940][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:23:45,435][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:23:45,931][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:23:46,428][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:23:46,925][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:23:47,421][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:23:47,919][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9707 tokens. +[2026-03-25 22:23:48,623][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 10.93%, Current % of VRAM taken: 57.36%, Block Peak % of device VRAM: 61.66%, ΔTime: 00:00:32 +[2026-03-25 22:23:49,390][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:23:49,392][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:23:49,394][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:23:50,054][__main__][INFO] - Iteration 25 took 49s (27.72% Gen, 70.93% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 28m 22s. Estimated total time: 40h 51m 23s. Time estimates for 10 more iterations: 8m 10s, 100 more iterations: 1h 21m 42s, 500 more iterations: 6h 48m 33s. +[2026-03-25 22:23:50,057][__main__][INFO] - Starting iteration 25. +[2026-03-25 22:23:50,459][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 2 and human policies 1. +[2026-03-25 22:23:50,459][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:24:03,992][__main__][INFO] - Number of regex retries in iteration 25: 0 +[2026-03-25 22:24:03,993][__main__][INFO] - agents played in iteration 25 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:24:04,773][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 50.41%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:24:04,797][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 50.41%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:24:04,821][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 50.41%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:24:04,844][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 50.41%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:24:04,845][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:24:04,845][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:24:05,512][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:24:05,964][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:24:06,471][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:24:06,969][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:24:07,466][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:24:07,971][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:24:08,480][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:24:08,982][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:24:09,480][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:24:09,979][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:24:10,475][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:24:10,970][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:24:11,465][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:24:11,959][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:24:12,455][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:24:12,952][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:24:13,451][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:24:13,949][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:24:14,451][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:24:14,948][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:24:15,447][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:24:15,944][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:24:16,440][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:24:16,936][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:24:17,433][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:24:17,929][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:24:18,423][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:24:18,919][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:24:19,415][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:24:19,911][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:24:20,408][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:24:20,904][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:24:21,406][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:24:21,909][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:24:22,406][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:24:22,900][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:24:23,399][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:24:23,894][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:24:24,390][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:24:24,884][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:24:25,379][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:24:25,875][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:24:26,372][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:24:26,866][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:24:27,362][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:24:27,859][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:24:28,354][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:24:28,850][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:24:29,345][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:24:29,841][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:24:30,337][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:24:30,831][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:24:31,325][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:24:31,819][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:24:32,315][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:24:32,814][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:24:33,310][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:24:33,806][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:24:34,302][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:24:34,797][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:24:35,292][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:24:35,787][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:24:36,283][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:24:36,779][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:24:37,277][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9633 tokens. +[2026-03-25 22:24:37,976][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.98%, Current % of VRAM taken: 60.41%, Block Peak % of device VRAM: 61.55%, ΔTime: 00:00:32 +[2026-03-25 22:24:38,737][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:24:38,739][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:24:38,741][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:24:39,405][__main__][INFO] - Iteration 26 took 48s (27.65% Gen, 70.99% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 23m 30s. Estimated total time: 40h 47m 21s. Time estimates for 10 more iterations: 8m 9s, 100 more iterations: 1h 21m 34s, 500 more iterations: 6h 47m 53s. +[2026-03-25 22:24:39,407][__main__][INFO] - Starting iteration 26. +[2026-03-25 22:24:39,809][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 2 and human policies 1. +[2026-03-25 22:24:39,810][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:24:53,955][__main__][INFO] - Number of regex retries in iteration 26: 0 +[2026-03-25 22:24:53,957][__main__][INFO] - agents played in iteration 26 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:24:54,746][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:24:54,770][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:24:54,793][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:24:54,817][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:24:54,817][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:24:54,818][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:24:55,478][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:24:55,933][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:24:56,430][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:24:56,926][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:24:57,428][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:24:57,931][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:24:58,432][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:24:58,927][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:24:59,427][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:24:59,924][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:25:00,423][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:25:00,919][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:25:01,416][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:25:01,912][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:25:02,408][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:25:02,904][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:25:03,400][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:25:03,896][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:25:04,393][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:25:04,892][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:25:05,387][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:25:05,894][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:25:06,391][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:25:06,888][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:25:07,383][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:25:07,881][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:25:08,378][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:25:08,876][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:25:09,375][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:25:09,871][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:25:10,369][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:25:10,868][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:25:11,366][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:25:11,860][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:25:12,356][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:25:12,850][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:25:13,349][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:25:13,848][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:25:14,355][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:25:14,850][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:25:15,344][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:25:15,839][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:25:16,334][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:25:16,830][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:25:17,327][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:25:17,825][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:25:18,323][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:25:18,819][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:25:19,315][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:25:19,812][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:25:20,311][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:25:20,806][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:25:21,302][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:25:21,798][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:25:22,293][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:25:22,788][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:25:23,283][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:25:23,778][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:25:24,272][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:25:24,768][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:25:25,263][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:25:25,758][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:25:26,253][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:25:26,749][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:25:27,244][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9704 tokens. +[2026-03-25 22:25:27,944][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.00%, Current % of VRAM taken: 60.43%, Block Peak % of device VRAM: 62.05%, ΔTime: 00:00:32 +[2026-03-25 22:25:28,701][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:25:28,703][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:25:28,705][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:25:29,361][__main__][INFO] - Iteration 27 took 49s (28.55% Gen, 70.12% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 52m 57s. Estimated total time: 41h 17m 37s. Time estimates for 10 more iterations: 8m 15s, 100 more iterations: 1h 22m 35s, 500 more iterations: 6h 52m 56s. +[2026-03-25 22:25:29,367][__main__][INFO] - Starting iteration 27. +[2026-03-25 22:25:29,770][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 2 and human policies 1. +[2026-03-25 22:25:29,770][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:25:30,832][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:25:36,490][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls + +Explanation: Given the per-item values, you have a significant advantage with hats and books compared to Bob. Hats are worth 10 points each to you, while Bob values them at 1 point each. Books are also valued highly by you compared to Bob. However, balls have low value for both of you. Therefore, it's in your best interest to claim all the hats to maximize your points. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:25:44,519][__main__][INFO] - Number of regex retries in iteration 27: 2 +[2026-03-25 22:25:44,520][__main__][INFO] - agents played in iteration 27 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:25:45,291][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:25:45,315][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:25:45,339][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:25:45,362][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:25:45,362][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:25:45,363][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:25:45,989][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:25:46,449][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:25:46,956][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:25:47,458][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:25:47,954][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:25:48,452][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:25:48,949][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:25:49,446][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:25:49,947][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:25:50,443][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:25:50,943][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:25:51,441][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:25:51,939][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:25:52,437][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:25:52,935][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:25:53,433][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:25:53,932][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:25:54,433][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:25:54,929][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:25:55,425][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:25:55,924][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:25:56,420][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:25:56,917][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:25:57,413][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:25:57,910][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:25:58,429][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:25:58,930][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:25:59,431][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:25:59,935][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:26:00,435][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:26:00,935][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:26:01,435][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:26:01,935][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:26:02,440][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:26:02,942][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:26:03,447][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:26:03,946][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:26:04,443][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:26:04,938][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:26:05,435][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:26:05,934][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:26:06,433][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:26:06,931][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:26:07,428][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:26:07,924][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:26:08,420][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:26:08,916][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:26:09,413][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:26:09,908][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:26:10,402][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:26:10,897][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:26:11,394][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:26:11,890][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:26:12,386][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:26:12,883][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:26:13,380][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:26:13,877][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:26:14,372][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:26:14,870][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:26:15,366][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:26:15,861][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:26:16,357][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:26:16,852][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:26:17,347][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:26:17,842][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9780 tokens. +[2026-03-25 22:26:18,535][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.95%, Current % of VRAM taken: 60.38%, Block Peak % of device VRAM: 61.83%, ΔTime: 00:00:32 +[2026-03-25 22:26:19,290][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:26:19,294][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:26:19,297][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:26:19,937][__main__][INFO] - Iteration 28 took 50s (29.40% Gen, 69.32% Train). Generation: 14s, Training: 34s. Estimated remaining time: 41h 22m 53s. Estimated total time: 41h 48m 24s. Time estimates for 10 more iterations: 8m 21s, 100 more iterations: 1h 23m 36s, 500 more iterations: 6h 58m 4s. +[2026-03-25 22:26:19,940][__main__][INFO] - Starting iteration 28. +[2026-03-25 22:26:20,340][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 2 and human policies 1. +[2026-03-25 22:26:20,341][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:26:33,935][__main__][INFO] - Number of regex retries in iteration 28: 0 +[2026-03-25 22:26:33,935][__main__][INFO] - agents played in iteration 28 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:26:34,713][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:26:34,737][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:26:34,760][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:26:34,783][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:26:34,784][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:26:34,784][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:26:35,402][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:26:35,860][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:26:36,366][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:26:36,868][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:26:37,366][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:26:37,867][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:26:38,364][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:26:38,860][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:26:39,358][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:26:39,854][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:26:40,353][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:26:40,850][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:26:41,348][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:26:41,844][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:26:42,341][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:26:42,837][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:26:43,337][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:26:43,833][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:26:44,328][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:26:44,824][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:26:45,320][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:26:45,816][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:26:46,312][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:26:46,808][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:26:47,304][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:26:47,802][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:26:48,303][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:26:48,802][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:26:49,300][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:26:49,797][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:26:50,294][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:26:50,795][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:26:51,297][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:26:51,806][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:26:52,307][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:26:52,807][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:26:53,302][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:26:53,797][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:26:54,295][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:26:54,791][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:26:55,287][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:26:55,783][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:26:56,279][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:26:56,776][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:26:57,275][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:26:57,772][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:26:58,271][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:26:58,767][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:26:59,264][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:26:59,758][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:27:00,255][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:27:00,751][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:27:01,246][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:27:01,741][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:27:02,235][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:27:02,729][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:27:03,224][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:27:03,720][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:27:04,214][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:27:04,711][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:27:05,205][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:27:05,700][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:27:06,197][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:27:06,694][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:27:07,191][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9720 tokens. +[2026-03-25 22:27:07,866][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.96%, Current % of VRAM taken: 60.38%, Block Peak % of device VRAM: 61.66%, ΔTime: 00:00:32 +[2026-03-25 22:27:08,619][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:27:08,622][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:27:08,624][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:27:09,271][__main__][INFO] - Iteration 29 took 48s (27.78% Gen, 70.89% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 20m 13s. Estimated total time: 40h 46m 34s. Time estimates for 10 more iterations: 8m 9s, 100 more iterations: 1h 21m 33s, 500 more iterations: 6h 47m 45s. +[2026-03-25 22:27:09,273][__main__][INFO] - Starting iteration 29. +[2026-03-25 22:27:09,675][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 2 and human policies 1. +[2026-03-25 22:27:09,675][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:27:14,390][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:27:23,302][__main__][INFO] - Number of regex retries in iteration 29: 1 +[2026-03-25 22:27:23,303][__main__][INFO] - agents played in iteration 29 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:27:24,066][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.41%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:27:24,090][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.41%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:27:24,113][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.41%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:27:24,136][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.41%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:27:24,137][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:27:24,137][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:27:24,783][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:27:25,240][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:27:25,752][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:27:26,258][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:27:26,756][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:27:27,255][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:27:27,752][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:27:28,250][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:27:28,747][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:27:29,242][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:27:29,740][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:27:30,235][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:27:30,731][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:27:31,226][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:27:31,723][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:27:32,219][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:27:32,715][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:27:33,211][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:27:33,709][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:27:34,206][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:27:34,703][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:27:35,198][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:27:35,695][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:27:36,191][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:27:36,686][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:27:37,182][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:27:37,679][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:27:38,176][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:27:38,673][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:27:39,173][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:27:39,668][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:27:40,165][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:27:40,661][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:27:41,159][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:27:41,659][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:27:42,160][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:27:42,660][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:27:43,155][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:27:43,651][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:27:44,147][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:27:44,641][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:27:45,137][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:27:45,633][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:27:46,129][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:27:46,624][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:27:47,120][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:27:47,617][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:27:48,113][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:27:48,610][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:27:49,105][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:27:49,601][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:27:50,101][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:27:50,597][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:27:51,093][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:27:51,588][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:27:52,082][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:27:52,577][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:27:53,075][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:27:53,574][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:27:54,071][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:27:54,568][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:27:55,065][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:27:55,562][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:27:56,061][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:27:56,558][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9734 tokens. +[2026-03-25 22:27:57,252][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.00%, Current % of VRAM taken: 60.42%, Block Peak % of device VRAM: 61.77%, ΔTime: 00:00:32 +[2026-03-25 22:27:58,010][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:27:58,012][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:27:58,014][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:27:58,666][__main__][INFO] - Iteration 30 took 48s (27.82% Gen, 70.85% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 22m 26s. Estimated total time: 40h 49m 36s. Time estimates for 10 more iterations: 8m 9s, 100 more iterations: 1h 21m 39s, 500 more iterations: 6h 48m 16s. +[2026-03-25 22:27:58,668][__main__][INFO] - Starting iteration 30. +[2026-03-25 22:27:59,068][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 2 and human policies 1. +[2026-03-25 22:27:59,069][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:28:04,619][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls + +This proposal is based on the per-item values provided. Since my value for hats is significantly higher (10) compared to Bob's (1), I will propose taking all 10 hats. For books and balls, Bob values them higher than I do, so it would not be in my best interest to propose taking any of these items. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:28:13,472][__main__][INFO] - Number of regex retries in iteration 30: 1 +[2026-03-25 22:28:13,473][__main__][INFO] - agents played in iteration 30 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:28:14,256][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:28:14,280][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:28:14,303][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:28:14,326][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:28:14,327][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:28:14,327][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:28:14,951][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:28:15,408][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:28:15,914][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:28:16,416][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:28:16,914][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:28:17,413][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:28:17,911][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:28:18,430][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:28:18,927][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:28:19,424][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:28:19,924][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:28:20,430][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:28:20,928][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:28:21,426][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:28:21,923][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:28:22,420][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:28:22,919][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:28:23,417][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:28:23,913][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:28:24,410][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:28:24,908][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:28:25,406][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:28:25,903][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:28:26,400][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:28:26,899][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:28:27,396][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:28:27,894][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:28:28,391][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:28:28,887][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:28:29,385][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:28:29,881][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:28:30,376][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:28:30,874][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:28:31,377][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:28:31,878][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:28:32,382][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:28:32,888][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:28:33,385][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:28:33,880][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:28:34,375][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:28:34,873][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:28:35,367][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:28:35,862][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:28:36,357][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:28:36,852][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:28:37,347][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:28:37,841][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:28:38,336][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:28:38,830][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:28:39,326][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:28:39,821][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:28:40,317][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:28:40,814][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:28:41,310][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:28:41,806][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:28:42,302][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:28:42,798][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:28:43,299][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:28:43,795][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:28:44,291][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:28:44,788][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:28:45,285][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:28:45,781][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:28:46,278][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:28:46,775][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9783 tokens. +[2026-03-25 22:28:47,447][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.05%, Current % of VRAM taken: 60.47%, Block Peak % of device VRAM: 61.92%, ΔTime: 00:00:32 +[2026-03-25 22:28:48,247][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:28:48,250][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:28:48,251][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:28:49,584][__main__][INFO] - Iteration 31 took 50s (28.51% Gen, 68.84% Train). Generation: 14s, Training: 34s. Estimated remaining time: 41h 37m 49s. Estimated total time: 42h 5m 49s. Time estimates for 10 more iterations: 8m 25s, 100 more iterations: 1h 24m 11s, 500 more iterations: 7h 0m 58s. +[2026-03-25 22:28:49,586][__main__][INFO] - Starting iteration 31. +[2026-03-25 22:28:49,990][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 3 and human policies 1. +[2026-03-25 22:28:49,991][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:29:03,596][__main__][INFO] - Number of regex retries in iteration 31: 0 +[2026-03-25 22:29:03,597][__main__][INFO] - agents played in iteration 31 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:29:04,375][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:29:04,399][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:29:04,422][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:29:04,445][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:29:04,445][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:29:04,446][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:29:05,083][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:29:05,541][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:29:06,055][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:29:06,558][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:29:07,059][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:29:07,559][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:29:08,057][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:29:08,554][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:29:09,051][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:29:09,547][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:29:10,042][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:29:10,539][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:29:11,035][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:29:11,531][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:29:12,027][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:29:12,527][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:29:13,023][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:29:13,521][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:29:14,018][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:29:14,518][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:29:15,016][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:29:15,515][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:29:16,013][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:29:16,510][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:29:17,007][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:29:17,508][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:29:18,006][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:29:18,505][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:29:19,003][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:29:19,501][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:29:20,001][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:29:20,500][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:29:21,001][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:29:21,500][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:29:21,996][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:29:22,495][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:29:22,996][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:29:23,500][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:29:23,997][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:29:24,493][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:29:24,988][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:29:25,484][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:29:25,981][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:29:26,476][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:29:26,973][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:29:27,469][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:29:27,964][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:29:28,462][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:29:28,956][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:29:29,451][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:29:29,946][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:29:30,441][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:29:30,937][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:29:31,434][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:29:31,930][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:29:32,424][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:29:32,918][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:29:33,413][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:29:33,912][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:29:34,408][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:29:34,903][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:29:35,398][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:29:35,893][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:29:36,388][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:29:36,883][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9739 tokens. +[2026-03-25 22:29:37,567][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.93%, Current % of VRAM taken: 60.35%, Block Peak % of device VRAM: 61.70%, ΔTime: 00:00:32 +[2026-03-25 22:29:38,327][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:29:38,330][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:29:38,331][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:29:38,989][__main__][INFO] - Iteration 32 took 48s (27.77% Gen, 70.89% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 21m 7s. Estimated total time: 40h 49m 57s. Time estimates for 10 more iterations: 8m 9s, 100 more iterations: 1h 21m 39s, 500 more iterations: 6h 48m 19s. +[2026-03-25 22:29:38,991][__main__][INFO] - Starting iteration 32. +[2026-03-25 22:29:39,395][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 3 and human policies 1. +[2026-03-25 22:29:39,396][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:29:53,624][__main__][INFO] - Number of regex retries in iteration 32: 0 +[2026-03-25 22:29:53,625][__main__][INFO] - agents played in iteration 32 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:29:54,491][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:29:54,515][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:29:54,539][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:29:54,562][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:29:54,562][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:29:54,563][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:29:55,257][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:29:55,715][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:29:56,218][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:29:56,718][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:29:57,220][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:29:57,717][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:29:58,218][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:29:58,715][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:29:59,213][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:29:59,711][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:30:00,207][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:30:00,702][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:30:01,201][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:30:01,697][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:30:02,194][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:30:02,695][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:30:03,194][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:30:03,695][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:30:04,192][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:30:04,689][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:30:05,185][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:30:05,682][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:30:06,179][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:30:06,675][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:30:07,170][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:30:07,667][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:30:08,165][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:30:08,663][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:30:09,160][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:30:09,658][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:30:10,155][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:30:10,655][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:30:11,154][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:30:11,650][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:30:12,147][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:30:12,644][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:30:13,145][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:30:13,644][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:30:14,147][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:30:14,646][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:30:15,142][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:30:15,637][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:30:16,134][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:30:16,631][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:30:17,125][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:30:17,621][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:30:18,117][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:30:18,612][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:30:19,108][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:30:19,606][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:30:20,103][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:30:20,602][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:30:21,100][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:30:21,596][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:30:22,093][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:30:22,589][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:30:23,085][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:30:23,582][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:30:24,079][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:30:24,575][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:30:25,072][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:30:25,569][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:30:26,065][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:30:26,561][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:30:27,057][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9733 tokens. +[2026-03-25 22:30:27,750][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.98%, Current % of VRAM taken: 60.41%, Block Peak % of device VRAM: 61.68%, ΔTime: 00:00:32 +[2026-03-25 22:30:28,507][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:30:28,509][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:30:28,511][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:30:29,166][__main__][INFO] - Iteration 33 took 49s (28.59% Gen, 70.09% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 58m 52s. Estimated total time: 41h 28m 33s. Time estimates for 10 more iterations: 8m 17s, 100 more iterations: 1h 22m 57s, 500 more iterations: 6h 54m 45s. +[2026-03-25 22:30:29,168][__main__][INFO] - Starting iteration 33. +[2026-03-25 22:30:29,569][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 3 and human policies 1. +[2026-03-25 22:30:29,570][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:30:43,175][__main__][INFO] - Number of regex retries in iteration 33: 0 +[2026-03-25 22:30:43,176][__main__][INFO] - agents played in iteration 33 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:30:43,964][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:30:43,988][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:30:44,011][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:30:44,034][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:30:44,035][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:30:44,036][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:30:44,690][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:30:45,149][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:30:45,659][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:30:46,159][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:30:46,665][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:30:47,167][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:30:47,665][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:30:48,163][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:30:48,660][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:30:49,164][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:30:49,664][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:30:50,163][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:30:50,661][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:30:51,159][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:30:51,657][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:30:52,156][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:30:52,655][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:30:53,152][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:30:53,650][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:30:54,147][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:30:54,644][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:30:55,142][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:30:55,640][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:30:56,139][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:30:56,636][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:30:57,134][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:30:57,632][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:30:58,130][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:30:58,627][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:30:59,126][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:30:59,623][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:31:00,122][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:31:00,618][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:31:01,115][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:31:01,611][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:31:02,108][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:31:02,604][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:31:03,099][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:31:03,596][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:31:04,091][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:31:04,591][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:31:05,095][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:31:05,596][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:31:06,093][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:31:06,592][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:31:07,092][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:31:07,590][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:31:08,087][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:31:08,584][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:31:09,080][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:31:09,577][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:31:10,073][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:31:10,568][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:31:11,062][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:31:11,557][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:31:12,052][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:31:12,549][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:31:13,044][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:31:13,540][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:31:14,038][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:31:14,535][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:31:15,030][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:31:15,525][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:31:16,020][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:31:16,516][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9757 tokens. +[2026-03-25 22:31:17,202][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.01%, Current % of VRAM taken: 60.44%, Block Peak % of device VRAM: 61.70%, ΔTime: 00:00:32 +[2026-03-25 22:31:17,953][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:31:17,955][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:31:17,957][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:31:18,618][__main__][INFO] - Iteration 34 took 49s (27.74% Gen, 70.91% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 22m 0s. Estimated total time: 40h 52m 30s. Time estimates for 10 more iterations: 8m 10s, 100 more iterations: 1h 21m 45s, 500 more iterations: 6h 48m 45s. +[2026-03-25 22:31:18,621][__main__][INFO] - Starting iteration 34. +[2026-03-25 22:31:19,024][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 3 and human policies 1. +[2026-03-25 22:31:19,024][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:31:20,274][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:31:33,000][__main__][INFO] - Number of regex retries in iteration 34: 1 +[2026-03-25 22:31:33,001][__main__][INFO] - agents played in iteration 34 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:31:33,773][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:31:33,797][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:31:33,820][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:31:33,844][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:31:33,844][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:31:33,845][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:31:34,453][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:31:34,910][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:31:35,417][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:31:35,924][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:31:36,422][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:31:36,922][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:31:37,422][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:31:37,917][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:31:38,413][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:31:38,907][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:31:39,404][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:31:39,899][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:31:40,396][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:31:40,892][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:31:41,388][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:31:41,883][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:31:42,379][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:31:42,878][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:31:43,375][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:31:43,872][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:31:44,372][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:31:44,868][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:31:45,362][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:31:45,857][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:31:46,355][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:31:46,851][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:31:47,348][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:31:47,842][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:31:48,338][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:31:48,835][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:31:49,333][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:31:49,830][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:31:50,330][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:31:50,825][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:31:51,320][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:31:51,816][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:31:52,311][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:31:52,806][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:31:53,303][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:31:53,799][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:31:54,293][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:31:54,792][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:31:55,295][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:31:55,797][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:31:56,306][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:31:56,804][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:31:57,303][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:31:57,803][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:31:58,301][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:31:58,796][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:31:59,292][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:31:59,788][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:32:00,282][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:32:00,776][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:32:01,270][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:32:01,765][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:32:02,259][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:32:02,755][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:32:03,252][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:32:03,749][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:32:04,245][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:32:04,741][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:32:05,239][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:32:05,736][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:32:06,232][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9751 tokens. +[2026-03-25 22:32:06,924][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.94%, Current % of VRAM taken: 60.36%, Block Peak % of device VRAM: 61.85%, ΔTime: 00:00:32 +[2026-03-25 22:32:07,703][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:32:07,705][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:32:07,707][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:32:08,359][__main__][INFO] - Iteration 35 took 49s (28.33% Gen, 70.35% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 35m 28s. Estimated total time: 41h 6m 47s. Time estimates for 10 more iterations: 8m 13s, 100 more iterations: 1h 22m 13s, 500 more iterations: 6h 51m 7s. +[2026-03-25 22:32:08,362][__main__][INFO] - Starting iteration 35. +[2026-03-25 22:32:08,763][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 3 and human policies 1. +[2026-03-25 22:32:08,764][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:32:16,860][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:32:22,950][__main__][INFO] - Number of regex retries in iteration 35: 1 +[2026-03-25 22:32:22,951][__main__][INFO] - agents played in iteration 35 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:32:23,726][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.40%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:32:23,750][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.40%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:32:23,773][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.40%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:32:23,796][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.40%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:32:23,797][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:32:23,797][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:32:24,439][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:32:24,900][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:32:25,408][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:32:25,910][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:32:26,413][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:32:26,912][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:32:27,410][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:32:27,911][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:32:28,410][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:32:28,906][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:32:29,402][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:32:29,898][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:32:30,397][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:32:30,894][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:32:31,389][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:32:31,885][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:32:32,382][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:32:32,880][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:32:33,378][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:32:33,875][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:32:34,373][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:32:34,871][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:32:35,370][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:32:35,869][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:32:36,366][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:32:36,866][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:32:37,364][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:32:37,860][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:32:38,359][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:32:38,857][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:32:39,355][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:32:39,853][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:32:40,353][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:32:40,851][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:32:41,346][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:32:41,841][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:32:42,337][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:32:42,835][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:32:43,331][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:32:43,829][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:32:44,326][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:32:44,825][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:32:45,327][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:32:45,830][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:32:46,328][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:32:46,824][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:32:47,322][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:32:47,820][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:32:48,318][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:32:48,814][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:32:49,309][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:32:49,803][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:32:50,303][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:32:50,799][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:32:51,293][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:32:51,793][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:32:52,288][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:32:52,788][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:32:53,286][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:32:53,783][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:32:54,280][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:32:54,778][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:32:55,276][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:32:55,774][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:32:56,273][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9802 tokens. +[2026-03-25 22:32:56,966][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 61.74%, ΔTime: 00:00:32 +[2026-03-25 22:32:57,721][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:32:57,724][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:32:57,726][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:32:58,373][__main__][INFO] - Iteration 36 took 49s (28.60% Gen, 70.10% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 48m 22s. Estimated total time: 41h 20m 32s. Time estimates for 10 more iterations: 8m 16s, 100 more iterations: 1h 22m 41s, 500 more iterations: 6h 53m 25s. +[2026-03-25 22:32:58,375][__main__][INFO] - Starting iteration 36. +[2026-03-25 22:32:58,777][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 3 and human policies 1. +[2026-03-25 22:32:58,777][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:33:12,800][__main__][INFO] - Number of regex retries in iteration 36: 0 +[2026-03-25 22:33:12,801][__main__][INFO] - agents played in iteration 36 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:33:13,610][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:33:13,635][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:33:13,658][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:33:13,682][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:33:13,682][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:33:13,683][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:33:14,346][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:33:14,805][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:33:15,319][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:33:15,822][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:33:16,322][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:33:16,823][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:33:17,324][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:33:17,821][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:33:18,319][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:33:18,818][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:33:19,316][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:33:19,815][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:33:20,315][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:33:20,812][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:33:21,310][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:33:21,810][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:33:22,308][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:33:22,806][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:33:23,305][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:33:23,803][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:33:24,304][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:33:24,801][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:33:25,301][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:33:25,800][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:33:26,297][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:33:26,795][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:33:27,293][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:33:27,790][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:33:28,290][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:33:28,788][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:33:29,288][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:33:29,786][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:33:30,284][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:33:30,781][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:33:31,280][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:33:31,778][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:33:32,276][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:33:32,772][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:33:33,269][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:33:33,767][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:33:34,263][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:33:34,761][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:33:35,257][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:33:35,755][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:33:36,252][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:33:36,748][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:33:37,244][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:33:37,743][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:33:38,245][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:33:38,743][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:33:39,243][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:33:39,740][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:33:40,238][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:33:40,737][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:33:41,236][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:33:41,733][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:33:42,230][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:33:42,725][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:33:43,219][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:33:43,714][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:33:44,209][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:33:44,702][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:33:45,198][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:33:45,694][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:33:46,190][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9803 tokens. +[2026-03-25 22:33:46,888][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.98%, Current % of VRAM taken: 60.41%, Block Peak % of device VRAM: 61.71%, ΔTime: 00:00:32 +[2026-03-25 22:33:47,653][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:33:47,656][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:33:47,657][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:33:48,315][__main__][INFO] - Iteration 37 took 49s (28.31% Gen, 70.36% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 43m 57s. Estimated total time: 41h 16m 57s. Time estimates for 10 more iterations: 8m 15s, 100 more iterations: 1h 22m 33s, 500 more iterations: 6h 52m 49s. +[2026-03-25 22:33:48,317][__main__][INFO] - Starting iteration 37. +[2026-03-25 22:33:48,719][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 3 and human policies 1. +[2026-03-25 22:33:48,720][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:34:02,268][__main__][INFO] - Number of regex retries in iteration 37: 0 +[2026-03-25 22:34:02,269][__main__][INFO] - agents played in iteration 37 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:34:03,079][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:34:03,103][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:34:03,126][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:34:03,149][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:34:03,150][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:34:03,150][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:34:03,776][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:34:04,228][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:34:04,735][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:34:05,238][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:34:05,739][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:34:06,238][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:34:06,734][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:34:07,233][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:34:07,728][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:34:08,230][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:34:08,756][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:34:09,256][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:34:09,753][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:34:10,251][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:34:10,747][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:34:11,243][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:34:11,741][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:34:12,239][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:34:12,734][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:34:13,231][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:34:13,727][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:34:14,226][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:34:14,722][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:34:15,219][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:34:15,716][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:34:16,213][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:34:16,713][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:34:17,210][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:34:17,709][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:34:18,208][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:34:18,707][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:34:19,206][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:34:19,702][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:34:20,203][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:34:20,700][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:34:21,201][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:34:21,699][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:34:22,201][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:34:22,697][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:34:23,198][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:34:23,697][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:34:24,194][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:34:24,692][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:34:25,189][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:34:25,685][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:34:26,181][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:34:26,676][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:34:27,175][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:34:27,675][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:34:28,178][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:34:28,674][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:34:29,170][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:34:29,668][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:34:30,163][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:34:30,657][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:34:31,153][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:34:31,647][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:34:32,143][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:34:32,639][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:34:33,135][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:34:33,632][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:34:34,128][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:34:34,624][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:34:35,120][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:34:35,619][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9793 tokens. +[2026-03-25 22:34:36,306][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 61.84%, ΔTime: 00:00:32 +[2026-03-25 22:34:37,059][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:34:37,061][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:34:37,063][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:34:37,707][__main__][INFO] - Iteration 38 took 48s (27.66% Gen, 71.02% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 15m 42s. Estimated total time: 40h 49m 31s. Time estimates for 10 more iterations: 8m 9s, 100 more iterations: 1h 21m 39s, 500 more iterations: 6h 48m 15s. +[2026-03-25 22:34:37,710][__main__][INFO] - Starting iteration 38. +[2026-03-25 22:34:38,111][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 3 and human policies 1. +[2026-03-25 22:34:38,112][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:34:51,405][__main__][INFO] - Number of regex retries in iteration 38: 0 +[2026-03-25 22:34:51,406][__main__][INFO] - agents played in iteration 38 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:34:52,221][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:34:52,244][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:34:52,268][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:34:52,290][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:34:52,291][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:34:52,292][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:34:52,935][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:34:53,391][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:34:53,893][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:34:54,391][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:34:54,898][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:34:55,403][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:34:55,906][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:34:56,404][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:34:56,904][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:34:57,403][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:34:57,901][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:34:58,396][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:34:58,893][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:34:59,389][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:34:59,887][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:35:00,386][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:35:00,882][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:35:01,379][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:35:01,876][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:35:02,373][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:35:02,869][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:35:03,366][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:35:03,862][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:35:04,359][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:35:04,860][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:35:05,357][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:35:05,853][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:35:06,349][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:35:06,845][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:35:07,340][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:35:07,836][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:35:08,333][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:35:08,831][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:35:09,327][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:35:09,824][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:35:10,321][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:35:10,821][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:35:11,321][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:35:11,819][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:35:12,315][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:35:12,811][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:35:13,307][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:35:13,802][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:35:14,298][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:35:14,793][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:35:15,290][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:35:15,787][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:35:16,282][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:35:16,781][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:35:17,283][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:35:17,785][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:35:18,289][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:35:18,791][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:35:19,289][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:35:19,785][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:35:20,283][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:35:20,779][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:35:21,276][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:35:21,772][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:35:22,268][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:35:22,764][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:35:23,260][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:35:23,758][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:35:24,254][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:35:24,753][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9717 tokens. +[2026-03-25 22:35:25,451][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.98%, Current % of VRAM taken: 60.40%, Block Peak % of device VRAM: 61.63%, ΔTime: 00:00:32 +[2026-03-25 22:35:26,207][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:35:26,209][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:35:26,211][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:35:26,860][__main__][INFO] - Iteration 39 took 48s (27.27% Gen, 71.40% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 2m 49s. Estimated total time: 40h 37m 27s. Time estimates for 10 more iterations: 8m 7s, 100 more iterations: 1h 21m 14s, 500 more iterations: 6h 46m 14s. +[2026-03-25 22:35:26,862][__main__][INFO] - Starting iteration 39. +[2026-03-25 22:35:27,267][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 3 and human policies 1. +[2026-03-25 22:35:27,268][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:35:40,939][__main__][INFO] - Number of regex retries in iteration 39: 0 +[2026-03-25 22:35:40,940][__main__][INFO] - agents played in iteration 39 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:35:41,721][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:35:41,745][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:35:41,768][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:35:41,791][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:35:41,792][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:35:41,792][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:35:42,401][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:35:42,854][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:35:43,354][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:35:43,851][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:35:44,348][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:35:44,855][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:35:45,359][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:35:45,860][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:35:46,357][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:35:46,858][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:35:47,359][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:35:47,858][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:35:48,356][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:35:48,854][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:35:49,349][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:35:49,847][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:35:50,345][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:35:50,839][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:35:51,336][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:35:51,834][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:35:52,332][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:35:52,828][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:35:53,325][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:35:53,822][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:35:54,319][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:35:54,815][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:35:55,315][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:35:55,812][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:35:56,312][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:35:56,810][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:35:57,309][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:35:57,809][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:35:58,312][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:35:58,807][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:35:59,301][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:35:59,798][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:36:00,296][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:36:00,794][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:36:01,291][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:36:01,786][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:36:02,281][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:36:02,777][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:36:03,274][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:36:03,775][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:36:04,273][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:36:04,768][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:36:05,265][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:36:05,761][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:36:06,257][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:36:06,755][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:36:07,251][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:36:07,747][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:36:08,250][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:36:08,751][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:36:09,249][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:36:09,743][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:36:10,238][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:36:10,735][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:36:11,233][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:36:11,731][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:36:12,228][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:36:12,725][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:36:13,222][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:36:13,718][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:36:14,215][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9727 tokens. +[2026-03-25 22:36:14,911][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.99%, Current % of VRAM taken: 60.42%, Block Peak % of device VRAM: 61.73%, ΔTime: 00:00:32 +[2026-03-25 22:36:15,670][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:36:15,672][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:36:15,674][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:36:16,325][__main__][INFO] - Iteration 40 took 49s (27.87% Gen, 70.80% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 17m 27s. Estimated total time: 40h 52m 55s. Time estimates for 10 more iterations: 8m 10s, 100 more iterations: 1h 21m 45s, 500 more iterations: 6h 48m 49s. +[2026-03-25 22:36:16,327][__main__][INFO] - Starting iteration 40. +[2026-03-25 22:36:16,729][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 3 and human policies 1. +[2026-03-25 22:36:16,730][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:36:30,542][__main__][INFO] - Number of regex retries in iteration 40: 0 +[2026-03-25 22:36:30,542][__main__][INFO] - agents played in iteration 40 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:36:31,344][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:36:31,368][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:36:31,391][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:36:31,414][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:36:31,415][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:36:31,415][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:36:32,028][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:36:32,480][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:36:32,982][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:36:33,479][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:36:33,983][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:36:34,487][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:36:34,988][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:36:35,493][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:36:35,994][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:36:36,493][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:36:36,993][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:36:37,493][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:36:37,993][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:36:38,491][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:36:38,988][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:36:39,484][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:36:39,981][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:36:40,479][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:36:40,977][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:36:41,473][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:36:41,971][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:36:42,469][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:36:42,964][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:36:43,462][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:36:43,960][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:36:44,458][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:36:44,957][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:36:45,456][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:36:45,955][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:36:46,452][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:36:46,949][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:36:47,451][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:36:47,975][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:36:48,474][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:36:48,971][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:36:49,466][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:36:49,964][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:36:50,461][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:36:50,958][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:36:51,455][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:36:51,950][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:36:52,446][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:36:52,943][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:36:53,439][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:36:53,939][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:36:54,436][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:36:54,934][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:36:55,430][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:36:55,928][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:36:56,424][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:36:56,918][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:36:57,414][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:36:57,911][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:36:58,410][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:36:58,909][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:36:59,414][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:36:59,911][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:37:00,407][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:37:00,903][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:37:01,399][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:37:01,895][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:37:02,391][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:37:02,885][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:37:03,381][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:37:03,877][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9736 tokens. +[2026-03-25 22:37:04,578][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.97%, Current % of VRAM taken: 60.40%, Block Peak % of device VRAM: 61.83%, ΔTime: 00:00:32 +[2026-03-25 22:37:05,336][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:37:05,338][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:37:05,340][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:37:06,573][__main__][INFO] - Iteration 41 took 49s (27.71% Gen, 69.81% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 55m 56s. Estimated total time: 41h 32m 14s. Time estimates for 10 more iterations: 8m 18s, 100 more iterations: 1h 23m 4s, 500 more iterations: 6h 55m 22s. +[2026-03-25 22:37:06,575][__main__][INFO] - Starting iteration 41. +[2026-03-25 22:37:06,975][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 4 and human policies 1. +[2026-03-25 22:37:06,976][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:37:08,549][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:37:12,696][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:37:21,041][__main__][INFO] - Number of regex retries in iteration 41: 2 +[2026-03-25 22:37:21,041][__main__][INFO] - agents played in iteration 41 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:37:21,826][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:37:21,850][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:37:21,873][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:37:21,896][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:37:21,897][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:37:21,897][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:37:22,503][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:37:22,957][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:37:23,457][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:37:23,956][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:37:24,458][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:37:24,962][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:37:25,466][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:37:25,963][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:37:26,462][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:37:26,961][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:37:27,459][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:37:27,955][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:37:28,451][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:37:28,948][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:37:29,442][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:37:29,937][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:37:30,433][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:37:30,931][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:37:31,428][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:37:31,923][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:37:32,420][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:37:32,916][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:37:33,413][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:37:33,911][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:37:34,410][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:37:34,908][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:37:35,403][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:37:35,899][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:37:36,396][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:37:36,890][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:37:37,386][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:37:37,881][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:37:38,378][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:37:38,877][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:37:39,374][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:37:39,872][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:37:40,369][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:37:40,876][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:37:41,392][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:37:41,891][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:37:42,387][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:37:42,885][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:37:43,381][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:37:43,878][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:37:44,376][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:37:44,873][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:37:45,371][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:37:45,869][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:37:46,365][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:37:46,862][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:37:47,359][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:37:47,857][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:37:48,361][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:37:48,863][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:37:49,362][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:37:49,860][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:37:50,359][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:37:50,856][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:37:51,350][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:37:51,845][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:37:52,340][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:37:52,834][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:37:53,329][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:37:53,823][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:37:54,319][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9775 tokens. +[2026-03-25 22:37:55,015][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.00%, Current % of VRAM taken: 60.42%, Block Peak % of device VRAM: 61.87%, ΔTime: 00:00:32 +[2026-03-25 22:37:55,770][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:37:55,772][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:37:55,774][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:37:56,413][__main__][INFO] - Iteration 42 took 49s (28.45% Gen, 70.25% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 34m 48s. Estimated total time: 41h 11m 56s. Time estimates for 10 more iterations: 8m 14s, 100 more iterations: 1h 22m 23s, 500 more iterations: 6h 51m 59s. +[2026-03-25 22:37:56,416][__main__][INFO] - Starting iteration 42. +[2026-03-25 22:37:56,815][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 4 and human policies 1. +[2026-03-25 22:37:56,816][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:38:10,849][__main__][INFO] - Number of regex retries in iteration 42: 0 +[2026-03-25 22:38:10,850][__main__][INFO] - agents played in iteration 42 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:38:11,633][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:38:11,657][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:38:11,680][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:38:11,703][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:38:11,704][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:38:11,704][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:38:12,314][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:38:12,765][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:38:13,272][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:38:13,775][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:38:14,277][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:38:14,784][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:38:15,291][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:38:15,795][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:38:16,297][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:38:16,797][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:38:17,296][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:38:17,796][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:38:18,292][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:38:18,789][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:38:19,285][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:38:19,782][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:38:20,283][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:38:20,780][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:38:21,278][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:38:21,774][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:38:22,271][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:38:22,770][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:38:23,271][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:38:23,769][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:38:24,265][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:38:24,761][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:38:25,259][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:38:25,756][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:38:26,255][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:38:26,751][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:38:27,248][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:38:27,745][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:38:28,241][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:38:28,740][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:38:29,243][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:38:29,739][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:38:30,237][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:38:30,736][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:38:31,237][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:38:31,735][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:38:32,234][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:38:32,730][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:38:33,227][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:38:33,723][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:38:34,221][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:38:34,718][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:38:35,215][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:38:35,714][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:38:36,213][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:38:36,709][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:38:37,208][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:38:37,707][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:38:38,209][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:38:38,705][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:38:39,200][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:38:39,695][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:38:40,192][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:38:40,688][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:38:41,185][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:38:41,682][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:38:42,179][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:38:42,675][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:38:43,173][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:38:43,667][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:38:44,161][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9715 tokens. +[2026-03-25 22:38:44,866][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.96%, Current % of VRAM taken: 60.38%, Block Peak % of device VRAM: 61.88%, ΔTime: 00:00:32 +[2026-03-25 22:38:45,626][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:38:45,628][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:38:45,630][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:38:46,628][__main__][INFO] - Iteration 43 took 49s (28.17% Gen, 69.82% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 52m 42s. Estimated total time: 41h 30m 40s. Time estimates for 10 more iterations: 8m 18s, 100 more iterations: 1h 23m 1s, 500 more iterations: 6h 55m 6s. +[2026-03-25 22:38:46,630][__main__][INFO] - Starting iteration 43. +[2026-03-25 22:38:47,031][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 4 and human policies 1. +[2026-03-25 22:38:47,032][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:38:48,560][mllm.models.large_language_model_local][WARNING] - Response Proposal: 5 hats, 5 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:38:49,652][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:38:51,601][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:38:52,906][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls + +This proposal aims to capitalize on the significant value advantage I have for hats, while assuming that Bob values books much more and may take most or all of them. I will hold onto all 10 hats, which is my highest-value item, and let Bob have the other items where he has a higher valuation. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:39:00,864][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:39:02,275][__main__][INFO] - Number of regex retries in iteration 43: 5 +[2026-03-25 22:39:02,276][__main__][INFO] - agents played in iteration 43 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:39:03,058][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:39:03,082][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:39:03,106][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:39:03,129][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:39:03,130][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:39:03,130][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:39:03,746][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:39:04,202][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:39:04,710][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:39:05,214][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:39:05,718][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:39:06,218][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:39:06,718][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:39:07,218][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:39:07,715][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:39:08,211][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:39:08,709][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:39:09,204][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:39:09,704][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:39:10,201][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:39:10,699][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:39:11,196][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:39:11,694][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:39:12,191][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:39:12,689][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:39:13,183][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:39:13,678][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:39:14,174][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:39:14,670][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:39:15,165][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:39:15,662][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:39:16,159][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:39:16,654][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:39:17,150][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:39:17,646][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:39:18,141][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:39:18,636][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:39:19,132][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:39:19,627][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:39:20,125][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:39:20,622][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:39:21,119][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:39:21,616][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:39:22,113][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:39:22,610][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:39:23,105][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:39:23,599][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:39:24,095][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:39:24,591][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:39:25,086][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:39:25,581][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:39:26,080][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:39:26,580][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:39:27,078][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:39:27,576][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:39:28,073][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:39:28,570][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:39:29,066][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:39:29,562][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:39:30,057][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:39:30,552][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:39:31,046][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:39:31,541][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:39:32,037][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:39:32,534][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:39:33,037][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:39:33,534][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:39:34,031][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:39:34,526][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:39:35,023][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:39:35,520][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9729 tokens. +[2026-03-25 22:39:36,214][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.48%, Block Peak % of device VRAM: 61.83%, ΔTime: 00:00:32 +[2026-03-25 22:39:36,979][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:39:36,981][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:39:36,983][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:39:37,633][__main__][INFO] - Iteration 44 took 50s (30.13% Gen, 68.59% Train). Generation: 15s, Training: 34s. Estimated remaining time: 41h 31m 17s. Estimated total time: 42h 10m 6s. Time estimates for 10 more iterations: 8m 26s, 100 more iterations: 1h 24m 20s, 500 more iterations: 7h 1m 41s. +[2026-03-25 22:39:37,635][__main__][INFO] - Starting iteration 44. +[2026-03-25 22:39:38,036][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 4 and human policies 1. +[2026-03-25 22:39:38,037][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:39:44,717][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:39:52,432][__main__][INFO] - Number of regex retries in iteration 44: 1 +[2026-03-25 22:39:52,434][__main__][INFO] - agents played in iteration 44 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:39:53,224][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:39:53,248][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:39:53,272][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:39:53,296][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:39:53,296][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:39:53,297][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:39:53,976][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:39:54,431][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:39:54,938][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:39:55,449][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:39:55,958][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:39:56,458][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:39:56,959][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:39:57,459][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:39:57,961][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:39:58,460][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:39:58,957][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:39:59,454][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:39:59,950][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:40:00,448][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:40:00,943][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:40:01,439][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:40:01,935][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:40:02,432][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:40:02,928][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:40:03,425][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:40:03,921][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:40:04,418][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:40:04,914][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:40:05,412][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:40:05,909][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:40:06,409][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:40:06,909][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:40:07,409][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:40:07,908][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:40:08,406][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:40:08,907][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:40:09,407][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:40:09,903][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:40:10,403][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:40:10,899][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:40:11,398][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:40:11,895][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:40:12,393][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:40:12,889][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:40:13,386][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:40:13,883][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:40:14,380][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:40:14,878][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:40:15,378][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:40:15,880][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:40:16,383][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:40:16,890][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:40:17,388][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:40:17,888][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:40:18,388][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:40:18,885][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:40:19,386][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:40:19,884][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:40:20,383][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:40:20,883][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:40:21,382][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:40:21,879][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:40:22,378][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:40:22,876][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:40:23,397][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:40:23,898][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:40:24,398][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:40:24,898][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:40:25,396][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:40:25,893][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9824 tokens. +[2026-03-25 22:40:26,619][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.02%, Current % of VRAM taken: 60.45%, Block Peak % of device VRAM: 61.78%, ΔTime: 00:00:32 +[2026-03-25 22:40:27,383][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:40:27,386][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:40:27,387][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:40:28,090][__main__][INFO] - Iteration 45 took 50s (28.76% Gen, 69.83% Train). Generation: 14s, Training: 34s. Estimated remaining time: 41h 3m 4s. Estimated total time: 41h 42m 43s. Time estimates for 10 more iterations: 8m 20s, 100 more iterations: 1h 23m 25s, 500 more iterations: 6h 57m 7s. +[2026-03-25 22:40:28,093][__main__][INFO] - Starting iteration 45. +[2026-03-25 22:40:28,495][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 4 and human policies 1. +[2026-03-25 22:40:28,496][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:40:30,786][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:40:34,849][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:40:42,518][__main__][INFO] - Number of regex retries in iteration 45: 2 +[2026-03-25 22:40:42,520][__main__][INFO] - agents played in iteration 45 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:40:43,300][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:40:43,323][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:40:43,347][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:40:43,370][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:40:43,370][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:40:43,371][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:40:44,029][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:40:44,484][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:40:44,996][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:40:45,504][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:40:46,007][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:40:46,504][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:40:47,003][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:40:47,502][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:40:48,001][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:40:48,499][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:40:48,998][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:40:49,496][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:40:49,995][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:40:50,494][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:40:50,992][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:40:51,489][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:40:51,987][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:40:52,484][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:40:52,981][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:40:53,484][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:40:53,983][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:40:54,482][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:40:54,982][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:40:55,482][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:40:55,979][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:40:56,477][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:40:56,976][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:40:57,474][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:40:57,973][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:40:58,472][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:40:58,969][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:40:59,466][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:40:59,963][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:41:00,459][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:41:00,959][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:41:01,457][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:41:01,954][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:41:02,449][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:41:02,944][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:41:03,439][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:41:03,941][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:41:04,440][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:41:04,943][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:41:05,442][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:41:05,937][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:41:06,432][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:41:06,927][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:41:07,423][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:41:07,918][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:41:08,415][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:41:08,912][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:41:09,409][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:41:09,905][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:41:10,401][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:41:10,899][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:41:11,398][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:41:11,897][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:41:12,393][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:41:12,889][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:41:13,386][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:41:13,882][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:41:14,377][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:41:14,872][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:41:15,369][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:41:15,865][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9885 tokens. +[2026-03-25 22:41:16,572][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.02%, Current % of VRAM taken: 60.44%, Block Peak % of device VRAM: 61.85%, ΔTime: 00:00:32 +[2026-03-25 22:41:17,331][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:41:17,333][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:41:17,335][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:41:17,985][__main__][INFO] - Iteration 46 took 49s (28.34% Gen, 70.35% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 34m 4s. Estimated total time: 41h 14m 33s. Time estimates for 10 more iterations: 8m 14s, 100 more iterations: 1h 22m 29s, 500 more iterations: 6h 52m 25s. +[2026-03-25 22:41:17,988][__main__][INFO] - Starting iteration 46. +[2026-03-25 22:41:18,394][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 4 and human policies 1. +[2026-03-25 22:41:18,395][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:41:19,945][mllm.models.large_language_model_local][WARNING] - Response Proposal: 5 hats, 5 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:41:26,990][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls + +Given the per-item values, hats are highly valued by both of us, while balls are valued the least by both. Books are valued by Alice but not by Bob. By proposing to take all 10 hats, I'm maximizing the value of the items I propose to keep. Since hats are equally valued by both, and I am taking all of them, I can be sure that Bob will not be able to propose taking any of them, ensuring that I get the full value of 10 points per hat. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:41:28,651][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:41:33,675][__main__][INFO] - Number of regex retries in iteration 46: 3 +[2026-03-25 22:41:33,676][__main__][INFO] - agents played in iteration 46 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:41:34,463][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:41:34,487][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:41:34,510][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:41:34,533][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:41:34,534][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:41:34,534][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:41:35,150][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:41:35,607][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:41:36,110][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:41:36,611][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:41:37,109][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:41:37,606][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:41:38,103][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:41:38,599][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:41:39,099][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:41:39,596][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:41:40,096][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:41:40,593][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:41:41,089][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:41:41,585][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:41:42,083][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:41:42,581][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:41:43,079][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:41:43,574][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:41:44,073][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:41:44,569][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:41:45,065][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:41:45,563][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:41:46,060][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:41:46,556][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:41:47,054][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:41:47,550][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:41:48,047][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:41:48,543][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:41:49,040][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:41:49,536][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:41:50,033][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:41:50,528][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:41:51,025][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:41:51,523][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:41:52,022][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:41:52,518][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:41:53,015][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:41:53,514][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:41:54,018][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:41:54,520][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:41:55,019][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:41:55,516][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:41:56,017][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:41:56,514][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:41:57,012][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:41:57,509][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:41:58,007][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:41:58,502][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:41:58,999][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:41:59,497][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:41:59,994][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:42:00,489][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:42:00,985][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:42:01,481][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:42:01,976][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:42:02,473][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:42:02,971][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:42:03,466][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:42:03,960][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:42:04,456][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:42:04,951][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:42:05,445][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:42:05,939][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:42:06,435][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:42:06,930][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9685 tokens. +[2026-03-25 22:42:07,630][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.95%, Current % of VRAM taken: 60.37%, Block Peak % of device VRAM: 61.69%, ΔTime: 00:00:32 +[2026-03-25 22:42:08,385][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:42:08,388][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:42:08,390][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:42:09,035][__main__][INFO] - Iteration 47 took 50s (30.18% Gen, 68.55% Train). Generation: 15s, Training: 34s. Estimated remaining time: 41h 30m 46s. Estimated total time: 42h 12m 7s. Time estimates for 10 more iterations: 8m 26s, 100 more iterations: 1h 24m 24s, 500 more iterations: 7h 2m 1s. +[2026-03-25 22:42:09,038][__main__][INFO] - Starting iteration 47. +[2026-03-25 22:42:09,440][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 4 and human policies 1. +[2026-03-25 22:42:09,441][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:42:15,193][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls + +Given the per-item values, I (Alice) have a higher value for hats compared to Bob and a lower value for books and balls compared to Bob. To maximize my points, I should focus on claiming as many hats as possible since they are high-value items for me, and avoid proposing for items that Bob values more highly. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:42:16,449][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls + +Given the per-item values, you have a significant advantage with hats, while Alice has a significant advantage with books and balls. By proposing to keep all 10 hats, you are trying to maximize your points given your high value for hats. Since the other items have lower values for you, it makes sense to propose to give them away to let Alice have a chance to get the items she values more. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:42:24,387][__main__][INFO] - Number of regex retries in iteration 47: 2 +[2026-03-25 22:42:24,388][__main__][INFO] - agents played in iteration 47 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:42:25,215][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:42:25,238][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:42:25,262][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:42:25,285][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:42:25,285][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:42:25,286][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:42:25,940][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:42:26,394][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:42:26,900][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:42:27,399][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:42:27,898][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:42:28,396][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:42:28,894][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:42:29,392][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:42:29,890][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:42:30,387][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:42:30,884][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:42:31,380][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:42:31,879][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:42:32,378][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:42:32,876][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:42:33,373][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:42:33,870][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:42:34,369][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:42:34,866][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:42:35,363][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:42:35,861][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:42:36,360][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:42:36,858][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:42:37,357][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:42:37,855][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:42:38,354][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:42:38,853][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:42:39,352][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:42:39,848][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:42:40,346][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:42:40,844][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:42:41,345][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:42:41,846][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:42:42,348][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:42:42,848][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:42:43,347][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:42:43,843][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:42:44,338][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:42:44,835][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:42:45,333][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:42:45,829][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:42:46,324][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:42:46,820][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:42:47,317][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:42:47,814][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:42:48,310][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:42:48,806][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:42:49,303][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:42:49,798][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:42:50,295][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:42:50,792][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:42:51,288][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:42:51,784][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:42:52,279][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:42:52,775][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:42:53,271][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:42:53,765][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:42:54,261][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:42:54,756][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:42:55,253][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:42:55,749][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:42:56,243][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:42:56,741][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:42:57,236][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:42:57,732][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9807 tokens. +[2026-03-25 22:42:58,442][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.01%, Current % of VRAM taken: 60.43%, Block Peak % of device VRAM: 61.84%, ΔTime: 00:00:32 +[2026-03-25 22:42:59,207][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:42:59,209][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:42:59,211][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:42:59,868][__main__][INFO] - Iteration 48 took 50s (29.64% Gen, 69.05% Train). Generation: 14s, Training: 34s. Estimated remaining time: 41h 19m 14s. Estimated total time: 42h 1m 25s. Time estimates for 10 more iterations: 8m 24s, 100 more iterations: 1h 24m 2s, 500 more iterations: 7h 0m 14s. +[2026-03-25 22:42:59,870][__main__][INFO] - Starting iteration 48. +[2026-03-25 22:43:00,272][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 4 and human policies 1. +[2026-03-25 22:43:00,272][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:43:14,093][__main__][INFO] - Number of regex retries in iteration 48: 0 +[2026-03-25 22:43:14,094][__main__][INFO] - agents played in iteration 48 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:43:14,880][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:43:14,904][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:43:14,928][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:43:14,951][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:43:14,952][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:43:14,952][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:43:15,578][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:43:16,041][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:43:16,543][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:43:17,043][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:43:17,542][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:43:18,042][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:43:18,540][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:43:19,038][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:43:19,535][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:43:20,030][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:43:20,525][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:43:21,020][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:43:21,515][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:43:22,010][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:43:22,504][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:43:22,998][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:43:23,492][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:43:23,990][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:43:24,487][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:43:24,983][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:43:25,480][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:43:25,976][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:43:26,475][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:43:26,971][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:43:27,467][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:43:27,962][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:43:28,458][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:43:28,958][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:43:29,455][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:43:29,953][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:43:30,448][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:43:30,945][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:43:31,448][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:43:31,950][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:43:32,455][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:43:32,962][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:43:33,457][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:43:33,955][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:43:34,451][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:43:34,947][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:43:35,441][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:43:35,937][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:43:36,433][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:43:36,927][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:43:37,421][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:43:37,916][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:43:38,411][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:43:38,907][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:43:39,402][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:43:39,897][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:43:40,394][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:43:40,890][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:43:41,385][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:43:41,882][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:43:42,378][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:43:42,875][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:43:43,372][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:43:43,867][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:43:44,363][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:43:44,858][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:43:45,356][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:43:45,852][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:43:46,347][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:43:46,842][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:43:47,338][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9720 tokens. +[2026-03-25 22:43:48,009][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.97%, Current % of VRAM taken: 60.40%, Block Peak % of device VRAM: 61.70%, ΔTime: 00:00:32 +[2026-03-25 22:43:48,773][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:43:48,775][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:43:48,777][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:43:49,419][__main__][INFO] - Iteration 49 took 49s (28.12% Gen, 70.57% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 14m 21s. Estimated total time: 40h 57m 22s. Time estimates for 10 more iterations: 8m 11s, 100 more iterations: 1h 21m 54s, 500 more iterations: 6h 49m 33s. +[2026-03-25 22:43:49,421][__main__][INFO] - Starting iteration 49. +[2026-03-25 22:43:49,862][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 4 and human policies 1. +[2026-03-25 22:43:49,862][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:43:51,390][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:44:04,255][__main__][INFO] - Number of regex retries in iteration 49: 1 +[2026-03-25 22:44:04,256][__main__][INFO] - agents played in iteration 49 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:44:05,057][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:44:05,081][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:44:05,105][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:44:05,128][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:44:05,129][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:44:05,130][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:44:05,747][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:44:06,200][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:44:06,704][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:44:07,206][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:44:07,703][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:44:08,205][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:44:08,702][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:44:09,198][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:44:09,696][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:44:10,193][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:44:10,690][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:44:11,187][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:44:11,683][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:44:12,180][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:44:12,678][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:44:13,176][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:44:13,674][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:44:14,169][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:44:14,666][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:44:15,161][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:44:15,656][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:44:16,152][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:44:16,648][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:44:17,142][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:44:17,638][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:44:18,135][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:44:18,635][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:44:19,131][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:44:19,628][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:44:20,125][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:44:20,624][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:44:21,120][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:44:21,616][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:44:22,116][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:44:22,622][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:44:23,122][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:44:23,623][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:44:24,122][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:44:24,618][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:44:25,114][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:44:25,611][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:44:26,106][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:44:26,603][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:44:27,099][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:44:27,594][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:44:28,092][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:44:28,588][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:44:29,084][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:44:29,580][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:44:30,077][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:44:30,574][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:44:31,074][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:44:31,576][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:44:32,076][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:44:32,574][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:44:33,071][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:44:33,569][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:44:34,067][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:44:34,563][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:44:35,060][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:44:35,557][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:44:36,054][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:44:36,553][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:44:37,047][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:44:37,542][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9760 tokens. +[2026-03-25 22:44:38,234][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.00%, Current % of VRAM taken: 60.42%, Block Peak % of device VRAM: 61.81%, ΔTime: 00:00:32 +[2026-03-25 22:44:38,986][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:44:38,988][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:44:38,990][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:44:39,644][__main__][INFO] - Iteration 50 took 49s (28.91% Gen, 69.77% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 45m 17s. Estimated total time: 41h 29m 8s. Time estimates for 10 more iterations: 8m 17s, 100 more iterations: 1h 22m 58s, 500 more iterations: 6h 54m 51s. +[2026-03-25 22:44:39,646][__main__][INFO] - Starting iteration 50. +[2026-03-25 22:44:40,049][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 4 and human policies 1. +[2026-03-25 22:44:40,049][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:44:48,798][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 20 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:44:54,273][__main__][INFO] - Number of regex retries in iteration 50: 1 +[2026-03-25 22:44:54,273][__main__][INFO] - agents played in iteration 50 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:44:55,071][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:44:55,095][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:44:55,123][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:44:55,148][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:44:55,148][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:44:55,149][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:44:55,763][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:44:56,213][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:44:56,720][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:44:57,220][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:44:57,717][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:44:58,215][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:44:58,710][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:44:59,206][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:44:59,702][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:45:00,198][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:45:00,694][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:45:01,192][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:45:01,690][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:45:02,185][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:45:02,680][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:45:03,175][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:45:03,673][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:45:04,171][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:45:04,675][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:45:05,178][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:45:05,677][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:45:06,204][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:45:06,701][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:45:07,201][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:45:07,702][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:45:08,199][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:45:08,697][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:45:09,193][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:45:09,691][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:45:10,189][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:45:10,687][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:45:11,185][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:45:11,686][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:45:12,185][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:45:12,684][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:45:13,179][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:45:13,675][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:45:14,168][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:45:14,663][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:45:15,157][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:45:15,655][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:45:16,153][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:45:16,650][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:45:17,146][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:45:17,641][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:45:18,138][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:45:18,634][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:45:19,134][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:45:19,631][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:45:20,131][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:45:20,628][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:45:21,123][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:45:21,621][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:45:22,119][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:45:22,617][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:45:23,114][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:45:23,612][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:45:24,109][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:45:24,605][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:45:25,102][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:45:25,598][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:45:26,094][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:45:26,589][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:45:27,085][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:45:27,583][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9860 tokens. +[2026-03-25 22:45:28,243][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.01%, ΔTime: 00:00:32 +[2026-03-25 22:45:29,005][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:45:29,008][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:45:29,010][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:45:30,202][__main__][INFO] - Iteration 51 took 50s (28.36% Gen, 69.26% Train). Generation: 14s, Training: 34s. Estimated remaining time: 41h 3m 0s. Estimated total time: 41h 47m 42s. Time estimates for 10 more iterations: 8m 21s, 100 more iterations: 1h 23m 35s, 500 more iterations: 6h 57m 57s. +[2026-03-25 22:45:30,204][__main__][INFO] - Starting iteration 51. +[2026-03-25 22:45:30,603][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 5 and human policies 1. +[2026-03-25 22:45:30,604][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:45:32,362][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:45:44,746][__main__][INFO] - Number of regex retries in iteration 51: 1 +[2026-03-25 22:45:44,747][__main__][INFO] - agents played in iteration 51 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:45:45,528][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:45:45,552][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:45:45,575][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:45:45,599][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:45:45,599][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:45:45,600][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:45:46,219][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:45:46,670][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:45:47,170][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:45:47,665][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:45:48,160][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:45:48,654][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:45:49,148][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:45:49,643][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:45:50,138][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:45:50,635][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:45:51,150][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:45:51,651][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:45:52,146][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:45:52,644][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:45:53,141][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:45:53,637][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:45:54,133][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:45:54,630][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:45:55,128][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:45:55,624][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:45:56,123][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:45:56,620][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:45:57,118][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:45:57,617][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:45:58,116][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:45:58,613][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:45:59,112][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:45:59,613][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:46:00,115][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:46:00,613][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:46:01,112][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:46:01,614][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:46:02,145][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:46:02,650][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:46:03,150][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:46:03,648][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:46:04,145][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:46:04,641][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:46:05,138][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:46:05,637][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:46:06,135][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:46:06,633][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:46:07,132][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:46:07,630][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:46:08,127][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:46:08,624][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:46:09,119][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:46:09,616][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:46:10,111][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:46:10,606][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:46:11,102][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:46:11,600][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:46:12,096][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:46:12,594][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:46:13,091][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:46:13,587][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:46:14,084][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:46:14,580][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:46:15,076][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:46:15,573][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:46:16,069][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:46:16,566][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:46:17,063][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:46:17,559][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:46:18,058][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9834 tokens. +[2026-03-25 22:46:18,706][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.02%, Current % of VRAM taken: 60.45%, Block Peak % of device VRAM: 61.82%, ΔTime: 00:00:32 +[2026-03-25 22:46:19,465][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:46:19,467][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:46:19,469][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:46:20,113][__main__][INFO] - Iteration 52 took 49s (28.57% Gen, 70.13% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 29m 58s. Estimated total time: 41h 15m 29s. Time estimates for 10 more iterations: 8m 15s, 100 more iterations: 1h 22m 30s, 500 more iterations: 6h 52m 34s. +[2026-03-25 22:46:20,115][__main__][INFO] - Starting iteration 52. +[2026-03-25 22:46:20,514][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 5 and human policies 1. +[2026-03-25 22:46:20,515][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:46:34,899][__main__][INFO] - Number of regex retries in iteration 52: 0 +[2026-03-25 22:46:34,899][__main__][INFO] - agents played in iteration 52 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:46:35,680][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:46:35,704][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:46:35,728][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:46:35,752][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:46:35,752][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:46:35,753][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:46:36,388][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:46:36,841][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:46:37,345][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:46:37,846][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:46:38,347][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:46:38,845][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:46:39,343][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:46:39,842][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:46:40,339][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:46:40,839][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:46:41,341][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:46:41,842][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:46:42,341][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:46:42,842][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:46:43,364][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:46:43,865][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:46:44,364][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:46:44,864][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:46:45,362][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:46:45,861][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:46:46,362][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:46:46,859][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:46:47,358][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:46:47,859][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:46:48,360][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:46:48,859][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:46:49,358][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:46:49,859][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:46:50,361][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:46:50,866][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:46:51,373][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:46:51,877][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:46:52,375][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:46:52,875][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:46:53,377][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:46:53,882][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:46:54,383][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:46:54,882][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:46:55,380][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:46:55,879][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:46:56,378][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:46:56,875][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:46:57,374][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:46:57,872][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:46:58,371][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:46:58,870][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:46:59,368][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:46:59,866][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:47:00,363][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:47:00,860][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:47:01,358][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:47:01,857][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:47:02,355][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:47:02,852][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:47:03,353][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:47:03,854][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:47:04,354][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:47:04,852][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:47:05,350][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:47:05,847][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:47:06,344][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:47:06,841][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:47:07,339][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:47:07,840][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:47:08,338][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9834 tokens. +[2026-03-25 22:47:09,019][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.05%, Current % of VRAM taken: 60.47%, Block Peak % of device VRAM: 61.94%, ΔTime: 00:00:32 +[2026-03-25 22:47:09,776][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:47:09,778][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:47:09,780][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:47:10,426][__main__][INFO] - Iteration 53 took 49s (28.82% Gen, 69.88% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 49m 15s. Estimated total time: 41h 35m 37s. Time estimates for 10 more iterations: 8m 19s, 100 more iterations: 1h 23m 11s, 500 more iterations: 6h 55m 56s. +[2026-03-25 22:47:10,428][__main__][INFO] - Starting iteration 53. +[2026-03-25 22:47:10,828][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 5 and human policies 1. +[2026-03-25 22:47:10,829][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:47:25,177][__main__][INFO] - Number of regex retries in iteration 53: 0 +[2026-03-25 22:47:25,178][__main__][INFO] - agents played in iteration 53 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:47:25,961][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:47:25,984][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:47:26,008][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:47:26,031][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:47:26,031][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:47:26,032][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:47:26,653][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:47:27,103][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:47:27,604][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:47:28,098][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:47:28,593][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:47:29,087][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:47:29,587][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:47:30,084][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:47:30,582][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:47:31,080][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:47:31,577][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:47:32,073][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:47:32,574][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:47:33,071][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:47:33,568][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:47:34,063][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:47:34,559][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:47:35,058][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:47:35,557][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:47:36,055][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:47:36,555][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:47:37,056][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:47:37,576][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:47:38,078][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:47:38,578][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:47:39,078][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:47:39,581][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:47:40,086][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:47:40,587][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:47:41,084][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:47:41,584][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:47:42,083][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:47:42,579][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:47:43,078][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:47:43,578][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:47:44,082][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:47:44,580][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:47:45,077][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:47:45,575][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:47:46,074][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:47:46,573][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:47:47,069][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:47:47,566][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:47:48,066][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:47:48,563][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:47:49,060][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:47:49,557][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:47:50,054][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:47:50,552][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:47:51,047][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:47:51,545][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:47:52,039][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:47:52,533][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:47:53,028][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:47:53,524][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:47:54,020][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:47:54,515][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:47:55,012][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:47:55,509][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:47:56,006][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:47:56,502][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:47:57,000][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:47:57,498][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:47:57,995][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:47:58,493][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9679 tokens. +[2026-03-25 22:47:59,178][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 61.81%, ΔTime: 00:00:32 +[2026-03-25 22:47:59,940][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:47:59,942][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:47:59,944][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:48:00,601][__main__][INFO] - Iteration 54 took 49s (28.83% Gen, 69.85% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 41m 27s. Estimated total time: 41h 28m 39s. Time estimates for 10 more iterations: 8m 17s, 100 more iterations: 1h 22m 57s, 500 more iterations: 6h 54m 46s. +[2026-03-25 22:48:00,603][__main__][INFO] - Starting iteration 54. +[2026-03-25 22:48:01,004][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 5 and human policies 1. +[2026-03-25 22:48:01,005][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:48:05,767][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:48:15,440][__main__][INFO] - Number of regex retries in iteration 54: 1 +[2026-03-25 22:48:15,441][__main__][INFO] - agents played in iteration 54 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:48:16,229][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:48:16,253][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:48:16,277][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:48:16,300][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:48:16,300][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:48:16,301][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:48:16,946][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:48:17,400][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:48:17,907][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:48:18,407][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:48:18,906][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:48:19,403][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:48:19,899][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:48:20,400][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:48:20,899][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:48:21,399][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:48:21,898][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:48:22,400][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:48:22,902][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:48:23,401][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:48:23,899][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:48:24,421][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:48:24,923][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:48:25,421][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:48:25,919][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:48:26,417][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:48:26,916][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:48:27,414][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:48:27,916][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:48:28,416][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:48:28,921][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:48:29,431][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:48:29,933][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:48:30,434][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:48:30,932][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:48:31,432][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:48:31,929][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:48:32,430][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:48:32,925][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:48:33,421][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:48:33,916][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:48:34,413][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:48:34,909][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:48:35,406][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:48:35,904][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:48:36,402][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:48:36,900][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:48:37,396][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:48:37,894][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:48:38,393][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:48:38,891][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:48:39,391][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:48:39,889][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:48:40,384][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:48:40,880][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:48:41,378][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:48:41,875][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:48:42,379][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:48:42,876][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:48:43,378][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:48:43,877][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:48:44,375][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:48:44,874][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:48:45,370][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:48:45,866][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:48:46,362][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:48:46,856][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:48:47,352][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:48:47,846][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:48:48,340][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:48:48,835][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9939 tokens. +[2026-03-25 22:48:49,508][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.02%, Current % of VRAM taken: 60.45%, Block Peak % of device VRAM: 61.95%, ΔTime: 00:00:32 +[2026-03-25 22:48:50,271][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:48:50,274][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:48:50,275][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:48:50,925][__main__][INFO] - Iteration 55 took 49s (28.92% Gen, 69.78% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 48m 0s. Estimated total time: 41h 36m 2s. Time estimates for 10 more iterations: 8m 19s, 100 more iterations: 1h 23m 12s, 500 more iterations: 6h 56m 0s. +[2026-03-25 22:48:50,927][__main__][INFO] - Starting iteration 55. +[2026-03-25 22:48:51,328][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 5 and human policies 1. +[2026-03-25 22:48:51,329][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:48:53,038][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:49:05,832][__main__][INFO] - Number of regex retries in iteration 55: 1 +[2026-03-25 22:49:05,833][__main__][INFO] - agents played in iteration 55 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:49:06,621][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:49:06,645][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:49:06,668][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:49:06,691][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:49:06,692][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:49:06,693][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:49:07,359][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:49:07,811][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:49:08,312][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:49:08,810][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:49:09,309][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:49:09,806][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:49:10,303][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:49:10,804][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:49:11,301][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:49:11,797][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:49:12,295][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:49:12,793][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:49:13,292][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:49:13,789][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:49:14,286][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:49:14,784][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:49:15,283][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:49:15,785][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:49:16,308][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:49:16,806][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:49:17,305][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:49:17,809][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:49:18,318][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:49:18,823][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:49:19,328][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:49:19,825][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:49:20,324][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:49:20,822][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:49:21,321][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:49:21,819][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:49:22,318][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:49:22,819][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:49:23,317][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:49:23,814][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:49:24,310][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:49:24,807][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:49:25,303][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:49:25,798][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:49:26,293][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:49:26,791][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:49:27,287][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:49:27,784][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:49:28,283][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:49:28,782][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:49:29,280][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:49:29,779][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:49:30,277][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:49:30,775][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:49:31,274][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:49:31,773][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:49:32,272][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:49:32,775][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:49:33,276][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:49:33,775][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:49:34,277][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:49:34,775][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:49:35,274][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:49:35,772][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:49:36,271][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:49:36,768][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:49:37,267][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:49:37,764][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:49:38,262][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:49:38,759][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:49:39,258][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9890 tokens. +[2026-03-25 22:49:39,953][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.04%, Current % of VRAM taken: 60.47%, Block Peak % of device VRAM: 61.93%, ΔTime: 00:00:32 +[2026-03-25 22:49:40,759][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:49:40,762][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:49:40,763][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:49:41,417][__main__][INFO] - Iteration 56 took 50s (28.96% Gen, 69.74% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 55m 34s. Estimated total time: 41h 44m 26s. Time estimates for 10 more iterations: 8m 20s, 100 more iterations: 1h 23m 28s, 500 more iterations: 6h 57m 24s. +[2026-03-25 22:49:41,419][__main__][INFO] - Starting iteration 56. +[2026-03-25 22:49:41,819][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 5 and human policies 1. +[2026-03-25 22:49:41,819][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:49:55,775][__main__][INFO] - Number of regex retries in iteration 56: 0 +[2026-03-25 22:49:55,776][__main__][INFO] - agents played in iteration 56 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:49:56,570][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:49:56,594][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:49:56,617][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:49:56,640][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:49:56,641][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:49:56,642][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:49:57,250][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:49:57,706][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:49:58,214][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:49:58,713][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:49:59,212][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:49:59,713][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:50:00,210][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:50:00,707][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:50:01,204][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:50:01,702][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:50:02,200][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:50:02,697][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:50:03,195][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:50:03,693][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:50:04,188][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:50:04,686][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:50:05,183][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:50:05,680][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:50:06,179][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:50:06,677][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:50:07,178][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:50:07,682][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:50:08,182][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:50:08,684][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:50:09,182][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:50:09,683][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:50:10,182][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:50:10,679][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:50:11,176][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:50:11,673][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:50:12,172][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:50:12,669][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:50:13,165][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:50:13,663][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:50:14,162][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:50:14,661][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:50:15,158][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:50:15,655][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:50:16,153][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:50:16,650][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:50:17,148][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:50:17,644][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:50:18,141][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:50:18,639][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:50:19,135][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:50:19,634][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:50:20,133][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:50:20,629][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:50:21,125][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:50:21,623][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:50:22,120][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:50:22,618][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:50:23,117][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:50:23,614][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:50:24,113][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:50:24,611][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:50:25,109][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:50:25,607][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:50:26,104][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:50:26,601][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:50:27,098][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:50:27,597][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:50:28,095][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:50:28,594][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:50:29,093][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9934 tokens. +[2026-03-25 22:50:29,792][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 61.85%, ΔTime: 00:00:32 +[2026-03-25 22:50:30,552][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:50:30,555][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:50:30,556][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:50:31,197][__main__][INFO] - Iteration 57 took 49s (28.26% Gen, 70.44% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 19m 13s. Estimated total time: 41h 8m 56s. Time estimates for 10 more iterations: 8m 13s, 100 more iterations: 1h 22m 17s, 500 more iterations: 6h 51m 29s. +[2026-03-25 22:50:31,199][__main__][INFO] - Starting iteration 57. +[2026-03-25 22:50:31,600][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 5 and human policies 1. +[2026-03-25 22:50:31,601][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:50:43,696][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:50:45,559][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:50:46,420][__main__][INFO] - Number of regex retries in iteration 57: 2 +[2026-03-25 22:50:46,421][__main__][INFO] - agents played in iteration 57 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:50:47,226][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:50:47,250][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:50:47,273][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:50:47,296][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:50:47,296][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:50:47,297][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:50:47,920][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:50:48,378][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:50:48,904][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:50:49,404][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:50:49,904][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:50:50,403][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:50:50,902][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:50:51,405][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:50:51,905][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:50:52,405][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:50:52,902][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:50:53,400][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:50:53,896][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:50:54,395][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:50:54,891][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:50:55,389][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:50:55,887][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:50:56,384][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:50:56,882][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:50:57,380][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:50:57,884][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:50:58,385][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:50:58,885][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:50:59,382][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:50:59,883][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:51:00,385][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:51:00,882][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:51:01,388][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:51:01,893][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:51:02,392][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:51:02,890][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:51:03,388][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:51:03,886][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:51:04,384][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:51:04,882][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:51:05,380][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:51:05,878][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:51:06,375][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:51:06,873][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:51:07,371][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:51:07,867][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:51:08,363][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:51:08,861][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:51:09,361][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:51:09,858][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:51:10,355][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:51:10,855][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:51:11,354][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:51:11,852][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:51:12,349][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:51:12,846][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:51:13,343][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:51:13,840][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:51:14,338][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:51:14,834][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:51:15,330][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:51:15,825][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:51:16,322][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:51:16,819][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:51:17,317][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:51:17,816][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:51:18,314][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:51:18,811][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:51:19,306][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:51:19,803][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9999 tokens. +[2026-03-25 22:51:20,485][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.06%, ΔTime: 00:00:32 +[2026-03-25 22:51:21,238][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:51:21,240][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:51:21,242][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:51:21,885][__main__][INFO] - Iteration 58 took 50s (29.47% Gen, 69.25% Train). Generation: 14s, Training: 34s. Estimated remaining time: 41h 3m 42s. Estimated total time: 41h 54m 15s. Time estimates for 10 more iterations: 8m 22s, 100 more iterations: 1h 23m 48s, 500 more iterations: 6h 59m 2s. +[2026-03-25 22:51:21,887][__main__][INFO] - Starting iteration 58. +[2026-03-25 22:51:22,289][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 5 and human policies 1. +[2026-03-25 22:51:22,290][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:51:31,150][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:51:36,176][__main__][INFO] - Number of regex retries in iteration 58: 1 +[2026-03-25 22:51:36,177][__main__][INFO] - agents played in iteration 58 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:51:36,968][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:51:36,992][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:51:37,016][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:51:37,039][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:51:37,040][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:51:37,040][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:51:37,668][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:51:38,121][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:51:38,623][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:51:39,120][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:51:39,618][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:51:40,116][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:51:40,614][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:51:41,114][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:51:41,613][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:51:42,115][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:51:42,613][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:51:43,110][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:51:43,624][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:51:44,125][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:51:44,622][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:51:45,119][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:51:45,615][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:51:46,112][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:51:46,608][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:51:47,103][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:51:47,600][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:51:48,101][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:51:48,601][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:51:49,106][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:51:49,610][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:51:50,111][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:51:50,609][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:51:51,106][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:51:51,604][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:51:52,104][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:51:52,603][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:51:53,099][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:51:53,599][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:51:54,097][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:51:54,594][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:51:55,090][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:51:55,587][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:51:56,085][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:51:56,581][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:51:57,079][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:51:57,576][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:51:58,072][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:51:58,568][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:51:59,066][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:51:59,562][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:52:00,058][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:52:00,554][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:52:01,051][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:52:01,548][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:52:02,044][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:52:02,542][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:52:03,039][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:52:03,536][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:52:04,031][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:52:04,529][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:52:05,026][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:52:05,522][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:52:06,019][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:52:06,516][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:52:07,014][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:52:07,510][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:52:08,007][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:52:08,506][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:52:09,003][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:52:09,500][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9919 tokens. +[2026-03-25 22:52:10,179][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 61.81%, ΔTime: 00:00:32 +[2026-03-25 22:52:10,939][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:52:10,941][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:52:10,943][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:52:11,583][__main__][INFO] - Iteration 59 took 49s (28.17% Gen, 70.52% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 13m 26s. Estimated total time: 41h 4m 49s. Time estimates for 10 more iterations: 8m 12s, 100 more iterations: 1h 22m 9s, 500 more iterations: 6h 50m 48s. +[2026-03-25 22:52:11,586][__main__][INFO] - Starting iteration 59. +[2026-03-25 22:52:11,983][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 5 and human policies 1. +[2026-03-25 22:52:11,984][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:52:19,971][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 20 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:52:26,467][__main__][INFO] - Number of regex retries in iteration 59: 1 +[2026-03-25 22:52:26,467][__main__][INFO] - agents played in iteration 59 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:52:27,255][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:52:27,279][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:52:27,302][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:52:27,325][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:52:27,326][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:52:27,326][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:52:27,940][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:52:28,394][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:52:28,897][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:52:29,397][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:52:29,896][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:52:30,395][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:52:30,892][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:52:31,393][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:52:31,891][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:52:32,389][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:52:32,886][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:52:33,384][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:52:33,882][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:52:34,381][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:52:34,880][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:52:35,377][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:52:35,875][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:52:36,373][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:52:36,869][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:52:37,366][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:52:37,867][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:52:38,369][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:52:38,877][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:52:39,377][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:52:39,878][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:52:40,380][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:52:40,879][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:52:41,380][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:52:41,883][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:52:42,383][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:52:42,883][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:52:43,383][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:52:43,881][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:52:44,380][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:52:44,900][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:52:45,402][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:52:45,904][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:52:46,401][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:52:46,900][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:52:47,401][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:52:47,899][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:52:48,396][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:52:48,892][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:52:49,387][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:52:49,883][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:52:50,380][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:52:50,877][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:52:51,374][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:52:51,871][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:52:52,369][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:52:52,868][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:52:53,365][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:52:53,863][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:52:54,361][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:52:54,862][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:52:55,361][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:52:55,859][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:52:56,358][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:52:56,856][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:52:57,354][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:52:57,858][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:52:58,356][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:52:58,855][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:52:59,354][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:52:59,853][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9969 tokens. +[2026-03-25 22:53:00,527][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 61.94%, ΔTime: 00:00:32 +[2026-03-25 22:53:01,285][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:53:01,287][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:53:01,289][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:53:01,930][__main__][INFO] - Iteration 60 took 49s (29.00% Gen, 69.72% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 45m 8s. Estimated total time: 41h 37m 22s. Time estimates for 10 more iterations: 8m 19s, 100 more iterations: 1h 23m 14s, 500 more iterations: 6h 56m 13s. +[2026-03-25 22:53:01,933][__main__][INFO] - Starting iteration 60. +[2026-03-25 22:53:02,331][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 5 and human policies 1. +[2026-03-25 22:53:02,332][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:53:06,671][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:53:12,595][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:53:16,420][__main__][INFO] - Number of regex retries in iteration 60: 2 +[2026-03-25 22:53:16,421][__main__][INFO] - agents played in iteration 60 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:53:17,215][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:53:17,239][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:53:17,262][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:53:17,285][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:53:17,285][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:53:17,286][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:53:17,929][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:53:18,383][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:53:18,914][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:53:19,415][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:53:19,913][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:53:20,414][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:53:20,913][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:53:21,411][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:53:21,908][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:53:22,403][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:53:22,899][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:53:23,394][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:53:23,889][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:53:24,384][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:53:24,879][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:53:25,376][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:53:25,871][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:53:26,367][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:53:26,864][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:53:27,364][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:53:27,865][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:53:28,365][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:53:28,865][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:53:29,362][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:53:29,860][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:53:30,364][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:53:30,867][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:53:31,365][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:53:31,864][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:53:32,362][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:53:32,862][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:53:33,359][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:53:33,858][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:53:34,357][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:53:34,854][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:53:35,352][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:53:35,848][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:53:36,344][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:53:36,839][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:53:37,335][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:53:37,831][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:53:38,327][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:53:38,825][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:53:39,321][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:53:39,818][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:53:40,315][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:53:40,813][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:53:41,312][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:53:41,809][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:53:42,305][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:53:42,804][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:53:43,305][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:53:43,805][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:53:44,303][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:53:44,801][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:53:45,298][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:53:45,796][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:53:46,292][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:53:46,789][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:53:47,284][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:53:47,782][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:53:48,281][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:53:48,777][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:53:49,274][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:53:49,769][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9799 tokens. +[2026-03-25 22:53:50,456][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.98%, Current % of VRAM taken: 60.41%, Block Peak % of device VRAM: 61.85%, ΔTime: 00:00:32 +[2026-03-25 22:53:51,110][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:53:51,112][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:53:51,114][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:53:52,345][__main__][INFO] - Iteration 61 took 50s (28.17% Gen, 69.37% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 47m 40s. Estimated total time: 41h 40m 43s. Time estimates for 10 more iterations: 8m 20s, 100 more iterations: 1h 23m 21s, 500 more iterations: 6h 56m 47s. +[2026-03-25 22:53:52,348][__main__][INFO] - Starting iteration 61. +[2026-03-25 22:53:52,750][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 6 and human policies 1. +[2026-03-25 22:53:52,750][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:54:10,497][__main__][INFO] - Number of regex retries in iteration 61: 0 +[2026-03-25 22:54:10,498][__main__][INFO] - agents played in iteration 61 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:54:11,280][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:54:11,303][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:54:11,327][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:54:11,349][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:54:11,350][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:54:11,351][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:54:11,957][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:54:12,409][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:54:12,912][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:54:13,416][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:54:13,917][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:54:14,415][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:54:14,916][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:54:15,417][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:54:15,917][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:54:16,413][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:54:16,908][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:54:17,403][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:54:17,902][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:54:18,400][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:54:18,900][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:54:19,400][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:54:19,896][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:54:20,400][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:54:20,900][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:54:21,398][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:54:21,895][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:54:22,394][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:54:22,895][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:54:23,394][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:54:23,890][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:54:24,385][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:54:24,881][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:54:25,377][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:54:25,873][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:54:26,368][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:54:26,862][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:54:27,357][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:54:27,853][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:54:28,349][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:54:28,846][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:54:29,343][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:54:29,840][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:54:30,338][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:54:30,834][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:54:31,331][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:54:31,833][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:54:32,330][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:54:32,827][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:54:33,325][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:54:33,821][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:54:34,317][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:54:34,814][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:54:35,312][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:54:35,808][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:54:36,305][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:54:36,801][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:54:37,299][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:54:37,796][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:54:38,293][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:54:38,791][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:54:39,286][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:54:39,783][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:54:40,278][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:54:40,775][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:54:41,274][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:54:41,770][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:54:42,266][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:54:42,760][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:54:43,258][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:54:43,756][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9789 tokens. +[2026-03-25 22:54:44,416][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.00%, Current % of VRAM taken: 60.42%, Block Peak % of device VRAM: 62.10%, ΔTime: 00:00:32 +[2026-03-25 22:54:45,172][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:54:45,174][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:54:45,176][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:54:45,839][__main__][INFO] - Iteration 62 took 53s (33.43% Gen, 65.32% Train). Generation: 17s, Training: 34s. Estimated remaining time: 43h 20m 33s. Estimated total time: 44h 14m 30s. Time estimates for 10 more iterations: 8m 50s, 100 more iterations: 1h 28m 29s, 500 more iterations: 7h 22m 25s. +[2026-03-25 22:54:45,842][__main__][INFO] - Starting iteration 62. +[2026-03-25 22:54:46,244][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 6 and human policies 1. +[2026-03-25 22:54:46,245][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:54:51,286][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:55:00,464][__main__][INFO] - Number of regex retries in iteration 62: 1 +[2026-03-25 22:55:00,465][__main__][INFO] - agents played in iteration 62 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:55:01,271][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:55:01,294][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:55:01,317][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:55:01,340][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:55:01,341][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:55:01,342][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:55:01,954][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:55:02,406][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:55:02,912][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:55:03,409][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:55:03,909][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:55:04,407][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:55:04,904][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:55:05,401][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:55:05,899][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:55:06,398][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:55:06,894][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:55:07,392][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:55:07,890][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:55:08,388][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:55:08,885][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:55:09,382][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:55:09,879][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:55:10,376][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:55:10,878][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:55:11,380][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:55:11,885][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:55:12,392][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:55:12,892][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:55:13,392][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:55:13,888][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:55:14,384][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:55:14,881][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:55:15,378][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:55:15,876][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:55:16,373][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:55:16,871][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:55:17,367][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:55:17,866][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:55:18,366][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:55:18,867][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:55:19,365][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:55:19,867][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:55:20,366][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:55:20,863][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:55:21,361][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:55:21,860][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:55:22,358][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:55:22,857][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:55:23,355][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:55:23,854][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:55:24,353][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:55:24,854][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:55:25,354][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:55:25,853][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:55:26,349][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:55:26,846][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:55:27,344][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:55:27,840][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:55:28,337][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:55:28,834][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:55:29,332][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:55:29,829][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:55:30,327][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:55:30,825][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:55:31,323][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:55:31,820][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:55:32,318][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:55:32,816][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:55:33,314][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:55:33,814][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9940 tokens. +[2026-03-25 22:55:34,496][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 61.87%, ΔTime: 00:00:32 +[2026-03-25 22:55:35,251][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:55:35,254][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:55:35,255][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:55:35,900][__main__][INFO] - Iteration 63 took 49s (28.64% Gen, 70.06% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 28m 2s. Estimated total time: 41h 22m 50s. Time estimates for 10 more iterations: 8m 16s, 100 more iterations: 1h 22m 45s, 500 more iterations: 6h 53m 48s. +[2026-03-25 22:55:35,902][__main__][INFO] - Starting iteration 63. +[2026-03-25 22:55:36,306][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 6 and human policies 1. +[2026-03-25 22:55:36,306][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:55:50,998][__main__][INFO] - Number of regex retries in iteration 63: 0 +[2026-03-25 22:55:50,999][__main__][INFO] - agents played in iteration 63 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:55:51,803][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:55:51,827][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:55:51,850][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:55:51,873][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:55:51,873][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:55:51,874][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:55:52,513][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:55:52,969][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:55:53,477][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:55:53,978][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:55:54,477][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:55:54,977][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:55:55,477][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:55:55,978][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:55:56,476][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:55:56,976][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:55:57,478][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:55:57,979][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:55:58,480][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:55:58,980][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:55:59,480][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:55:59,982][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:56:00,482][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:56:00,983][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:56:01,488][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:56:01,992][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:56:02,502][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:56:03,002][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:56:03,500][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:56:03,998][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:56:04,498][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:56:05,003][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:56:05,502][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:56:06,001][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:56:06,499][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:56:06,998][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:56:07,496][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:56:07,995][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:56:08,495][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:56:08,995][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:56:09,493][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:56:09,990][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:56:10,489][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:56:10,986][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:56:11,484][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:56:11,984][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:56:12,483][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:56:12,980][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:56:13,476][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:56:13,972][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:56:14,468][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:56:14,965][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:56:15,461][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:56:15,957][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:56:16,451][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:56:16,973][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:56:17,474][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:56:17,978][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:56:18,478][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:56:18,980][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:56:19,484][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:56:19,983][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:56:20,487][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:56:20,985][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:56:21,483][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:56:21,982][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:56:22,479][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:56:22,978][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:56:23,478][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:56:23,977][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:56:24,474][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9903 tokens. +[2026-03-25 22:56:25,173][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.08%, ΔTime: 00:00:32 +[2026-03-25 22:56:25,975][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:56:25,978][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:56:25,979][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:56:26,637][__main__][INFO] - Iteration 64 took 50s (29.19% Gen, 69.50% Train). Generation: 14s, Training: 34s. Estimated remaining time: 41h 1m 0s. Estimated total time: 41h 56m 38s. Time estimates for 10 more iterations: 8m 23s, 100 more iterations: 1h 23m 53s, 500 more iterations: 6h 59m 26s. +[2026-03-25 22:56:26,640][__main__][INFO] - Starting iteration 64. +[2026-03-25 22:56:27,042][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 6 and human policies 1. +[2026-03-25 22:56:27,042][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:56:29,009][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:56:41,842][__main__][INFO] - Number of regex retries in iteration 64: 1 +[2026-03-25 22:56:41,843][__main__][INFO] - agents played in iteration 64 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:56:42,637][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:56:42,661][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:56:42,684][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:56:42,707][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:56:42,708][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:56:42,708][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:56:43,360][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:56:43,813][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:56:44,315][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:56:44,815][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:56:45,311][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:56:45,809][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:56:46,309][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:56:46,808][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:56:47,306][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:56:47,806][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:56:48,305][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:56:48,802][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:56:49,298][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:56:49,794][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:56:50,296][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:56:50,794][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:56:51,296][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:56:51,796][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:56:52,304][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:56:52,805][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:56:53,304][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:56:53,806][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:56:54,304][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:56:54,803][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:56:55,303][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:56:55,801][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:56:56,299][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:56:56,800][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:56:57,297][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:56:57,794][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:56:58,291][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:56:58,788][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:56:59,286][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:56:59,783][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:57:00,281][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:57:00,778][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:57:01,275][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:57:01,772][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:57:02,268][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:57:02,763][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:57:03,259][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:57:03,754][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:57:04,250][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:57:04,746][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:57:05,241][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:57:05,736][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:57:06,233][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:57:06,729][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:57:07,225][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:57:07,721][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:57:08,219][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:57:08,714][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:57:09,211][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:57:09,705][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:57:10,201][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:57:10,696][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:57:11,191][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:57:11,688][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:57:12,186][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:57:12,683][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:57:13,181][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:57:13,680][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:57:14,176][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:57:14,672][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:57:15,173][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9822 tokens. +[2026-03-25 22:57:15,858][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.48%, Block Peak % of device VRAM: 61.76%, ΔTime: 00:00:32 +[2026-03-25 22:57:16,600][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:57:16,602][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:57:16,604][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:57:17,274][__main__][INFO] - Iteration 65 took 50s (29.46% Gen, 69.20% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 55m 11s. Estimated total time: 41h 51m 40s. Time estimates for 10 more iterations: 8m 22s, 100 more iterations: 1h 23m 43s, 500 more iterations: 6h 58m 36s. +[2026-03-25 22:57:17,277][__main__][INFO] - Starting iteration 65. +[2026-03-25 22:57:17,680][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 6 and human policies 1. +[2026-03-25 22:57:17,681][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:57:32,398][__main__][INFO] - Number of regex retries in iteration 65: 0 +[2026-03-25 22:57:32,398][__main__][INFO] - agents played in iteration 65 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:57:33,181][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:57:33,204][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:57:33,228][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:57:33,251][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:57:33,252][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:57:33,252][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:57:33,878][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:57:34,334][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:57:34,834][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:57:35,331][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:57:35,829][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:57:36,328][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:57:36,825][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:57:37,323][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:57:37,823][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:57:38,320][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:57:38,817][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:57:39,314][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:57:39,812][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:57:40,309][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:57:40,811][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:57:41,314][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:57:41,814][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:57:42,316][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:57:42,817][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:57:43,315][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:57:43,815][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:57:44,313][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:57:44,812][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:57:45,311][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:57:45,809][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:57:46,309][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:57:46,830][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:57:47,333][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:57:47,833][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:57:48,336][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:57:48,839][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:57:49,337][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:57:49,836][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:57:50,339][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:57:50,837][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:57:51,338][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:57:51,837][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:57:52,335][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:57:52,833][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:57:53,332][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:57:53,831][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:57:54,327][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:57:54,823][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:57:55,321][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:57:55,818][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:57:56,316][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:57:56,813][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:57:57,310][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:57:57,807][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:57:58,305][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:57:58,804][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:57:59,302][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:57:59,800][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:58:00,298][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:58:00,797][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:58:01,296][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:58:01,794][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:58:02,292][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:58:02,790][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:58:03,288][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:58:03,786][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:58:04,284][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:58:04,782][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:58:05,280][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:58:05,778][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9997 tokens. +[2026-03-25 22:58:06,454][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.07%, ΔTime: 00:00:32 +[2026-03-25 22:58:07,242][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:58:07,244][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:58:07,246][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:58:07,945][__main__][INFO] - Iteration 66 took 50s (29.28% Gen, 69.33% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 56m 0s. Estimated total time: 41h 53m 19s. Time estimates for 10 more iterations: 8m 22s, 100 more iterations: 1h 23m 46s, 500 more iterations: 6h 58m 53s. +[2026-03-25 22:58:07,947][__main__][INFO] - Starting iteration 66. +[2026-03-25 22:58:08,349][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 6 and human policies 1. +[2026-03-25 22:58:08,349][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:58:15,350][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 22:58:23,391][__main__][INFO] - Number of regex retries in iteration 66: 1 +[2026-03-25 22:58:23,392][__main__][INFO] - agents played in iteration 66 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:58:24,168][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:58:24,191][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:58:24,214][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:58:24,237][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:58:24,238][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:58:24,239][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:58:24,854][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:58:25,308][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:58:25,810][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:58:26,307][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:58:26,805][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:58:27,302][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:58:27,802][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:58:28,300][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:58:28,799][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:58:29,297][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:58:29,801][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:58:30,300][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:58:30,803][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:58:31,301][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:58:31,799][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:58:32,298][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:58:32,797][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:58:33,296][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:58:33,793][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:58:34,290][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:58:34,787][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:58:35,286][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:58:35,783][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:58:36,280][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:58:36,778][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:58:37,276][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:58:37,774][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:58:38,272][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:58:38,769][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:58:39,266][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:58:39,764][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:58:40,260][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:58:40,758][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:58:41,261][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:58:41,763][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:58:42,260][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:58:42,760][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:58:43,258][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:58:43,758][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:58:44,256][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:58:44,757][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:58:45,255][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:58:45,751][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:58:46,248][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:58:46,745][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:58:47,242][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:58:47,738][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:58:48,237][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:58:48,735][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:58:49,233][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:58:49,733][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:58:50,234][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:58:50,732][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:58:51,228][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:58:51,725][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:58:52,221][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:58:52,719][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:58:53,215][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:58:53,712][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:58:54,208][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:58:54,706][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:58:55,204][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:58:55,701][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:58:56,197][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:58:56,693][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9907 tokens. +[2026-03-25 22:58:57,372][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.95%, Current % of VRAM taken: 60.38%, Block Peak % of device VRAM: 61.98%, ΔTime: 00:00:32 +[2026-03-25 22:58:58,132][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:58:58,134][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:58:58,136][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:58:58,780][__main__][INFO] - Iteration 67 took 50s (29.83% Gen, 68.89% Train). Generation: 15s, Training: 34s. Estimated remaining time: 41h 3m 24s. Estimated total time: 42h 1m 35s. Time estimates for 10 more iterations: 8m 24s, 100 more iterations: 1h 24m 3s, 500 more iterations: 7h 0m 15s. +[2026-03-25 22:58:58,782][__main__][INFO] - Starting iteration 67. +[2026-03-25 22:58:59,185][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 6 and human policies 1. +[2026-03-25 22:58:59,185][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 22:59:13,469][__main__][INFO] - Number of regex retries in iteration 67: 0 +[2026-03-25 22:59:13,469][__main__][INFO] - agents played in iteration 67 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 22:59:14,239][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:59:14,263][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:59:14,286][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:59:14,309][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 22:59:14,310][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 22:59:14,310][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 22:59:14,941][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 22:59:15,395][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 22:59:15,898][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 22:59:16,398][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 22:59:16,897][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 22:59:17,396][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 22:59:17,893][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 22:59:18,392][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 22:59:18,893][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 22:59:19,394][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 22:59:19,898][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 22:59:20,400][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 22:59:20,898][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 22:59:21,392][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 22:59:21,888][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 22:59:22,384][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 22:59:22,885][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 22:59:23,382][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 22:59:23,880][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 22:59:24,377][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 22:59:24,877][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 22:59:25,376][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 22:59:25,875][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 22:59:26,379][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 22:59:26,880][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 22:59:27,378][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 22:59:27,875][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 22:59:28,371][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 22:59:28,867][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 22:59:29,363][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 22:59:29,861][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 22:59:30,357][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 22:59:30,854][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 22:59:31,349][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 22:59:31,844][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 22:59:32,338][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 22:59:32,834][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 22:59:33,330][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 22:59:33,824][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 22:59:34,320][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 22:59:34,816][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 22:59:35,316][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 22:59:35,815][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 22:59:36,336][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 22:59:36,835][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 22:59:37,334][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 22:59:37,834][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 22:59:38,331][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 22:59:38,830][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 22:59:39,325][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 22:59:39,820][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 22:59:40,316][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 22:59:40,812][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 22:59:41,307][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 22:59:41,804][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 22:59:42,300][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 22:59:42,796][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 22:59:43,293][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 22:59:43,791][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 22:59:44,288][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 22:59:44,784][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 22:59:45,281][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 22:59:45,777][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 22:59:46,275][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 22:59:46,773][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9935 tokens. +[2026-03-25 22:59:47,425][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 61.91%, ΔTime: 00:00:32 +[2026-03-25 22:59:48,166][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 22:59:48,168][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 22:59:48,170][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 22:59:48,815][__main__][INFO] - Iteration 68 took 49s (28.78% Gen, 69.92% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 22m 33s. Estimated total time: 41h 21m 33s. Time estimates for 10 more iterations: 8m 16s, 100 more iterations: 1h 22m 43s, 500 more iterations: 6h 53m 35s. +[2026-03-25 22:59:48,818][__main__][INFO] - Starting iteration 68. +[2026-03-25 22:59:49,216][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 6 and human policies 1. +[2026-03-25 22:59:49,217][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:00:03,077][__main__][INFO] - Number of regex retries in iteration 68: 0 +[2026-03-25 23:00:03,079][__main__][INFO] - agents played in iteration 68 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:00:03,849][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:00:03,872][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:00:03,896][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:00:03,919][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:00:03,919][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:00:03,920][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:00:04,540][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:00:04,995][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:00:05,498][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:00:06,000][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:00:06,499][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:00:06,997][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:00:07,494][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:00:07,993][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:00:08,491][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:00:08,990][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:00:09,495][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:00:10,000][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:00:10,507][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:00:11,007][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:00:11,506][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:00:12,008][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:00:12,507][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:00:13,004][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:00:13,501][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:00:14,001][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:00:14,500][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:00:14,999][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:00:15,497][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:00:15,994][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:00:16,492][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:00:16,989][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:00:17,485][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:00:17,983][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:00:18,479][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:00:18,976][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:00:19,472][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:00:19,970][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:00:20,468][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:00:20,964][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:00:21,461][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:00:21,959][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:00:22,456][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:00:22,954][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:00:23,453][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:00:23,949][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:00:24,445][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:00:24,943][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:00:25,441][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:00:25,938][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:00:26,435][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:00:26,931][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:00:27,427][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:00:27,924][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:00:28,421][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:00:28,919][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:00:29,419][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:00:29,916][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:00:30,412][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:00:30,908][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:00:31,404][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:00:31,902][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:00:32,399][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:00:32,897][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:00:33,394][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:00:33,892][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:00:34,390][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:00:34,886][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:00:35,384][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:00:35,881][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:00:36,378][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9985 tokens. +[2026-03-25 23:00:37,061][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.05%, Current % of VRAM taken: 60.47%, Block Peak % of device VRAM: 61.95%, ΔTime: 00:00:32 +[2026-03-25 23:00:37,827][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:00:37,829][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:00:37,831][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:00:38,520][__main__][INFO] - Iteration 69 took 49s (28.11% Gen, 70.49% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 5m 21s. Estimated total time: 41h 5m 11s. Time estimates for 10 more iterations: 8m 13s, 100 more iterations: 1h 22m 10s, 500 more iterations: 6h 50m 51s. +[2026-03-25 23:00:38,522][__main__][INFO] - Starting iteration 69. +[2026-03-25 23:00:38,922][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 6 and human policies 1. +[2026-03-25 23:00:38,923][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:00:40,872][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:00:52,743][__main__][INFO] - Number of regex retries in iteration 69: 1 +[2026-03-25 23:00:52,744][__main__][INFO] - agents played in iteration 69 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:00:53,515][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:00:53,538][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:00:53,562][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:00:53,585][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:00:53,585][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:00:53,586][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:00:54,210][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:00:54,669][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:00:55,172][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:00:55,670][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:00:56,167][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:00:56,663][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:00:57,161][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:00:57,658][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:00:58,160][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:00:58,661][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:00:59,169][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:00:59,672][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:01:00,176][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:01:00,681][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:01:01,183][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:01:01,684][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:01:02,185][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:01:02,684][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:01:03,182][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:01:03,680][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:01:04,182][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:01:04,686][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:01:05,186][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:01:05,686][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:01:06,187][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:01:06,686][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:01:07,185][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:01:07,682][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:01:08,180][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:01:08,678][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:01:09,174][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:01:09,671][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:01:10,167][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:01:10,666][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:01:11,164][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:01:11,663][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:01:12,162][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:01:12,661][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:01:13,159][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:01:13,656][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:01:14,154][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:01:14,650][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:01:15,147][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:01:15,643][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:01:16,137][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:01:16,636][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:01:17,137][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:01:17,636][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:01:18,137][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:01:18,635][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:01:19,134][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:01:19,631][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:01:20,131][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:01:20,629][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:01:21,125][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:01:21,622][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:01:22,119][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:01:22,618][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:01:23,116][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:01:23,616][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:01:24,113][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:01:24,610][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:01:25,109][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:01:25,607][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:01:26,104][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9940 tokens. +[2026-03-25 23:01:26,777][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 61.89%, ΔTime: 00:00:32 +[2026-03-25 23:01:27,521][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:01:27,524][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:01:27,525][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:01:28,572][__main__][INFO] - Iteration 70 took 49s (27.84% Gen, 70.05% Train). Generation: 13s, Training: 34s. Estimated remaining time: 40h 21m 51s. Estimated total time: 41h 22m 31s. Time estimates for 10 more iterations: 8m 16s, 100 more iterations: 1h 22m 45s, 500 more iterations: 6h 53m 45s. +[2026-03-25 23:01:28,574][__main__][INFO] - Starting iteration 70. +[2026-03-25 23:01:28,973][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 6 and human policies 1. +[2026-03-25 23:01:28,974][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:01:32,979][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:01:43,305][__main__][INFO] - Number of regex retries in iteration 70: 1 +[2026-03-25 23:01:43,306][__main__][INFO] - agents played in iteration 70 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:01:44,087][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:01:44,111][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:01:44,134][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:01:44,157][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:01:44,158][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:01:44,158][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:01:44,818][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:01:45,273][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:01:45,778][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:01:46,281][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:01:46,780][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:01:47,280][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:01:47,783][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:01:48,288][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:01:48,792][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:01:49,295][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:01:49,794][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:01:50,295][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:01:50,793][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:01:51,290][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:01:51,789][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:01:52,292][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:01:52,793][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:01:53,288][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:01:53,787][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:01:54,284][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:01:54,780][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:01:55,279][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:01:55,777][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:01:56,275][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:01:56,775][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:01:57,274][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:01:57,773][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:01:58,272][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:01:58,770][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:01:59,266][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:01:59,763][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:02:00,259][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:02:00,758][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:02:01,255][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:02:01,750][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:02:02,246][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:02:02,745][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:02:03,241][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:02:03,736][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:02:04,232][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:02:04,728][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:02:05,224][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:02:05,722][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:02:06,218][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:02:06,715][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:02:07,210][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:02:07,706][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:02:08,203][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:02:08,701][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:02:09,201][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:02:09,698][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:02:10,197][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:02:10,695][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:02:11,193][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:02:11,689][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:02:12,185][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:02:12,681][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:02:13,178][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:02:13,676][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:02:14,175][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:02:14,676][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:02:15,175][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:02:15,672][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:02:16,167][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:02:16,665][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9904 tokens. +[2026-03-25 23:02:17,360][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.99%, Current % of VRAM taken: 60.41%, Block Peak % of device VRAM: 61.80%, ΔTime: 00:00:32 +[2026-03-25 23:02:18,127][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:02:18,130][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:02:18,131][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:02:19,397][__main__][INFO] - Iteration 71 took 50s (28.42% Gen, 69.06% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 59m 42s. Estimated total time: 42h 1m 13s. Time estimates for 10 more iterations: 8m 24s, 100 more iterations: 1h 24m 2s, 500 more iterations: 7h 0m 12s. +[2026-03-25 23:02:19,399][__main__][INFO] - Starting iteration 71. +[2026-03-25 23:02:19,801][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 7 and human policies 1. +[2026-03-25 23:02:19,802][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:02:34,204][__main__][INFO] - Number of regex retries in iteration 71: 0 +[2026-03-25 23:02:34,205][__main__][INFO] - agents played in iteration 71 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:02:35,013][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:02:35,037][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:02:35,061][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:02:35,084][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:02:35,085][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:02:35,085][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:02:35,743][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:02:36,202][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:02:36,707][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:02:37,208][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:02:37,712][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:02:38,220][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:02:38,723][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:02:39,223][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:02:39,723][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:02:40,225][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:02:40,725][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:02:41,225][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:02:41,726][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:02:42,226][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:02:42,727][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:02:43,228][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:02:43,733][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:02:44,232][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:02:44,731][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:02:45,233][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:02:45,733][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:02:46,232][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:02:46,736][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:02:47,236][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:02:47,737][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:02:48,238][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:02:48,740][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:02:49,241][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:02:49,742][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:02:50,242][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:02:50,739][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:02:51,236][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:02:51,734][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:02:52,236][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:02:52,738][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:02:53,242][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:02:53,743][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:02:54,242][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:02:54,743][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:02:55,244][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:02:55,743][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:02:56,239][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:02:56,734][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:02:57,229][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:02:57,724][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:02:58,220][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:02:58,715][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:02:59,210][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:02:59,705][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:03:00,203][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:03:00,702][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:03:01,199][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:03:01,697][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:03:02,193][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:03:02,693][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:03:03,191][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:03:03,689][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:03:04,187][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:03:04,685][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:03:05,182][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:03:05,681][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:03:06,179][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:03:06,679][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:03:07,178][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:03:07,677][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10063 tokens. +[2026-03-25 23:03:08,364][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 61.94%, ΔTime: 00:00:32 +[2026-03-25 23:03:09,127][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:03:09,129][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:03:09,131][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:03:09,786][__main__][INFO] - Iteration 72 took 49s (28.81% Gen, 69.87% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 36m 58s. Estimated total time: 41h 39m 19s. Time estimates for 10 more iterations: 8m 19s, 100 more iterations: 1h 23m 18s, 500 more iterations: 6h 56m 33s. +[2026-03-25 23:03:09,789][__main__][INFO] - Starting iteration 72. +[2026-03-25 23:03:10,190][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 7 and human policies 1. +[2026-03-25 23:03:10,191][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:03:24,245][__main__][INFO] - Number of regex retries in iteration 72: 0 +[2026-03-25 23:03:24,246][__main__][INFO] - agents played in iteration 72 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:03:25,035][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:03:25,058][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:03:25,082][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:03:25,105][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:03:25,106][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:03:25,106][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:03:25,746][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:03:26,202][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:03:26,705][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:03:27,204][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:03:27,707][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:03:28,211][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:03:28,715][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:03:29,218][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:03:29,722][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:03:30,219][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:03:30,718][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:03:31,216][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:03:31,713][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:03:32,211][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:03:32,708][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:03:33,206][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:03:33,703][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:03:34,200][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:03:34,698][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:03:35,196][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:03:35,696][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:03:36,196][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:03:36,695][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:03:37,199][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:03:37,700][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:03:38,198][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:03:38,697][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:03:39,197][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:03:39,699][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:03:40,200][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:03:40,700][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:03:41,198][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:03:41,696][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:03:42,195][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:03:42,693][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:03:43,190][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:03:43,688][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:03:44,186][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:03:44,682][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:03:45,179][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:03:45,677][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:03:46,174][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:03:46,670][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:03:47,165][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:03:47,660][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:03:48,157][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:03:48,653][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:03:49,147][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:03:49,644][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:03:50,146][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:03:50,647][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:03:51,145][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:03:51,645][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:03:52,162][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:03:52,665][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:03:53,167][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:03:53,665][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:03:54,166][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:03:54,663][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:03:55,160][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:03:55,656][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:03:56,154][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:03:56,652][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:03:57,148][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:03:57,644][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9875 tokens. +[2026-03-25 23:03:58,336][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.02%, Current % of VRAM taken: 60.45%, Block Peak % of device VRAM: 61.81%, ΔTime: 00:00:32 +[2026-03-25 23:03:59,097][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:03:59,100][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:03:59,101][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:04:00,015][__main__][INFO] - Iteration 73 took 49s (28.21% Gen, 69.96% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 28m 5s. Estimated total time: 41h 31m 17s. Time estimates for 10 more iterations: 8m 18s, 100 more iterations: 1h 23m 2s, 500 more iterations: 6h 55m 12s. +[2026-03-25 23:04:00,017][__main__][INFO] - Starting iteration 73. +[2026-03-25 23:04:00,416][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 7 and human policies 1. +[2026-03-25 23:04:00,417][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:04:14,725][__main__][INFO] - Number of regex retries in iteration 73: 0 +[2026-03-25 23:04:14,727][__main__][INFO] - agents played in iteration 73 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:04:15,504][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:04:15,528][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:04:15,551][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:04:15,574][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:04:15,575][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:04:15,576][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:04:16,230][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:04:16,685][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:04:17,187][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:04:17,689][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:04:18,197][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:04:18,699][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:04:19,200][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:04:19,699][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:04:20,199][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:04:20,697][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:04:21,195][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:04:21,691][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:04:22,187][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:04:22,684][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:04:23,180][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:04:23,678][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:04:24,175][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:04:24,675][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:04:25,175][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:04:25,674][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:04:26,172][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:04:26,671][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:04:27,168][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:04:27,668][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:04:28,166][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:04:28,662][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:04:29,157][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:04:29,654][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:04:30,154][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:04:30,649][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:04:31,144][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:04:31,638][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:04:32,133][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:04:32,627][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:04:33,122][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:04:33,618][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:04:34,114][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:04:34,610][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:04:35,106][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:04:35,602][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:04:36,098][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:04:36,594][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:04:37,093][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:04:37,590][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:04:38,087][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:04:38,583][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:04:39,081][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:04:39,580][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:04:40,078][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:04:40,578][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:04:41,076][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:04:41,576][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:04:42,075][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:04:42,575][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:04:43,074][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:04:43,573][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:04:44,072][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:04:44,570][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:04:45,068][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:04:45,565][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:04:46,062][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:04:46,561][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:04:47,060][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:04:47,558][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:04:48,056][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9878 tokens. +[2026-03-25 23:04:48,741][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 61.80%, ΔTime: 00:00:32 +[2026-03-25 23:04:49,496][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:04:49,498][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:04:49,500][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:04:50,156][__main__][INFO] - Iteration 74 took 49s (28.77% Gen, 69.91% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 22m 59s. Estimated total time: 41h 27m 0s. Time estimates for 10 more iterations: 8m 17s, 100 more iterations: 1h 22m 54s, 500 more iterations: 6h 54m 30s. +[2026-03-25 23:04:50,158][__main__][INFO] - Starting iteration 74. +[2026-03-25 23:04:50,560][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 7 and human policies 1. +[2026-03-25 23:04:50,560][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:04:57,476][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls + +Given the per-item values, you have a significant advantage with hats and books compared to Bob, while the values for balls are similar. Since you have a much higher value for hats and Bob values books much more than you do, it's strategic to take all the hats and none of the books or balls. This maximizes your points based on your high value for hats. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:04:59,735][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:05:05,847][__main__][INFO] - Number of regex retries in iteration 74: 2 +[2026-03-25 23:05:05,848][__main__][INFO] - agents played in iteration 74 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:05:06,623][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:05:06,646][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:05:06,670][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:05:06,693][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:05:06,694][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:05:06,694][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:05:07,322][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:05:07,780][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:05:08,286][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:05:08,789][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:05:09,295][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:05:09,793][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:05:10,290][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:05:10,789][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:05:11,287][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:05:11,785][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:05:12,285][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:05:12,783][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:05:13,282][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:05:13,779][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:05:14,277][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:05:14,775][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:05:15,273][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:05:15,771][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:05:16,268][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:05:16,765][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:05:17,264][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:05:17,762][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:05:18,267][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:05:18,766][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:05:19,264][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:05:19,767][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:05:20,294][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:05:20,799][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:05:21,302][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:05:21,801][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:05:22,299][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:05:22,800][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:05:23,300][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:05:23,796][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:05:24,293][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:05:24,790][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:05:25,287][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:05:25,785][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:05:26,282][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:05:26,779][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:05:27,276][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:05:27,773][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:05:28,269][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:05:28,764][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:05:29,261][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:05:29,757][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:05:30,254][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:05:30,750][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:05:31,247][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:05:31,744][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:05:32,243][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:05:32,740][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:05:33,238][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:05:33,735][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:05:34,236][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:05:34,734][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:05:35,231][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:05:35,728][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:05:36,227][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:05:36,727][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:05:37,226][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:05:37,725][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:05:38,221][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:05:38,719][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:05:39,219][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9995 tokens. +[2026-03-25 23:05:39,885][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 61.96%, ΔTime: 00:00:32 +[2026-03-25 23:05:40,644][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:05:40,646][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:05:40,648][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:05:41,305][__main__][INFO] - Iteration 75 took 50s (30.13% Gen, 68.58% Train). Generation: 15s, Training: 34s. Estimated remaining time: 41h 12m 25s. Estimated total time: 42h 17m 18s. Time estimates for 10 more iterations: 8m 27s, 100 more iterations: 1h 24m 34s, 500 more iterations: 7h 2m 53s. +[2026-03-25 23:05:41,307][__main__][INFO] - Starting iteration 75. +[2026-03-25 23:05:41,706][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 7 and human policies 1. +[2026-03-25 23:05:41,707][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:05:56,468][__main__][INFO] - Number of regex retries in iteration 75: 0 +[2026-03-25 23:05:56,469][__main__][INFO] - agents played in iteration 75 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:05:57,254][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:05:57,277][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:05:57,301][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:05:57,324][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:05:57,325][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:05:57,325][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:05:57,961][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:05:58,427][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:05:58,931][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:05:59,433][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:05:59,936][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:06:00,438][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:06:00,940][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:06:01,439][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:06:01,941][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:06:02,439][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:06:02,936][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:06:03,436][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:06:03,933][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:06:04,435][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:06:04,935][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:06:05,434][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:06:05,933][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:06:06,432][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:06:06,933][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:06:07,433][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:06:07,935][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:06:08,435][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:06:08,939][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:06:09,440][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:06:09,940][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:06:10,462][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:06:10,961][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:06:11,463][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:06:11,962][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:06:12,460][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:06:12,981][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:06:13,479][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:06:13,980][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:06:14,478][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:06:14,978][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:06:15,476][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:06:15,975][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:06:16,472][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:06:16,971][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:06:17,470][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:06:17,968][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:06:18,467][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:06:18,966][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:06:19,463][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:06:19,963][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:06:20,464][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:06:20,965][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:06:21,462][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:06:21,961][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:06:22,458][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:06:22,955][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:06:23,451][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:06:23,947][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:06:24,443][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:06:24,938][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:06:25,434][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:06:25,930][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:06:26,426][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:06:26,922][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:06:27,417][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:06:27,913][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:06:28,408][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:06:28,904][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:06:29,401][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:06:29,898][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9955 tokens. +[2026-03-25 23:06:30,574][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 61.94%, ΔTime: 00:00:32 +[2026-03-25 23:06:31,345][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:06:31,347][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:06:31,349][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:06:31,933][__main__][INFO] - Iteration 76 took 50s (29.39% Gen, 69.44% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 45m 37s. Estimated total time: 41h 51m 20s. Time estimates for 10 more iterations: 8m 22s, 100 more iterations: 1h 23m 42s, 500 more iterations: 6h 58m 33s. +[2026-03-25 23:06:31,940][__main__][INFO] - Starting iteration 76. +[2026-03-25 23:06:32,349][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 7 and human policies 1. +[2026-03-25 23:06:32,350][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:06:37,035][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:06:47,157][__main__][INFO] - Number of regex retries in iteration 76: 1 +[2026-03-25 23:06:47,158][__main__][INFO] - agents played in iteration 76 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:06:47,925][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:06:47,948][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:06:47,971][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:06:47,995][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:06:47,995][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:06:47,996][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:06:48,637][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:06:49,092][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:06:49,593][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:06:50,092][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:06:50,590][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:06:51,088][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:06:51,585][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:06:52,081][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:06:52,580][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:06:53,081][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:06:53,580][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:06:54,078][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:06:54,595][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:06:55,095][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:06:55,596][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:06:56,098][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:06:56,598][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:06:57,100][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:06:57,600][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:06:58,101][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:06:58,600][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:06:59,100][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:06:59,599][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:07:00,098][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:07:00,596][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:07:01,095][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:07:01,595][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:07:02,096][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:07:02,596][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:07:03,094][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:07:03,594][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:07:04,093][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:07:04,591][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:07:05,087][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:07:05,583][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:07:06,079][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:07:06,576][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:07:07,072][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:07:07,568][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:07:08,068][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:07:08,565][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:07:09,062][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:07:09,559][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:07:10,056][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:07:10,557][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:07:11,056][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:07:11,553][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:07:12,050][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:07:12,547][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:07:13,043][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:07:13,544][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:07:14,040][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:07:14,537][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:07:15,036][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:07:15,536][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:07:16,034][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:07:16,531][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:07:17,026][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:07:17,521][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:07:18,019][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:07:18,515][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:07:19,012][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:07:19,507][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:07:20,004][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:07:20,502][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9852 tokens. +[2026-03-25 23:07:21,211][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.01%, Current % of VRAM taken: 60.43%, Block Peak % of device VRAM: 61.83%, ΔTime: 00:00:32 +[2026-03-25 23:07:21,969][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:07:21,971][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:07:21,973][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:07:22,623][__main__][INFO] - Iteration 77 took 50s (29.45% Gen, 69.24% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 47m 24s. Estimated total time: 41h 53m 58s. Time estimates for 10 more iterations: 8m 22s, 100 more iterations: 1h 23m 47s, 500 more iterations: 6h 58m 59s. +[2026-03-25 23:07:22,625][__main__][INFO] - Starting iteration 77. +[2026-03-25 23:07:23,025][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 7 and human policies 1. +[2026-03-25 23:07:23,026][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:08:00,452][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:08:04,532][__main__][INFO] - Number of regex retries in iteration 77: 1 +[2026-03-25 23:08:04,533][__main__][INFO] - agents played in iteration 77 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:08:05,336][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:08:05,359][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:08:05,383][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:08:05,406][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:08:05,406][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:08:05,407][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:08:06,016][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:08:06,470][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:08:06,972][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:08:07,469][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:08:07,967][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:08:08,465][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:08:08,963][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:08:09,459][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:08:09,958][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:08:10,461][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:08:10,959][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:08:11,462][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:08:11,961][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:08:12,459][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:08:12,958][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:08:13,480][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:08:13,983][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:08:14,480][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:08:14,979][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:08:15,477][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:08:15,975][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:08:16,472][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:08:16,969][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:08:17,466][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:08:17,963][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:08:18,459][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:08:18,955][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:08:19,455][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:08:19,951][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:08:20,447][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:08:20,942][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:08:21,439][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:08:21,933][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:08:22,429][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:08:22,928][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:08:23,425][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:08:23,922][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:08:24,421][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:08:24,918][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:08:25,415][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:08:25,914][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:08:26,412][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:08:26,910][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:08:27,406][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:08:27,904][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:08:28,402][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:08:28,899][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:08:29,397][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:08:29,895][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:08:30,397][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:08:30,898][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:08:31,395][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:08:31,897][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:08:32,397][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:08:32,898][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:08:33,398][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:08:33,898][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:08:34,402][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:08:34,904][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:08:35,405][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:08:35,901][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:08:36,400][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:08:36,899][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:08:37,397][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:08:37,894][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9991 tokens. +[2026-03-25 23:08:38,606][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.04%, Current % of VRAM taken: 60.47%, Block Peak % of device VRAM: 62.06%, ΔTime: 00:00:32 +[2026-03-25 23:08:39,360][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:08:39,363][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:08:39,364][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:08:40,014][__main__][INFO] - Iteration 78 took 1m 16s (53.91% Gen, 45.24% Train). Generation: 41s, Training: 34s. Estimated remaining time: 63h 1m 38s. Estimated total time: 64h 9m 29s. Time estimates for 10 more iterations: 12m 49s, 100 more iterations: 2h 8m 18s, 500 more iterations: 10h 41m 34s. +[2026-03-25 23:08:40,017][__main__][INFO] - Starting iteration 78. +[2026-03-25 23:08:40,417][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 7 and human policies 1. +[2026-03-25 23:08:40,418][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:08:42,622][mllm.models.large_language_model_local][WARNING] - Response Propposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:08:55,346][__main__][INFO] - Number of regex retries in iteration 78: 1 +[2026-03-25 23:08:55,347][__main__][INFO] - agents played in iteration 78 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:08:56,159][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:08:56,182][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:08:56,206][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:08:56,228][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:08:56,229][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:08:56,230][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:08:56,859][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:08:57,314][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:08:57,818][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:08:58,323][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:08:58,825][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:08:59,328][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:08:59,827][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:09:00,324][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:09:00,821][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:09:01,322][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:09:01,820][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:09:02,316][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:09:02,812][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:09:03,309][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:09:03,805][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:09:04,303][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:09:04,800][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:09:05,297][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:09:05,798][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:09:06,296][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:09:06,795][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:09:07,293][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:09:07,792][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:09:08,289][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:09:08,786][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:09:09,289][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:09:09,789][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:09:10,289][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:09:10,790][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:09:11,290][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:09:11,788][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:09:12,288][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:09:12,787][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:09:13,284][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:09:13,783][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:09:14,280][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:09:14,777][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:09:15,273][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:09:15,770][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:09:16,266][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:09:16,764][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:09:17,261][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:09:17,758][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:09:18,256][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:09:18,754][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:09:19,252][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:09:19,750][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:09:20,250][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:09:20,747][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:09:21,245][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:09:21,743][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:09:22,240][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:09:22,741][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:09:23,239][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:09:23,737][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:09:24,237][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:09:24,743][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:09:25,250][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:09:25,756][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:09:26,257][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:09:26,757][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:09:27,257][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:09:27,758][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:09:28,257][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:09:28,757][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9902 tokens. +[2026-03-25 23:09:29,517][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 61.81%, ΔTime: 00:00:32 +[2026-03-25 23:09:30,272][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:09:30,274][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:09:30,275][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:09:30,934][__main__][INFO] - Iteration 79 took 50s (29.55% Gen, 69.14% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 57m 10s. Estimated total time: 42h 5m 52s. Time estimates for 10 more iterations: 8m 25s, 100 more iterations: 1h 24m 11s, 500 more iterations: 7h 0m 58s. +[2026-03-25 23:09:30,936][__main__][INFO] - Starting iteration 79. +[2026-03-25 23:09:31,340][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 7 and human policies 1. +[2026-03-25 23:09:31,341][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:09:46,112][__main__][INFO] - Number of regex retries in iteration 79: 0 +[2026-03-25 23:09:46,113][__main__][INFO] - agents played in iteration 79 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:09:46,878][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:09:46,901][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:09:46,924][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:09:46,947][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:09:46,948][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:09:46,949][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:09:47,583][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:09:48,036][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:09:48,540][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:09:49,038][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:09:49,536][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:09:50,035][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:09:50,536][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:09:51,039][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:09:51,539][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:09:52,038][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:09:52,540][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:09:53,042][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:09:53,541][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:09:54,039][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:09:54,539][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:09:55,038][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:09:55,537][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:09:56,036][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:09:56,535][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:09:57,034][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:09:57,533][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:09:58,031][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:09:58,529][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:09:59,029][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:09:59,529][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:10:00,030][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:10:00,531][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:10:01,053][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:10:01,551][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:10:02,054][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:10:02,555][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:10:03,056][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:10:03,553][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:10:04,052][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:10:04,550][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:10:05,051][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:10:05,551][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:10:06,049][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:10:06,548][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:10:07,049][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:10:07,547][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:10:08,045][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:10:08,542][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:10:09,040][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:10:09,538][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:10:10,037][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:10:10,534][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:10:11,031][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:10:11,529][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:10:12,027][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:10:12,525][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:10:13,024][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:10:13,522][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:10:14,020][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:10:14,519][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:10:15,021][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:10:15,527][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:10:16,028][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:10:16,530][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:10:17,029][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:10:17,525][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:10:18,024][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:10:18,522][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:10:19,020][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:10:19,519][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9986 tokens. +[2026-03-25 23:10:20,218][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.05%, Current % of VRAM taken: 60.48%, Block Peak % of device VRAM: 62.06%, ΔTime: 00:00:32 +[2026-03-25 23:10:20,967][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:10:20,969][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:10:20,971][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:10:21,630][__main__][INFO] - Iteration 80 took 50s (29.37% Gen, 69.31% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 45m 4s. Estimated total time: 41h 54m 37s. Time estimates for 10 more iterations: 8m 22s, 100 more iterations: 1h 23m 49s, 500 more iterations: 6h 59m 6s. +[2026-03-25 23:10:21,633][__main__][INFO] - Starting iteration 80. +[2026-03-25 23:10:22,032][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 7 and human policies 1. +[2026-03-25 23:10:22,033][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:10:36,746][__main__][INFO] - Number of regex retries in iteration 80: 0 +[2026-03-25 23:10:36,747][__main__][INFO] - agents played in iteration 80 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:10:37,540][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:10:37,564][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:10:37,587][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:10:37,610][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:10:37,611][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:10:37,611][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:10:38,220][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:10:38,674][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:10:39,182][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:10:39,683][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:10:40,185][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:10:40,684][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:10:41,185][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:10:41,686][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:10:42,186][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:10:42,689][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:10:43,190][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:10:43,690][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:10:44,188][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:10:44,686][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:10:45,183][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:10:45,682][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:10:46,179][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:10:46,677][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:10:47,175][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:10:47,672][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:10:48,170][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:10:48,667][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:10:49,164][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:10:49,661][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:10:50,162][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:10:50,661][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:10:51,161][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:10:51,660][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:10:52,156][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:10:52,655][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:10:53,153][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:10:53,652][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:10:54,151][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:10:54,649][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:10:55,146][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:10:55,644][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:10:56,143][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:10:56,640][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:10:57,138][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:10:57,636][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:10:58,135][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:10:58,635][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:10:59,133][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:10:59,632][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:11:00,133][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:11:00,632][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:11:01,138][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:11:01,640][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:11:02,139][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:11:02,661][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:11:03,164][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:11:03,665][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:11:04,167][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:11:04,669][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:11:05,169][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:11:05,670][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:11:06,172][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:11:06,675][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:11:07,181][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:11:07,687][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:11:08,186][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:11:08,688][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:11:09,190][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:11:09,692][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:11:10,192][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10000 tokens. +[2026-03-25 23:11:10,861][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.01%, ΔTime: 00:00:32 +[2026-03-25 23:11:11,790][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:11:11,793][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:11:11,794][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:11:13,086][__main__][INFO] - Iteration 81 took 51s (28.82% Gen, 68.65% Train). Generation: 14s, Training: 35s. Estimated remaining time: 41h 22m 16s. Estimated total time: 42h 32m 41s. Time estimates for 10 more iterations: 8m 30s, 100 more iterations: 1h 25m 5s, 500 more iterations: 7h 5m 26s. +[2026-03-25 23:11:13,088][__main__][INFO] - Starting iteration 81. +[2026-03-25 23:11:13,490][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 8 and human policies 1. +[2026-03-25 23:11:13,491][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:11:27,222][__main__][INFO] - Number of regex retries in iteration 81: 0 +[2026-03-25 23:11:27,223][__main__][INFO] - agents played in iteration 81 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:11:27,994][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:11:28,017][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:11:28,040][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:11:28,063][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:11:28,064][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:11:28,065][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:11:28,707][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:11:29,159][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:11:29,662][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:11:30,158][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:11:30,658][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:11:31,155][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:11:31,651][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:11:32,153][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:11:32,653][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:11:33,152][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:11:33,650][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:11:34,147][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:11:34,644][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:11:35,143][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:11:35,639][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:11:36,138][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:11:36,636][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:11:37,134][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:11:37,632][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:11:38,130][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:11:38,628][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:11:39,128][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:11:39,625][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:11:40,122][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:11:40,619][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:11:41,117][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:11:41,614][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:11:42,112][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:11:42,608][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:11:43,104][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:11:43,601][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:11:44,098][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:11:44,595][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:11:45,093][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:11:45,591][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:11:46,090][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:11:46,590][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:11:47,108][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:11:47,607][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:11:48,109][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:11:48,610][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:11:49,106][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:11:49,604][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:11:50,102][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:11:50,601][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:11:51,098][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:11:51,596][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:11:52,094][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:11:52,591][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:11:53,089][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:11:53,587][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:11:54,087][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:11:54,585][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:11:55,084][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:11:55,580][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:11:56,078][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:11:56,577][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:11:57,077][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:11:57,582][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:11:58,087][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:11:58,591][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:11:59,090][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:11:59,590][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:12:00,091][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:12:00,594][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9968 tokens. +[2026-03-25 23:12:01,275][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 61.83%, ΔTime: 00:00:32 +[2026-03-25 23:12:02,032][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:12:02,034][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:12:02,036][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:12:02,723][__main__][INFO] - Iteration 82 took 49s (27.89% Gen, 70.71% Train). Generation: 13s, Training: 34s. Estimated remaining time: 39h 50m 28s. Estimated total time: 41h 1m 42s. Time estimates for 10 more iterations: 8m 12s, 100 more iterations: 1h 22m 3s, 500 more iterations: 6h 50m 17s. +[2026-03-25 23:12:02,726][__main__][INFO] - Starting iteration 82. +[2026-03-25 23:12:03,125][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 8 and human policies 1. +[2026-03-25 23:12:03,125][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:12:16,961][__main__][INFO] - Number of regex retries in iteration 82: 0 +[2026-03-25 23:12:16,962][__main__][INFO] - agents played in iteration 82 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:12:17,732][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:12:17,756][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:12:17,779][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:12:17,802][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:12:17,803][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:12:17,803][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:12:18,434][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:12:18,887][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:12:19,389][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:12:19,886][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:12:20,384][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:12:20,881][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:12:21,380][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:12:21,877][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:12:22,372][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:12:22,870][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:12:23,366][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:12:23,865][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:12:24,362][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:12:24,859][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:12:25,357][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:12:25,855][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:12:26,353][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:12:26,853][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:12:27,351][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:12:27,869][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:12:28,366][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:12:28,865][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:12:29,365][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:12:29,865][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:12:30,363][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:12:30,861][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:12:31,358][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:12:31,854][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:12:32,355][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:12:32,852][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:12:33,348][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:12:33,844][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:12:34,340][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:12:34,835][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:12:35,331][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:12:35,828][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:12:36,325][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:12:36,821][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:12:37,317][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:12:37,814][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:12:38,312][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:12:38,813][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:12:39,314][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:12:39,814][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:12:40,314][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:12:40,817][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:12:41,319][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:12:41,819][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:12:42,319][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:12:42,819][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:12:43,316][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:12:43,814][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:12:44,315][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:12:44,812][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:12:45,313][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:12:45,811][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:12:46,310][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:12:46,808][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:12:47,307][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:12:47,805][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:12:48,306][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:12:48,811][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:12:49,317][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:12:49,822][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:12:50,323][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10044 tokens. +[2026-03-25 23:12:51,011][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.05%, Current % of VRAM taken: 60.47%, Block Peak % of device VRAM: 62.05%, ΔTime: 00:00:32 +[2026-03-25 23:12:51,753][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:12:51,755][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:12:51,757][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:12:52,463][__main__][INFO] - Iteration 83 took 49s (28.04% Gen, 70.52% Train). Generation: 13s, Training: 34s. Estimated remaining time: 39h 54m 53s. Estimated total time: 41h 6m 57s. Time estimates for 10 more iterations: 8m 13s, 100 more iterations: 1h 22m 13s, 500 more iterations: 6h 51m 9s. +[2026-03-25 23:12:52,467][__main__][INFO] - Starting iteration 83. +[2026-03-25 23:12:52,906][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 8 and human policies 1. +[2026-03-25 23:12:52,906][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:12:55,014][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:12:56,049][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:13:07,404][__main__][INFO] - Number of regex retries in iteration 83: 2 +[2026-03-25 23:13:07,405][__main__][INFO] - agents played in iteration 83 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:13:08,175][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:13:08,199][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:13:08,222][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:13:08,246][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:13:08,247][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:13:08,247][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:13:08,884][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:13:09,342][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:13:09,847][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:13:10,351][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:13:10,850][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:13:11,349][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:13:11,850][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:13:12,348][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:13:12,845][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:13:13,344][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:13:13,844][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:13:14,343][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:13:14,844][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:13:15,343][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:13:15,843][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:13:16,342][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:13:16,841][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:13:17,340][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:13:17,843][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:13:18,344][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:13:18,844][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:13:19,344][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:13:19,842][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:13:20,342][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:13:20,840][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:13:21,339][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:13:21,837][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:13:22,335][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:13:22,834][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:13:23,332][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:13:23,829][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:13:24,327][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:13:24,825][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:13:25,321][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:13:25,817][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:13:26,312][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:13:26,807][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:13:27,302][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:13:27,799][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:13:28,296][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:13:28,793][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:13:29,292][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:13:29,788][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:13:30,285][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:13:30,783][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:13:31,280][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:13:31,777][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:13:32,277][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:13:32,775][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:13:33,274][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:13:33,773][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:13:34,269][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:13:34,767][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:13:35,264][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:13:35,762][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:13:36,260][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:13:36,757][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:13:37,254][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:13:37,751][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:13:38,249][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:13:38,746][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:13:39,244][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:13:39,748][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:13:40,255][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:13:40,758][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9923 tokens. +[2026-03-25 23:13:41,435][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 61.89%, ΔTime: 00:00:32 +[2026-03-25 23:13:42,294][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:13:42,296][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:13:42,298][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:13:43,053][__main__][INFO] - Iteration 84 took 50s (28.91% Gen, 69.58% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 34m 29s. Estimated total time: 41h 47m 24s. Time estimates for 10 more iterations: 8m 21s, 100 more iterations: 1h 23m 34s, 500 more iterations: 6h 57m 54s. +[2026-03-25 23:13:43,056][__main__][INFO] - Starting iteration 84. +[2026-03-25 23:13:43,499][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 8 and human policies 1. +[2026-03-25 23:13:43,500][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:13:46,009][mllm.models.large_language_model_local][WARNING] - Response Proposeral: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:13:46,848][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:13:58,340][__main__][INFO] - Number of regex retries in iteration 84: 2 +[2026-03-25 23:13:58,341][__main__][INFO] - agents played in iteration 84 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:13:59,112][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:13:59,136][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:13:59,160][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:13:59,184][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:13:59,184][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:13:59,185][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:13:59,829][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:14:00,284][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:14:00,791][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:14:01,291][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:14:01,789][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:14:02,288][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:14:02,785][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:14:03,285][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:14:03,784][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:14:04,286][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:14:04,786][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:14:05,283][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:14:05,782][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:14:06,280][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:14:06,776][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:14:07,274][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:14:07,772][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:14:08,271][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:14:08,779][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:14:09,279][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:14:09,778][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:14:10,278][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:14:10,777][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:14:11,276][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:14:11,774][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:14:12,275][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:14:12,775][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:14:13,275][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:14:13,775][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:14:14,274][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:14:14,776][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:14:15,276][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:14:15,776][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:14:16,276][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:14:16,775][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:14:17,274][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:14:17,774][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:14:18,272][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:14:18,770][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:14:19,265][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:14:19,761][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:14:20,260][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:14:20,759][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:14:21,256][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:14:21,754][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:14:22,251][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:14:22,748][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:14:23,245][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:14:23,742][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:14:24,238][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:14:24,735][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:14:25,232][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:14:25,728][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:14:26,225][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:14:26,724][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:14:27,220][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:14:27,718][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:14:28,216][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:14:28,713][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:14:29,211][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:14:29,708][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:14:30,205][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:14:30,702][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:14:31,206][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:14:31,710][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9950 tokens. +[2026-03-25 23:14:32,412][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.48%, Block Peak % of device VRAM: 62.10%, ΔTime: 00:00:32 +[2026-03-25 23:14:33,158][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:14:33,160][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:14:33,162][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:14:33,864][__main__][INFO] - Iteration 85 took 50s (29.47% Gen, 69.13% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 44m 32s. Estimated total time: 41h 58m 17s. Time estimates for 10 more iterations: 8m 23s, 100 more iterations: 1h 23m 56s, 500 more iterations: 6h 59m 42s. +[2026-03-25 23:14:33,866][__main__][INFO] - Starting iteration 85. +[2026-03-25 23:14:34,297][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 8 and human policies 1. +[2026-03-25 23:14:34,298][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:14:48,709][__main__][INFO] - Number of regex retries in iteration 85: 0 +[2026-03-25 23:14:48,710][__main__][INFO] - agents played in iteration 85 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:14:49,481][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:14:49,504][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:14:49,527][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:14:49,550][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:14:49,551][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:14:49,551][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:14:50,186][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:14:50,640][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:14:51,145][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:14:51,643][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:14:52,141][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:14:52,638][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:14:53,139][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:14:53,637][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:14:54,134][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:14:54,635][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:14:55,135][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:14:55,634][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:14:56,132][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:14:56,634][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:14:57,132][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:14:57,633][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:14:58,131][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:14:58,628][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:14:59,123][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:14:59,622][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:15:00,119][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:15:00,617][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:15:01,115][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:15:01,611][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:15:02,108][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:15:02,605][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:15:03,103][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:15:03,600][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:15:04,096][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:15:04,594][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:15:05,091][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:15:05,586][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:15:06,084][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:15:06,585][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:15:07,110][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:15:07,613][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:15:08,111][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:15:08,612][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:15:09,114][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:15:09,614][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:15:10,115][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:15:10,611][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:15:11,108][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:15:11,605][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:15:12,102][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:15:12,600][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:15:13,098][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:15:13,596][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:15:14,094][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:15:14,594][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:15:15,093][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:15:15,594][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:15:16,094][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:15:16,594][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:15:17,094][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:15:17,593][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:15:18,092][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:15:18,590][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:15:19,086][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:15:19,581][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:15:20,082][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:15:20,586][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:15:21,092][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:15:21,596][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:15:22,104][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9951 tokens. +[2026-03-25 23:15:22,782][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.08%, ΔTime: 00:00:32 +[2026-03-25 23:15:23,548][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:15:23,550][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:15:23,552][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:15:24,211][__main__][INFO] - Iteration 86 took 49s (28.87% Gen, 69.80% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 21m 6s. Estimated total time: 41h 35m 42s. Time estimates for 10 more iterations: 8m 19s, 100 more iterations: 1h 23m 11s, 500 more iterations: 6h 55m 57s. +[2026-03-25 23:15:24,213][__main__][INFO] - Starting iteration 86. +[2026-03-25 23:15:24,612][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 8 and human policies 1. +[2026-03-25 23:15:24,612][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:15:34,610][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:15:38,680][__main__][INFO] - Number of regex retries in iteration 86: 1 +[2026-03-25 23:15:38,681][__main__][INFO] - agents played in iteration 86 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:15:39,449][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:15:39,472][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:15:39,495][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:15:39,518][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:15:39,519][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:15:39,519][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:15:40,145][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:15:40,599][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:15:41,104][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:15:41,602][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:15:42,100][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:15:42,598][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:15:43,092][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:15:43,588][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:15:44,084][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:15:44,582][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:15:45,080][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:15:45,578][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:15:46,075][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:15:46,571][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:15:47,069][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:15:47,565][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:15:48,062][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:15:48,561][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:15:49,062][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:15:49,561][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:15:50,059][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:15:50,557][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:15:51,054][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:15:51,552][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:15:52,050][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:15:52,548][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:15:53,045][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:15:53,543][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:15:54,040][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:15:54,538][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:15:55,037][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:15:55,539][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:15:56,037][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:15:56,537][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:15:57,036][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:15:57,535][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:15:58,033][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:15:58,532][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:15:59,030][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:15:59,528][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:16:00,025][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:16:00,552][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:16:01,052][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:16:01,552][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:16:02,053][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:16:02,552][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:16:03,056][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:16:03,555][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:16:04,056][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:16:04,552][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:16:05,049][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:16:05,547][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:16:06,045][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:16:06,541][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:16:07,037][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:16:07,535][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:16:08,033][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:16:08,530][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:16:09,027][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:16:09,524][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:16:10,020][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:16:10,520][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:16:11,018][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:16:11,517][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:16:12,022][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9966 tokens. +[2026-03-25 23:16:12,704][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 61.99%, ΔTime: 00:00:32 +[2026-03-25 23:16:13,462][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:16:13,464][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:16:13,466][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:16:14,144][__main__][INFO] - Iteration 87 took 49s (28.40% Gen, 70.22% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 1m 14s. Estimated total time: 41h 16m 39s. Time estimates for 10 more iterations: 8m 15s, 100 more iterations: 1h 22m 33s, 500 more iterations: 6h 52m 46s. +[2026-03-25 23:16:14,147][__main__][INFO] - Starting iteration 87. +[2026-03-25 23:16:14,545][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 8 and human policies 1. +[2026-03-25 23:16:14,546][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:16:29,904][__main__][INFO] - Number of regex retries in iteration 87: 0 +[2026-03-25 23:16:29,905][__main__][INFO] - agents played in iteration 87 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:16:30,675][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:16:30,699][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:16:30,722][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:16:30,745][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:16:30,746][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:16:30,746][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:16:31,366][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:16:31,823][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:16:32,329][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:16:32,830][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:16:33,330][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:16:33,830][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:16:34,329][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:16:34,829][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:16:35,328][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:16:35,826][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:16:36,325][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:16:36,823][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:16:37,322][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:16:37,820][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:16:38,318][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:16:38,817][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:16:39,318][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:16:39,815][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:16:40,314][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:16:40,812][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:16:41,309][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:16:41,806][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:16:42,307][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:16:42,803][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:16:43,301][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:16:43,808][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:16:44,311][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:16:44,814][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:16:45,316][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:16:45,819][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:16:46,321][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:16:46,825][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:16:47,327][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:16:47,826][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:16:48,324][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:16:48,824][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:16:49,322][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:16:49,822][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:16:50,323][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:16:50,822][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:16:51,320][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:16:51,818][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:16:52,318][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:16:52,818][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:16:53,320][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:16:53,820][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:16:54,321][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:16:54,820][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:16:55,319][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:16:55,819][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:16:56,321][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:16:56,820][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:16:57,319][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:16:57,819][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:16:58,319][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:16:58,817][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:16:59,317][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:16:59,818][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:17:00,318][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:17:00,818][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:17:01,321][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:17:01,820][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:17:02,322][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:17:02,834][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:17:03,347][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10096 tokens. +[2026-03-25 23:17:04,036][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.11%, ΔTime: 00:00:32 +[2026-03-25 23:17:04,793][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:17:04,796][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:17:04,797][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:17:05,419][__main__][INFO] - Iteration 88 took 50s (30.19% Gen, 68.58% Train). Generation: 15s, Training: 34s. Estimated remaining time: 41h 7m 24s. Estimated total time: 42h 23m 41s. Time estimates for 10 more iterations: 8m 28s, 100 more iterations: 1h 24m 47s, 500 more iterations: 7h 3m 56s. +[2026-03-25 23:17:05,421][__main__][INFO] - Starting iteration 88. +[2026-03-25 23:17:05,831][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 8 and human policies 1. +[2026-03-25 23:17:05,832][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:17:10,377][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:17:16,461][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls + +Given the values: +- I value hats at 10, while Bob values them at 1. +- I value books at 10, while Bob values them at 10. +- I value balls at 1, while Bob values them at 10. + +Since I have the highest value for hats, it is beneficial to claim all of them. Books have a similar value for both of us, but since my value is lower than Bob's, it's best to avoid sharing those to maximize my points. Balls, with the lowest value for both, can be entirely avoided without much cost to my overall score. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:17:21,302][__main__][INFO] - Number of regex retries in iteration 88: 2 +[2026-03-25 23:17:21,303][__main__][INFO] - agents played in iteration 88 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:17:22,087][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:17:22,110][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:17:22,134][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:17:22,157][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:17:22,157][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:17:22,158][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:17:22,775][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:17:23,227][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:17:23,733][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:17:24,232][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:17:24,731][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:17:25,232][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:17:25,731][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:17:26,229][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:17:26,727][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:17:27,226][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:17:27,724][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:17:28,221][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:17:28,720][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:17:29,218][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:17:29,715][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:17:30,215][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:17:30,713][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:17:31,216][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:17:31,716][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:17:32,214][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:17:32,720][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:17:33,221][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:17:33,745][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:17:34,246][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:17:34,746][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:17:35,244][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:17:35,742][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:17:36,240][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:17:36,737][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:17:37,235][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:17:37,732][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:17:38,230][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:17:38,727][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:17:39,226][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:17:39,722][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:17:40,223][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:17:40,721][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:17:41,220][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:17:41,718][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:17:42,217][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:17:42,714][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:17:43,217][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:17:43,717][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:17:44,215][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:17:44,716][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:17:45,215][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:17:45,716][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:17:46,214][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:17:46,712][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:17:47,210][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:17:47,709][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:17:48,209][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:17:48,708][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:17:49,206][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:17:49,707][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:17:50,206][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:17:50,705][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:17:51,203][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:17:51,702][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:17:52,201][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:17:52,699][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:17:53,197][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:17:53,696][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:17:54,199][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:17:54,710][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10137 tokens. +[2026-03-25 23:17:55,395][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.61%, Block Peak % of device VRAM: 62.22%, ΔTime: 00:00:32 +[2026-03-25 23:17:56,147][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:17:56,149][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:17:56,151][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:17:56,819][__main__][INFO] - Iteration 89 took 50s (30.34% Gen, 68.34% Train). Generation: 15s, Training: 34s. Estimated remaining time: 41h 12m 24s. Estimated total time: 42h 29m 32s. Time estimates for 10 more iterations: 8m 29s, 100 more iterations: 1h 24m 59s, 500 more iterations: 7h 4m 55s. +[2026-03-25 23:17:56,822][__main__][INFO] - Starting iteration 89. +[2026-03-25 23:17:57,222][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 8 and human policies 1. +[2026-03-25 23:17:57,223][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:17:59,629][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:18:12,047][__main__][INFO] - Number of regex retries in iteration 89: 1 +[2026-03-25 23:18:12,048][__main__][INFO] - agents played in iteration 89 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:18:12,856][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:18:12,879][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:18:12,903][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:18:12,926][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:18:12,926][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:18:12,927][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:18:13,554][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:18:14,008][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:18:14,515][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:18:15,018][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:18:15,516][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:18:16,017][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:18:16,515][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:18:17,012][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:18:17,510][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:18:18,008][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:18:18,504][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:18:19,000][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:18:19,497][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:18:19,996][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:18:20,495][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:18:20,992][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:18:21,489][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:18:21,987][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:18:22,485][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:18:22,983][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:18:23,481][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:18:23,977][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:18:24,474][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:18:24,972][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:18:25,469][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:18:25,968][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:18:26,466][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:18:26,964][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:18:27,465][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:18:27,963][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:18:28,463][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:18:28,962][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:18:29,461][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:18:29,959][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:18:30,456][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:18:30,956][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:18:31,456][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:18:31,955][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:18:32,452][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:18:32,948][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:18:33,446][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:18:33,945][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:18:34,444][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:18:34,951][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:18:35,449][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:18:35,947][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:18:36,445][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:18:36,942][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:18:37,443][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:18:37,944][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:18:38,443][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:18:38,941][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:18:39,438][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:18:39,939][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:18:40,438][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:18:40,936][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:18:41,436][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:18:41,934][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:18:42,431][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:18:42,927][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:18:43,425][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:18:43,921][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:18:44,424][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:18:44,927][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:18:45,438][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10111 tokens. +[2026-03-25 23:18:46,122][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.04%, Current % of VRAM taken: 60.46%, Block Peak % of device VRAM: 62.03%, ΔTime: 00:00:32 +[2026-03-25 23:18:46,878][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:18:46,881][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:18:46,882][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:18:47,548][__main__][INFO] - Iteration 90 took 50s (29.46% Gen, 69.22% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 38m 21s. Estimated total time: 41h 56m 20s. Time estimates for 10 more iterations: 8m 23s, 100 more iterations: 1h 23m 52s, 500 more iterations: 6h 59m 23s. +[2026-03-25 23:18:47,550][__main__][INFO] - Starting iteration 90. +[2026-03-25 23:18:47,955][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 8 and human policies 1. +[2026-03-25 23:18:47,956][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:18:56,699][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:19:02,077][__main__][INFO] - Number of regex retries in iteration 90: 1 +[2026-03-25 23:19:02,078][__main__][INFO] - agents played in iteration 90 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:19:02,870][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:19:02,894][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:19:02,917][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:19:02,941][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:19:02,941][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:19:02,942][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:19:03,562][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:19:04,016][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:19:04,518][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:19:05,016][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:19:05,515][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:19:06,014][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:19:06,515][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:19:07,013][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:19:07,512][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:19:08,010][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:19:08,507][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:19:09,005][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:19:09,503][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:19:10,001][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:19:10,499][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:19:10,998][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:19:11,498][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:19:11,996][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:19:12,494][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:19:12,990][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:19:13,489][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:19:13,986][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:19:14,483][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:19:14,980][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:19:15,477][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:19:15,976][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:19:16,475][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:19:16,972][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:19:17,469][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:19:17,969][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:19:18,468][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:19:18,965][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:19:19,462][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:19:19,960][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:19:20,460][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:19:20,957][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:19:21,455][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:19:21,952][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:19:22,449][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:19:22,946][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:19:23,444][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:19:23,940][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:19:24,437][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:19:24,935][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:19:25,433][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:19:25,929][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:19:26,426][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:19:26,922][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:19:27,422][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:19:27,926][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:19:28,423][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:19:28,922][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:19:29,425][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:19:29,926][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:19:30,425][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:19:30,925][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:19:31,425][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:19:31,923][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:19:32,421][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:19:32,919][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:19:33,416][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:19:33,915][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:19:34,413][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:19:34,912][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:19:35,418][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10112 tokens. +[2026-03-25 23:19:36,107][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.02%, ΔTime: 00:00:32 +[2026-03-25 23:19:36,873][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:19:36,875][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:19:36,877][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:19:38,144][__main__][INFO] - Iteration 91 took 50s (28.14% Gen, 69.34% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 30m 37s. Estimated total time: 41h 49m 26s. Time estimates for 10 more iterations: 8m 21s, 100 more iterations: 1h 23m 38s, 500 more iterations: 6h 58m 14s. +[2026-03-25 23:19:38,146][__main__][INFO] - Starting iteration 91. +[2026-03-25 23:19:38,551][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 9 and human policies 1. +[2026-03-25 23:19:38,551][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:19:48,612][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:19:53,621][__main__][INFO] - Number of regex retries in iteration 91: 1 +[2026-03-25 23:19:53,622][__main__][INFO] - agents played in iteration 91 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:19:54,432][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:19:54,455][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:19:54,478][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:19:54,501][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:19:54,502][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:19:54,502][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:19:55,116][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:19:55,570][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:19:56,076][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:19:56,575][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:19:57,074][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:19:57,572][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:19:58,071][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:19:58,569][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:19:59,066][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:19:59,565][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:20:00,063][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:20:00,561][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:20:01,060][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:20:01,557][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:20:02,054][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:20:02,551][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:20:03,050][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:20:03,554][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:20:04,073][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:20:04,577][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:20:05,076][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:20:05,581][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:20:06,080][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:20:06,581][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:20:07,082][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:20:07,579][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:20:08,077][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:20:08,577][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:20:09,075][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:20:09,571][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:20:10,069][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:20:10,566][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:20:11,067][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:20:11,566][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:20:12,067][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:20:12,567][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:20:13,065][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:20:13,563][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:20:14,063][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:20:14,561][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:20:15,060][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:20:15,558][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:20:16,057][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:20:16,555][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:20:17,052][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:20:17,550][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:20:18,048][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:20:18,545][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:20:19,043][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:20:19,539][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:20:20,036][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:20:20,534][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:20:21,031][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:20:21,528][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:20:22,026][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:20:22,523][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:20:23,019][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:20:23,519][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:20:24,018][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:20:24,518][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:20:25,017][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:20:25,520][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:20:26,023][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:20:26,526][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:20:27,033][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10050 tokens. +[2026-03-25 23:20:27,711][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.15%, Current % of VRAM taken: 60.58%, Block Peak % of device VRAM: 62.02%, ΔTime: 00:00:32 +[2026-03-25 23:20:28,454][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:20:28,456][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:20:28,458][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:20:29,118][__main__][INFO] - Iteration 92 took 50s (29.80% Gen, 68.89% Train). Generation: 15s, Training: 34s. Estimated remaining time: 40h 48m 43s. Estimated total time: 42h 8m 23s. Time estimates for 10 more iterations: 8m 25s, 100 more iterations: 1h 24m 16s, 500 more iterations: 7h 1m 23s. +[2026-03-25 23:20:29,120][__main__][INFO] - Starting iteration 92. +[2026-03-25 23:20:29,522][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 9 and human policies 1. +[2026-03-25 23:20:29,523][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:20:44,909][__main__][INFO] - Number of regex retries in iteration 92: 0 +[2026-03-25 23:20:44,910][__main__][INFO] - agents played in iteration 92 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:20:45,696][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:20:45,719][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:20:45,743][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:20:45,765][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:20:45,766][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:20:45,766][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:20:46,395][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:20:46,848][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:20:47,350][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:20:47,850][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:20:48,351][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:20:48,848][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:20:49,345][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:20:49,843][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:20:50,342][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:20:50,840][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:20:51,338][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:20:51,835][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:20:52,333][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:20:52,830][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:20:53,327][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:20:53,824][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:20:54,322][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:20:54,819][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:20:55,317][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:20:55,815][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:20:56,314][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:20:56,812][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:20:57,310][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:20:57,807][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:20:58,303][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:20:58,801][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:20:59,301][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:20:59,804][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:21:00,301][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:21:00,799][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:21:01,295][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:21:01,798][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:21:02,296][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:21:02,796][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:21:03,293][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:21:03,790][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:21:04,288][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:21:04,784][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:21:05,280][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:21:05,779][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:21:06,276][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:21:06,775][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:21:07,276][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:21:07,775][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:21:08,273][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:21:08,774][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:21:09,275][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:21:09,773][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:21:10,271][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:21:10,767][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:21:11,263][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:21:11,763][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:21:13,596][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:21:14,093][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:21:14,591][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:21:15,088][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:21:15,585][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:21:16,083][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:21:16,588][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:21:17,090][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:21:17,600][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:21:18,100][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:21:18,607][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:21:19,110][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:21:19,615][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10119 tokens. +[2026-03-25 23:21:20,313][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.07%, ΔTime: 00:00:33 +[2026-03-25 23:21:21,060][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:21:21,062][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:21:21,064][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:21:21,739][__main__][INFO] - Iteration 93 took 52s (29.47% Gen, 69.24% Train). Generation: 15s, Training: 36s. Estimated remaining time: 42h 10m 17s. Estimated total time: 43h 30m 50s. Time estimates for 10 more iterations: 8m 42s, 100 more iterations: 1h 27m 1s, 500 more iterations: 7h 15m 8s. +[2026-03-25 23:21:21,741][__main__][INFO] - Starting iteration 93. +[2026-03-25 23:21:22,143][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 9 and human policies 1. +[2026-03-25 23:21:22,144][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:21:35,730][__main__][INFO] - Number of regex retries in iteration 93: 0 +[2026-03-25 23:21:35,731][__main__][INFO] - agents played in iteration 93 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:21:36,504][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:21:36,528][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:21:36,551][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:21:36,574][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:21:36,574][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:21:36,575][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:21:37,212][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:21:37,666][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:21:38,168][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:21:38,664][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:21:39,162][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:21:39,664][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:21:40,162][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:21:40,659][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:21:41,157][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:21:41,655][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:21:42,153][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:21:42,651][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:21:43,146][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:21:43,645][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:21:44,143][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:21:44,641][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:21:45,139][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:21:45,636][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:21:46,134][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:21:46,631][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:21:47,128][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:21:47,625][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:21:48,121][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:21:48,618][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:21:49,114][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:21:49,610][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:21:50,109][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:21:50,606][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:21:51,104][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:21:51,601][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:21:52,099][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:21:52,596][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:21:53,093][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:21:53,597][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:21:54,094][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:21:54,593][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:21:55,092][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:21:55,589][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:21:56,088][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:21:56,586][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:21:57,084][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:21:57,580][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:21:58,077][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:21:58,573][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:21:59,068][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:21:59,564][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:22:00,060][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:22:00,555][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:22:01,052][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:22:01,549][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:22:02,046][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:22:02,544][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:22:03,041][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:22:03,539][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:22:04,037][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:22:04,536][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:22:05,035][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:22:05,531][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:22:06,027][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:22:06,523][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:22:07,019][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:22:07,519][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:22:08,020][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:22:08,522][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:22:09,021][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9989 tokens. +[2026-03-25 23:22:09,692][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.00%, Current % of VRAM taken: 60.43%, Block Peak % of device VRAM: 62.03%, ΔTime: 00:00:32 +[2026-03-25 23:22:10,439][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:22:10,442][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:22:10,444][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:22:11,106][__main__][INFO] - Iteration 94 took 48s (27.75% Gen, 70.90% Train). Generation: 13s, Training: 34s. Estimated remaining time: 39h 26m 48s. Estimated total time: 40h 48m 10s. Time estimates for 10 more iterations: 8m 9s, 100 more iterations: 1h 21m 36s, 500 more iterations: 6h 48m 1s. +[2026-03-25 23:22:11,109][__main__][INFO] - Starting iteration 94. +[2026-03-25 23:22:11,514][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 9 and human policies 1. +[2026-03-25 23:22:11,514][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:22:26,928][__main__][INFO] - Number of regex retries in iteration 94: 0 +[2026-03-25 23:22:26,929][__main__][INFO] - agents played in iteration 94 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:22:27,706][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:22:27,730][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:22:27,753][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:22:27,776][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:22:27,777][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:22:27,777][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:22:28,395][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:22:28,851][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:22:29,357][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:22:29,859][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:22:30,360][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:22:30,860][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:22:31,362][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:22:31,862][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:22:32,360][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:22:32,858][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:22:33,356][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:22:33,853][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:22:34,353][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:22:34,854][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:22:35,353][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:22:35,852][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:22:36,348][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:22:36,848][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:22:37,349][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:22:37,847][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:22:38,350][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:22:38,850][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:22:39,351][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:22:39,850][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:22:40,352][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:22:40,850][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:22:41,348][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:22:41,846][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:22:42,343][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:22:42,842][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:22:43,339][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:22:43,840][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:22:44,339][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:22:44,840][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:22:45,340][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:22:45,838][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:22:46,338][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:22:46,837][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:22:47,336][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:22:47,835][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:22:48,331][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:22:48,832][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:22:49,328][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:22:49,826][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:22:50,328][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:22:50,828][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:22:51,327][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:22:51,826][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:22:52,323][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:22:52,820][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:22:53,316][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:22:53,816][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:22:54,311][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:22:54,811][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:22:55,308][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:22:55,805][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:22:56,302][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:22:56,798][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:22:57,300][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:22:57,806][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:22:58,309][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:22:58,806][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:22:59,303][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:22:59,801][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:23:00,298][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10085 tokens. +[2026-03-25 23:23:00,978][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.99%, Current % of VRAM taken: 60.42%, Block Peak % of device VRAM: 62.08%, ΔTime: 00:00:32 +[2026-03-25 23:23:01,730][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:23:01,733][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:23:01,735][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:23:02,400][__main__][INFO] - Iteration 95 took 50s (30.29% Gen, 68.40% Train). Generation: 15s, Training: 34s. Estimated remaining time: 41h 2m 8s. Estimated total time: 42h 24m 22s. Time estimates for 10 more iterations: 8m 28s, 100 more iterations: 1h 24m 48s, 500 more iterations: 7h 4m 3s. +[2026-03-25 23:23:02,403][__main__][INFO] - Starting iteration 95. +[2026-03-25 23:23:02,805][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 9 and human policies 1. +[2026-03-25 23:23:02,806][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:23:18,565][__main__][INFO] - Number of regex retries in iteration 95: 0 +[2026-03-25 23:23:18,566][__main__][INFO] - agents played in iteration 95 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:23:19,340][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:23:19,364][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:23:19,387][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:23:19,410][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:23:19,411][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:23:19,411][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:23:20,040][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:23:20,495][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:23:21,001][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:23:21,500][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:23:22,003][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:23:22,501][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:23:23,000][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:23:23,498][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:23:23,997][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:23:24,496][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:23:24,995][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:23:25,493][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:23:25,990][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:23:26,487][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:23:26,985][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:23:27,482][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:23:27,981][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:23:28,479][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:23:28,976][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:23:29,474][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:23:29,971][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:23:30,468][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:23:30,964][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:23:31,461][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:23:31,958][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:23:32,456][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:23:32,956][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:23:33,455][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:23:33,952][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:23:34,450][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:23:34,949][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:23:35,446][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:23:35,943][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:23:36,440][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:23:36,938][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:23:37,435][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:23:37,931][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:23:38,428][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:23:38,924][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:23:39,419][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:23:39,914][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:23:40,411][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:23:40,909][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:23:41,405][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:23:41,903][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:23:42,402][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:23:42,900][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:23:43,398][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:23:43,894][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:23:44,391][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:23:44,889][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:23:45,386][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:23:45,884][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:23:46,382][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:23:46,888][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:23:47,392][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:23:47,896][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:23:48,395][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:23:48,892][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:23:49,389][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:23:49,886][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:23:50,385][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:23:50,881][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:23:51,378][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:23:51,881][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10012 tokens. +[2026-03-25 23:23:52,540][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.01%, Current % of VRAM taken: 60.44%, Block Peak % of device VRAM: 61.86%, ΔTime: 00:00:32 +[2026-03-25 23:23:53,279][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:23:53,281][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:23:53,283][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:23:53,945][__main__][INFO] - Iteration 96 took 51s (30.82% Gen, 67.88% Train). Generation: 15s, Training: 34s. Estimated remaining time: 41h 13m 59s. Estimated total time: 42h 37m 4s. Time estimates for 10 more iterations: 8m 31s, 100 more iterations: 1h 25m 14s, 500 more iterations: 7h 6m 10s. +[2026-03-25 23:23:53,947][__main__][INFO] - Starting iteration 96. +[2026-03-25 23:23:54,348][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 9 and human policies 1. +[2026-03-25 23:23:54,349][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:23:56,801][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:24:09,281][__main__][INFO] - Number of regex retries in iteration 96: 1 +[2026-03-25 23:24:09,282][__main__][INFO] - agents played in iteration 96 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:24:10,067][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:24:10,091][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:24:10,114][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:24:10,138][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:24:10,138][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:24:10,139][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:24:10,759][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:24:11,217][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:24:11,723][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:24:12,225][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:24:12,724][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:24:13,224][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:24:13,727][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:24:14,226][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:24:14,726][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:24:15,226][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:24:15,724][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:24:16,222][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:24:16,721][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:24:17,220][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:24:17,718][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:24:18,216][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:24:18,714][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:24:19,214][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:24:19,713][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:24:20,215][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:24:20,716][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:24:21,214][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:24:21,715][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:24:22,212][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:24:22,709][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:24:23,204][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:24:23,700][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:24:24,196][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:24:24,693][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:24:25,189][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:24:25,685][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:24:26,181][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:24:26,678][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:24:27,177][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:24:27,677][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:24:28,177][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:24:28,676][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:24:29,175][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:24:29,674][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:24:30,172][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:24:30,669][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:24:31,165][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:24:31,662][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:24:32,158][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:24:32,656][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:24:33,154][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:24:33,652][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:24:34,148][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:24:34,644][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:24:35,141][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:24:35,638][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:24:36,135][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:24:36,630][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:24:37,130][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:24:37,632][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:24:38,135][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:24:38,636][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:24:39,132][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:24:39,629][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:24:40,126][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:24:40,623][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:24:41,120][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:24:41,617][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:24:42,115][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:24:42,614][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10076 tokens. +[2026-03-25 23:24:43,267][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 62.07%, ΔTime: 00:00:32 +[2026-03-25 23:24:44,007][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:24:44,009][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:24:44,011][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:24:44,668][__main__][INFO] - Iteration 97 took 50s (29.68% Gen, 69.01% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 32m 6s. Estimated total time: 41h 56m 2s. Time estimates for 10 more iterations: 8m 23s, 100 more iterations: 1h 23m 52s, 500 more iterations: 6h 59m 20s. +[2026-03-25 23:24:44,671][__main__][INFO] - Starting iteration 97. +[2026-03-25 23:24:45,070][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 9 and human policies 1. +[2026-03-25 23:24:45,071][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:25:00,270][__main__][INFO] - Number of regex retries in iteration 97: 0 +[2026-03-25 23:25:00,270][__main__][INFO] - agents played in iteration 97 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:25:01,048][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:25:01,072][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:25:01,095][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:25:01,118][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:25:01,118][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:25:01,119][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:25:01,732][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:25:02,188][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:25:02,698][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:25:03,196][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:25:03,697][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:25:04,199][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:25:04,698][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:25:05,198][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:25:05,699][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:25:06,201][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:25:06,703][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:25:07,205][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:25:07,704][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:25:08,203][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:25:08,704][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:25:09,205][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:25:09,706][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:25:10,205][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:25:10,704][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:25:11,204][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:25:11,704][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:25:12,205][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:25:12,704][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:25:13,203][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:25:13,704][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:25:14,202][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:25:14,700][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:25:15,199][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:25:15,697][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:25:16,198][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:25:16,698][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:25:17,198][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:25:17,697][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:25:18,200][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:25:18,695][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:25:19,192][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:25:19,691][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:25:20,188][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:25:20,683][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:25:21,182][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:25:21,679][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:25:22,177][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:25:22,675][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:25:23,174][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:25:23,671][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:25:24,169][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:25:24,669][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:25:25,167][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:25:25,666][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:25:26,161][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:25:26,659][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:25:27,156][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:25:27,651][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:25:28,147][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:25:28,642][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:25:29,139][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:25:29,635][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:25:30,132][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:25:30,629][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:25:31,129][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:25:31,631][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:25:32,133][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:25:32,638][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:25:33,141][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:25:33,638][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10080 tokens. +[2026-03-25 23:25:34,304][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 61.93%, ΔTime: 00:00:32 +[2026-03-25 23:25:35,051][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:25:35,053][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:25:35,055][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:25:35,733][__main__][INFO] - Iteration 98 took 50s (30.00% Gen, 68.66% Train). Generation: 15s, Training: 34s. Estimated remaining time: 40h 48m 22s. Estimated total time: 42h 13m 9s. Time estimates for 10 more iterations: 8m 26s, 100 more iterations: 1h 24m 26s, 500 more iterations: 7h 2m 11s. +[2026-03-25 23:25:35,735][__main__][INFO] - Starting iteration 98. +[2026-03-25 23:25:36,135][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 9 and human policies 1. +[2026-03-25 23:25:36,135][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:25:42,956][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls + +Explanation: Given your per-item values and Bob's per-item values, hats are valuable to you and books are valuable to Bob. By proposing to take all 10 hats, you maximize your potential points from this round since hats are your highest valued item. Since books and balls are valued more by Bob, it would be strategic to keep those for him. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:25:51,621][__main__][INFO] - Number of regex retries in iteration 98: 1 +[2026-03-25 23:25:51,622][__main__][INFO] - agents played in iteration 98 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:25:52,401][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:25:52,424][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:25:52,448][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:25:52,471][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:25:52,472][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:25:52,472][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:25:53,095][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:25:53,554][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:25:54,056][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:25:54,557][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:25:55,057][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:25:55,561][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:25:56,058][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:25:56,559][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:25:57,060][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:25:57,558][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:25:58,057][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:25:58,557][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:25:59,055][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:25:59,552][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:26:00,050][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:26:00,550][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:26:01,047][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:26:01,542][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:26:02,041][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:26:02,539][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:26:03,037][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:26:03,535][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:26:04,032][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:26:04,530][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:26:05,027][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:26:05,523][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:26:06,019][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:26:06,515][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:26:07,013][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:26:07,510][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:26:08,006][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:26:08,503][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:26:09,000][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:26:09,496][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:26:09,994][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:26:10,491][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:26:10,988][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:26:11,484][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:26:11,980][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:26:12,477][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:26:12,974][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:26:13,472][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:26:13,969][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:26:14,467][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:26:14,966][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:26:15,464][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:26:15,962][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:26:16,461][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:26:16,961][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:26:17,460][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:26:17,960][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:26:18,458][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:26:18,956][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:26:19,455][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:26:19,961][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:26:20,461][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:26:20,963][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:26:21,458][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:26:21,954][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:26:22,449][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:26:22,945][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:26:23,441][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:26:23,937][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:26:24,435][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:26:24,932][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10101 tokens. +[2026-03-25 23:26:25,589][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.48%, Block Peak % of device VRAM: 62.06%, ΔTime: 00:00:32 +[2026-03-25 23:26:26,334][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:26:26,336][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:26:26,338][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:26:27,007][__main__][INFO] - Iteration 99 took 50s (30.44% Gen, 68.24% Train). Generation: 15s, Training: 34s. Estimated remaining time: 40h 58m 1s. Estimated total time: 42h 23m 39s. Time estimates for 10 more iterations: 8m 28s, 100 more iterations: 1h 24m 47s, 500 more iterations: 7h 3m 56s. +[2026-03-25 23:26:27,010][__main__][INFO] - Starting iteration 99. +[2026-03-25 23:26:27,412][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 9 and human policies 1. +[2026-03-25 23:26:27,413][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:26:42,840][__main__][INFO] - Number of regex retries in iteration 99: 0 +[2026-03-25 23:26:42,841][__main__][INFO] - agents played in iteration 99 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:26:43,620][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:26:43,643][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:26:43,667][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:26:43,690][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:26:43,690][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:26:43,691][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:26:44,314][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:26:44,773][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:26:45,280][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:26:45,781][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:26:46,283][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:26:46,784][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:26:47,286][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:26:47,790][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:26:48,293][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:26:48,792][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:26:49,291][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:26:49,791][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:26:50,293][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:26:50,795][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:26:51,294][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:26:51,793][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:26:52,294][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:26:52,793][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:26:53,294][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:26:53,793][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:26:54,294][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:26:54,792][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:26:55,291][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:26:55,789][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:26:56,292][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:26:56,791][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:26:57,290][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:26:57,790][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:26:58,290][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:26:58,791][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:26:59,289][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:26:59,789][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:27:00,289][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:27:00,787][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:27:01,288][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:27:01,787][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:27:02,286][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:27:02,784][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:27:03,283][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:27:03,781][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:27:04,282][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:27:04,779][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:27:05,278][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:27:05,776][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:27:06,274][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:27:06,772][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:27:07,269][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:27:07,767][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:27:08,264][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:27:08,762][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:27:09,260][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:27:09,758][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:27:10,256][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:27:10,755][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:27:11,253][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:27:11,751][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:27:12,249][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:27:12,748][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:27:13,247][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:27:13,745][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:27:14,243][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:27:14,740][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:27:15,238][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:27:15,736][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:27:16,233][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10107 tokens. +[2026-03-25 23:27:16,876][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 61.99%, ΔTime: 00:00:32 +[2026-03-25 23:27:17,620][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:27:17,622][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:27:17,624][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:27:18,282][__main__][INFO] - Iteration 100 took 50s (30.33% Gen, 68.37% Train). Generation: 15s, Training: 34s. Estimated remaining time: 40h 57m 8s. Estimated total time: 42h 23m 38s. Time estimates for 10 more iterations: 8m 28s, 100 more iterations: 1h 24m 47s, 500 more iterations: 7h 3m 56s. +[2026-03-25 23:27:18,284][__main__][INFO] - Starting iteration 100. +[2026-03-25 23:27:18,706][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 9 and human policies 1. +[2026-03-25 23:27:18,706][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:27:34,408][__main__][INFO] - Number of regex retries in iteration 100: 0 +[2026-03-25 23:27:34,409][__main__][INFO] - agents played in iteration 100 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:27:35,200][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:27:35,224][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:27:35,247][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:27:35,270][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:27:35,271][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:27:35,272][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:27:35,931][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:27:36,388][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:27:36,891][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:27:37,393][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:27:37,890][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:27:38,388][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:27:38,885][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:27:39,381][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:27:39,879][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:27:40,380][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:27:40,880][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:27:41,383][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:27:41,882][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:27:42,380][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:27:42,905][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:27:43,406][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:27:43,906][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:27:44,404][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:27:44,903][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:27:45,404][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:27:45,902][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:27:46,400][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:27:46,898][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:27:47,399][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:27:47,898][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:27:48,396][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:27:48,895][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:27:49,394][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:27:49,892][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:27:50,395][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:27:50,893][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:27:51,390][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:27:51,888][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:27:52,385][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:27:52,881][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:27:53,381][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:27:53,879][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:27:54,377][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:27:54,875][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:27:55,373][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:27:55,870][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:27:56,367][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:27:56,869][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:27:57,367][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:27:57,864][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:27:58,365][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:27:58,862][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:27:59,361][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:27:59,861][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:28:00,359][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:28:00,856][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:28:01,353][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:28:01,850][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:28:02,347][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:28:02,843][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:28:03,340][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:28:03,838][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:28:04,336][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:28:04,831][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:28:05,326][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:28:05,822][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:28:06,318][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:28:06,813][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:28:07,307][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:28:07,802][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10121 tokens. +[2026-03-25 23:28:08,453][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.08%, ΔTime: 00:00:32 +[2026-03-25 23:28:09,219][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:28:09,221][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:28:09,223][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:28:10,509][__main__][INFO] - Iteration 101 took 51s (30.31% Gen, 67.20% Train). Generation: 15s, Training: 34s. Estimated remaining time: 41h 42m 51s. Estimated total time: 43h 10m 13s. Time estimates for 10 more iterations: 8m 38s, 100 more iterations: 1h 26m 20s, 500 more iterations: 7h 11m 42s. +[2026-03-25 23:28:10,514][__main__][INFO] - Starting iteration 101. +[2026-03-25 23:28:10,915][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 10 and human policies 1. +[2026-03-25 23:28:10,916][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:28:26,762][__main__][INFO] - Number of regex retries in iteration 101: 0 +[2026-03-25 23:28:26,763][__main__][INFO] - agents played in iteration 101 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:28:27,538][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:28:27,561][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:28:27,585][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:28:27,608][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:28:27,609][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:28:27,609][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:28:28,232][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:28:28,686][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:28:29,190][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:28:29,690][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:28:30,189][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:28:30,686][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:28:31,186][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:28:31,684][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:28:32,187][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:28:32,685][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:28:33,182][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:28:33,677][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:28:34,174][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:28:34,669][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:28:35,164][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:28:35,659][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:28:36,153][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:28:36,650][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:28:37,146][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:28:37,643][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:28:38,142][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:28:38,641][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:28:39,142][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:28:39,641][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:28:40,142][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:28:40,641][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:28:41,139][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:28:41,639][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:28:42,136][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:28:42,636][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:28:43,133][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:28:43,631][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:28:44,131][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:28:44,627][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:28:45,125][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:28:45,623][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:28:46,121][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:28:46,618][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:28:47,116][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:28:47,613][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:28:48,110][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:28:48,610][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:28:49,107][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:28:49,607][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:28:50,107][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:28:50,604][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:28:51,104][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:28:51,600][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:28:52,099][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:28:52,598][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:28:53,123][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:28:53,621][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:28:54,119][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:28:54,619][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:28:55,119][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:28:55,618][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:28:56,122][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:28:56,617][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:28:57,114][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:28:57,610][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:28:58,105][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:28:58,600][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:28:59,099][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:28:59,603][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:29:00,104][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10123 tokens. +[2026-03-25 23:29:00,761][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.25%, ΔTime: 00:00:32 +[2026-03-25 23:29:01,521][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:29:01,524][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:29:01,526][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:29:02,184][__main__][INFO] - Iteration 102 took 51s (30.91% Gen, 67.80% Train). Generation: 15s, Training: 34s. Estimated remaining time: 41h 15m 15s. Estimated total time: 42h 43m 29s. Time estimates for 10 more iterations: 8m 32s, 100 more iterations: 1h 25m 26s, 500 more iterations: 7h 7m 14s. +[2026-03-25 23:29:02,187][__main__][INFO] - Starting iteration 102. +[2026-03-25 23:29:02,588][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 10 and human policies 1. +[2026-03-25 23:29:02,588][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:29:05,068][mllm.models.large_language_model_local][WARNING] - Response Proposal: 5 hats, 5 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:29:06,195][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:29:17,545][__main__][INFO] - Number of regex retries in iteration 102: 2 +[2026-03-25 23:29:17,546][__main__][INFO] - agents played in iteration 102 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:29:18,328][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:29:18,352][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:29:18,375][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:29:18,399][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:29:18,400][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:29:18,400][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:29:19,028][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:29:19,481][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:29:19,984][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:29:20,484][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:29:20,982][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:29:21,479][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:29:21,979][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:29:22,476][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:29:22,978][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:29:23,476][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:29:23,972][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:29:24,467][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:29:24,965][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:29:25,463][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:29:25,959][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:29:26,456][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:29:26,951][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:29:27,448][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:29:27,945][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:29:28,442][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:29:28,939][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:29:29,459][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:29:29,957][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:29:30,455][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:29:30,952][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:29:31,450][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:29:31,947][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:29:32,445][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:29:32,944][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:29:33,443][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:29:33,941][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:29:34,438][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:29:34,937][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:29:35,434][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:29:35,933][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:29:36,430][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:29:36,928][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:29:37,430][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:29:37,929][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:29:38,428][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:29:38,923][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:29:39,422][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:29:39,918][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:29:40,414][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:29:40,914][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:29:41,410][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:29:41,910][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:29:42,406][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:29:42,905][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:29:43,405][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:29:43,902][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:29:44,400][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:29:44,897][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:29:45,393][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:29:45,889][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:29:46,388][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:29:46,884][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:29:47,381][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:29:47,885][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:29:48,386][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:29:48,885][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:29:49,383][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:29:49,883][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:29:50,390][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:29:50,900][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10092 tokens. +[2026-03-25 23:29:51,594][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.16%, Current % of VRAM taken: 60.58%, Block Peak % of device VRAM: 62.13%, ΔTime: 00:00:32 +[2026-03-25 23:29:52,349][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:29:52,351][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:29:52,353][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:29:53,010][__main__][INFO] - Iteration 103 took 50s (29.66% Gen, 69.03% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 32m 4s. Estimated total time: 42h 1m 9s. Time estimates for 10 more iterations: 8m 24s, 100 more iterations: 1h 24m 2s, 500 more iterations: 7h 0m 11s. +[2026-03-25 23:29:53,013][__main__][INFO] - Starting iteration 103. +[2026-03-25 23:29:53,412][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 10 and human policies 1. +[2026-03-25 23:29:53,413][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:30:09,125][__main__][INFO] - Number of regex retries in iteration 103: 0 +[2026-03-25 23:30:09,126][__main__][INFO] - agents played in iteration 103 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:30:09,902][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:30:09,926][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:30:09,949][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:30:09,972][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:30:09,972][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:30:09,973][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:30:10,612][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:30:11,064][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:30:11,568][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:30:12,065][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:30:12,564][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:30:13,061][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:30:13,560][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:30:14,060][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:30:14,558][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:30:15,061][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:30:15,560][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:30:16,060][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:30:16,564][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:30:17,062][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:30:17,560][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:30:18,058][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:30:18,560][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:30:19,060][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:30:19,564][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:30:20,061][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:30:20,561][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:30:21,058][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:30:21,558][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:30:22,058][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:30:22,556][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:30:23,058][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:30:23,556][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:30:24,053][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:30:24,551][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:30:25,047][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:30:25,544][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:30:26,040][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:30:26,536][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:30:27,034][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:30:27,535][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:30:28,034][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:30:28,531][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:30:29,030][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:30:29,529][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:30:30,028][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:30:30,527][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:30:31,025][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:30:31,524][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:30:32,022][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:30:32,522][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:30:33,022][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:30:33,522][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:30:34,022][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:30:34,544][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:30:35,042][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:30:35,541][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:30:36,041][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:30:36,541][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:30:37,045][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:30:37,545][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:30:38,046][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:30:38,548][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:30:39,050][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:30:39,549][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:30:40,048][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:30:40,543][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:30:41,038][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:30:41,534][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:30:42,028][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:30:42,523][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10191 tokens. +[2026-03-25 23:30:43,180][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 62.26%, ΔTime: 00:00:32 +[2026-03-25 23:30:43,937][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:30:43,939][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:30:43,941][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:30:44,624][__main__][INFO] - Iteration 104 took 51s (30.68% Gen, 67.98% Train). Generation: 15s, Training: 34s. Estimated remaining time: 41h 10m 41s. Estimated total time: 42h 40m 37s. Time estimates for 10 more iterations: 8m 32s, 100 more iterations: 1h 25m 21s, 500 more iterations: 7h 6m 46s. +[2026-03-25 23:30:44,627][__main__][INFO] - Starting iteration 104. +[2026-03-25 23:30:45,026][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 10 and human policies 1. +[2026-03-25 23:30:45,027][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:31:00,290][__main__][INFO] - Number of regex retries in iteration 104: 0 +[2026-03-25 23:31:00,291][__main__][INFO] - agents played in iteration 104 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:31:01,063][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:31:01,087][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:31:01,110][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:31:01,134][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:31:01,134][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:31:01,135][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:31:01,754][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:31:02,210][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:31:02,709][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:31:03,205][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:31:03,703][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:31:04,199][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:31:04,697][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:31:05,193][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:31:05,688][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:31:06,186][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:31:06,684][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:31:07,184][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:31:07,685][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:31:08,184][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:31:08,681][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:31:09,179][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:31:09,678][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:31:10,176][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:31:10,678][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:31:11,175][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:31:11,671][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:31:12,168][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:31:12,665][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:31:13,164][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:31:13,662][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:31:14,187][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:31:14,688][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:31:15,188][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:31:15,688][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:31:16,187][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:31:16,687][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:31:17,187][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:31:17,698][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:31:18,197][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:31:18,696][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:31:19,193][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:31:19,688][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:31:20,185][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:31:20,681][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:31:21,177][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:31:21,675][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:31:22,174][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:31:22,673][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:31:23,171][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:31:23,667][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:31:24,166][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:31:24,664][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:31:25,164][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:31:25,662][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:31:26,158][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:31:26,654][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:31:27,150][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:31:27,648][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:31:28,146][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:31:28,647][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:31:29,145][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:31:29,641][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:31:30,137][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:31:30,634][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:31:31,128][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:31:31,624][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:31:32,119][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:31:32,615][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:31:33,112][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:31:33,608][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10053 tokens. +[2026-03-25 23:31:34,267][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.36%, ΔTime: 00:00:32 +[2026-03-25 23:31:35,026][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:31:35,029][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:31:35,030][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:31:35,699][__main__][INFO] - Iteration 105 took 50s (30.12% Gen, 68.56% Train). Generation: 15s, Training: 34s. Estimated remaining time: 40h 42m 54s. Estimated total time: 42h 13m 41s. Time estimates for 10 more iterations: 8m 26s, 100 more iterations: 1h 24m 27s, 500 more iterations: 7h 2m 16s. +[2026-03-25 23:31:35,701][__main__][INFO] - Starting iteration 105. +[2026-03-25 23:31:36,101][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 10 and human policies 1. +[2026-03-25 23:31:36,102][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:31:51,256][__main__][INFO] - Number of regex retries in iteration 105: 0 +[2026-03-25 23:31:51,258][__main__][INFO] - agents played in iteration 105 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:31:52,033][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:31:52,056][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:31:52,080][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:31:52,103][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:31:52,103][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:31:52,104][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:31:52,744][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:31:53,196][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:31:53,699][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:31:54,196][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:31:54,692][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:31:55,190][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:31:55,688][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:31:56,186][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:31:56,684][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:31:57,183][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:31:57,681][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:31:58,181][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:31:58,680][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:31:59,178][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:31:59,678][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:32:00,176][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:32:00,675][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:32:01,173][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:32:01,678][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:32:02,177][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:32:02,677][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:32:03,177][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:32:03,677][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:32:04,184][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:32:04,684][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:32:05,181][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:32:05,680][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:32:06,179][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:32:06,678][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:32:07,177][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:32:07,677][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:32:08,177][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:32:08,676][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:32:09,174][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:32:09,674][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:32:10,172][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:32:10,674][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:32:11,170][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:32:11,671][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:32:12,167][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:32:12,664][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:32:13,164][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:32:13,660][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:32:14,157][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:32:14,657][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:32:15,155][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:32:15,652][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:32:16,148][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:32:16,644][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:32:17,141][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:32:17,642][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:32:18,145][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:32:18,648][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:32:19,154][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:32:19,650][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:32:20,152][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:32:20,652][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:32:21,150][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:32:21,648][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:32:22,148][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:32:22,644][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:32:23,143][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:32:23,643][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:32:24,142][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:32:24,640][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10024 tokens. +[2026-03-25 23:32:25,325][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.05%, Current % of VRAM taken: 60.47%, Block Peak % of device VRAM: 61.97%, ΔTime: 00:00:32 +[2026-03-25 23:32:26,098][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:32:26,101][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:32:26,103][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:32:26,778][__main__][INFO] - Iteration 106 took 50s (29.91% Gen, 68.76% Train). Generation: 15s, Training: 34s. Estimated remaining time: 40h 42m 12s. Estimated total time: 42h 13m 50s. Time estimates for 10 more iterations: 8m 26s, 100 more iterations: 1h 24m 27s, 500 more iterations: 7h 2m 18s. +[2026-03-25 23:32:26,780][__main__][INFO] - Starting iteration 106. +[2026-03-25 23:32:27,182][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 10 and human policies 1. +[2026-03-25 23:32:27,183][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:32:38,214][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls + +Given the per-item values, I aim to maximize my points by prioritizing items based on their value to me. Hats are worth 10 points each to me, while books are worth 10 points each to Bob, and balls are worth 1 point each to me. Since I don't need to consider proportionality issues and I have the highest value for hats, I will propose to take all 10 hats. This maximizes my potential points for this round since taking all hats ensures no chance of proportional distribution that would reduce my points. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:32:44,274][__main__][INFO] - Number of regex retries in iteration 106: 1 +[2026-03-25 23:32:44,276][__main__][INFO] - agents played in iteration 106 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:32:45,052][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:32:45,075][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:32:45,099][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:32:45,122][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:32:45,122][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:32:45,123][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:32:45,742][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:32:46,194][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:32:46,697][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:32:47,194][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:32:47,691][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:32:48,191][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:32:48,686][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:32:49,184][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:32:49,682][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:32:50,183][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:32:50,681][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:32:51,179][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:32:51,678][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:32:52,177][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:32:52,675][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:32:53,175][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:32:53,674][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:32:54,171][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:32:54,672][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:32:55,172][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:32:55,671][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:32:56,168][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:32:56,665][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:32:57,163][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:32:57,661][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:32:58,157][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:32:58,653][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:32:59,150][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:32:59,648][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:33:00,146][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:33:00,646][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:33:01,142][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:33:01,638][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:33:02,135][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:33:02,635][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:33:03,133][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:33:03,650][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:33:04,146][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:33:04,642][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:33:05,140][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:33:05,637][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:33:06,133][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:33:06,628][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:33:07,125][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:33:07,620][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:33:08,115][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:33:08,609][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:33:09,106][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:33:09,609][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:33:10,116][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:33:10,616][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:33:11,113][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:33:11,615][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:33:12,113][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:33:12,609][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:33:13,111][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:33:13,610][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:33:14,106][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:33:14,601][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:33:15,097][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:33:15,592][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:33:16,088][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:33:16,585][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:33:17,081][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:33:17,580][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10166 tokens. +[2026-03-25 23:33:18,237][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.01%, ΔTime: 00:00:32 +[2026-03-25 23:33:19,003][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:33:19,006][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:33:19,007][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:33:19,676][__main__][INFO] - Iteration 107 took 52s (32.56% Gen, 66.16% Train). Generation: 17s, Training: 34s. Estimated remaining time: 42h 12m 13s. Estimated total time: 43h 44m 44s. Time estimates for 10 more iterations: 8m 44s, 100 more iterations: 1h 27m 29s, 500 more iterations: 7h 17m 27s. +[2026-03-25 23:33:19,678][__main__][INFO] - Starting iteration 107. +[2026-03-25 23:33:20,081][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 10 and human policies 1. +[2026-03-25 23:33:20,081][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:33:36,042][__main__][INFO] - Number of regex retries in iteration 107: 0 +[2026-03-25 23:33:36,043][__main__][INFO] - agents played in iteration 107 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:33:36,821][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:33:36,845][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:33:36,869][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:33:36,893][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:33:36,893][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:33:36,894][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:33:37,519][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:33:37,972][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:33:38,478][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:33:38,978][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:33:39,479][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:33:39,980][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:33:40,478][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:33:40,978][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:33:41,479][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:33:41,983][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:33:42,481][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:33:42,983][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:33:43,483][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:33:43,983][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:33:44,482][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:33:44,981][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:33:45,484][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:33:45,983][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:33:46,481][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:33:46,979][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:33:47,478][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:33:47,978][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:33:48,477][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:33:48,975][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:33:49,473][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:33:49,970][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:33:50,467][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:33:50,964][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:33:51,462][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:33:51,962][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:33:52,461][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:33:52,958][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:33:53,456][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:33:53,953][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:33:54,450][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:33:54,947][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:33:55,443][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:33:55,940][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:33:56,438][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:33:56,940][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:33:57,438][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:33:57,937][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:33:58,433][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:33:58,930][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:33:59,427][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:33:59,926][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:34:00,428][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:34:00,932][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:34:01,436][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:34:01,940][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:34:02,438][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:34:02,936][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:34:03,434][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:34:03,934][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:34:04,433][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:34:04,930][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:34:05,428][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:34:05,927][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:34:06,430][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:34:06,927][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:34:07,428][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:34:07,928][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:34:08,428][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:34:08,927][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:34:09,426][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10207 tokens. +[2026-03-25 23:34:10,092][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.16%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.04%, ΔTime: 00:00:32 +[2026-03-25 23:34:10,875][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:34:10,878][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:34:10,880][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:34:11,571][__main__][INFO] - Iteration 108 took 51s (31.00% Gen, 67.66% Train). Generation: 15s, Training: 34s. Estimated remaining time: 41h 21m 9s. Estimated total time: 42h 54m 32s. Time estimates for 10 more iterations: 8m 34s, 100 more iterations: 1h 25m 49s, 500 more iterations: 7h 9m 5s. +[2026-03-25 23:34:11,573][__main__][INFO] - Starting iteration 108. +[2026-03-25 23:34:11,974][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 10 and human policies 1. +[2026-03-25 23:34:11,975][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:34:27,002][__main__][INFO] - Number of regex retries in iteration 108: 0 +[2026-03-25 23:34:27,003][__main__][INFO] - agents played in iteration 108 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:34:27,777][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:34:27,800][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:34:27,823][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:34:27,847][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:34:27,847][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:34:27,848][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:34:28,481][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:34:28,933][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:34:29,437][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:34:29,933][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:34:30,429][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:34:30,926][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:34:31,422][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:34:31,919][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:34:32,417][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:34:32,917][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:34:33,416][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:34:33,917][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:34:34,417][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:34:34,915][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:34:35,411][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:34:35,913][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:34:36,413][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:34:36,911][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:34:37,407][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:34:37,903][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:34:38,401][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:34:38,896][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:34:39,392][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:34:39,888][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:34:40,386][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:34:40,886][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:34:41,384][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:34:41,882][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:34:42,381][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:34:42,879][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:34:43,403][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:34:43,901][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:34:44,399][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:34:44,895][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:34:45,392][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:34:45,890][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:34:46,388][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:34:46,885][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:34:47,381][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:34:47,878][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:34:48,376][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:34:48,874][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:34:49,371][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:34:49,869][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:34:50,375][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:34:50,883][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:34:51,385][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:34:51,884][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:34:52,384][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:34:52,883][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:34:53,381][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:34:53,879][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:34:54,379][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:34:54,878][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:34:55,377][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:34:55,876][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:34:56,373][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:34:56,873][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:34:57,370][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:34:57,868][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:34:58,365][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:34:58,862][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:34:59,360][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:34:59,858][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:35:00,358][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10025 tokens. +[2026-03-25 23:35:01,022][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.15%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 61.94%, ΔTime: 00:00:32 +[2026-03-25 23:35:01,783][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:35:01,786][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:35:01,787][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:35:02,456][__main__][INFO] - Iteration 109 took 50s (29.77% Gen, 68.90% Train). Generation: 15s, Training: 34s. Estimated remaining time: 40h 29m 51s. Estimated total time: 42h 4m 5s. Time estimates for 10 more iterations: 8m 24s, 100 more iterations: 1h 24m 8s, 500 more iterations: 7h 0m 40s. +[2026-03-25 23:35:02,458][__main__][INFO] - Starting iteration 109. +[2026-03-25 23:35:02,857][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 10 and human policies 1. +[2026-03-25 23:35:02,858][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:35:12,282][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:35:13,746][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:35:18,372][__main__][INFO] - Number of regex retries in iteration 109: 2 +[2026-03-25 23:35:18,373][__main__][INFO] - agents played in iteration 109 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:35:19,152][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:35:19,175][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:35:19,198][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:35:19,221][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:35:19,222][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:35:19,223][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:35:19,842][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:35:20,296][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:35:20,801][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:35:21,301][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:35:21,801][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:35:22,301][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:35:22,798][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:35:23,297][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:35:23,796][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:35:24,294][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:35:24,803][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:35:25,305][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:35:25,803][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:35:26,301][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:35:26,804][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:35:27,307][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:35:27,809][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:35:28,307][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:35:28,803][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:35:29,302][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:35:29,800][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:35:30,299][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:35:30,796][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:35:31,294][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:35:31,792][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:35:32,290][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:35:32,787][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:35:33,285][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:35:33,781][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:35:34,279][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:35:34,779][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:35:35,277][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:35:35,775][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:35:36,273][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:35:36,770][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:35:37,267][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:35:37,763][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:35:38,262][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:35:38,760][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:35:39,258][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:35:39,756][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:35:40,252][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:35:40,749][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:35:41,250][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:35:41,750][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:35:42,247][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:35:42,742][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:35:43,237][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:35:43,731][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:35:44,227][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:35:44,721][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:35:45,216][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:35:45,714][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:35:46,210][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:35:46,704][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:35:47,201][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:35:47,699][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:35:48,198][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:35:48,696][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:35:49,194][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:35:49,690][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:35:50,188][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:35:50,685][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:35:51,183][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:35:51,683][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10059 tokens. +[2026-03-25 23:35:52,342][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.60%, Block Peak % of device VRAM: 62.15%, ΔTime: 00:00:32 +[2026-03-25 23:35:53,099][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:35:53,102][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:35:53,104][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:35:53,825][__main__][INFO] - Iteration 110 took 50s (30.44% Gen, 68.14% Train). Generation: 15s, Training: 34s. Estimated remaining time: 40h 53m 22s. Estimated total time: 42h 28m 27s. Time estimates for 10 more iterations: 8m 29s, 100 more iterations: 1h 24m 56s, 500 more iterations: 7h 4m 44s. +[2026-03-25 23:35:53,828][__main__][INFO] - Starting iteration 110. +[2026-03-25 23:35:54,228][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 10 and human policies 1. +[2026-03-25 23:35:54,229][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:36:09,097][__main__][INFO] - Number of regex retries in iteration 110: 0 +[2026-03-25 23:36:09,097][__main__][INFO] - agents played in iteration 110 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:36:09,868][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:36:09,891][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:36:09,914][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:36:09,937][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:36:09,938][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:36:09,938][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:36:10,555][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:36:11,011][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:36:11,512][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:36:12,009][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:36:12,508][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:36:13,007][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:36:13,505][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:36:14,003][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:36:14,501][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:36:14,998][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:36:15,496][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:36:15,995][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:36:16,495][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:36:16,994][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:36:17,491][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:36:17,994][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:36:18,494][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:36:18,991][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:36:19,489][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:36:19,987][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:36:20,489][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:36:20,985][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:36:21,482][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:36:21,979][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:36:22,476][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:36:22,974][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:36:23,471][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:36:23,971][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:36:24,469][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:36:24,967][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:36:25,466][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:36:25,963][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:36:26,461][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:36:26,959][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:36:27,456][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:36:27,952][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:36:28,449][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:36:28,947][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:36:29,445][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:36:29,942][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:36:30,446][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:36:30,949][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:36:31,454][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:36:31,950][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:36:32,445][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:36:32,940][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:36:33,437][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:36:33,934][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:36:34,429][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:36:34,926][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:36:35,422][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:36:35,918][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:36:36,413][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:36:36,909][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:36:37,405][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:36:37,901][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:36:38,397][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:36:38,893][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:36:39,390][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:36:39,890][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:36:40,388][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:36:40,884][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:36:41,382][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:36:41,881][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:36:42,378][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10062 tokens. +[2026-03-25 23:36:43,036][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 61.96%, ΔTime: 00:00:32 +[2026-03-25 23:36:43,794][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:36:43,797][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:36:43,798][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:36:45,114][__main__][INFO] - Iteration 111 took 50s (29.22% Gen, 68.19% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 48m 22s. Estimated total time: 42h 24m 18s. Time estimates for 10 more iterations: 8m 28s, 100 more iterations: 1h 24m 48s, 500 more iterations: 7h 4m 3s. +[2026-03-25 23:36:45,117][__main__][INFO] - Starting iteration 111. +[2026-03-25 23:36:45,518][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 11 and human policies 1. +[2026-03-25 23:36:45,518][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:36:48,212][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:36:54,059][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:36:56,802][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls + +Given the values: +- You value hats at 10 and Bob values them at 1. +- You value books at 10 and Bob values them at 10. +- You value balls at 1 and Bob values them at 10. + +Since you have a much higher value for hats compared to Bob, and you have 10 hats to split, it makes sense to propose keeping all 10 hats. This ensures you get the maximum possible points from the hats. The books and balls have lower comparative values for you, so it's better to allocate them to Bob. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:37:02,066][__main__][INFO] - Number of regex retries in iteration 111: 3 +[2026-03-25 23:37:02,067][__main__][INFO] - agents played in iteration 111 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:37:02,836][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:37:02,860][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:37:02,883][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:37:02,906][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:37:02,906][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:37:02,907][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:37:03,517][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:37:03,974][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:37:04,479][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:37:04,980][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:37:05,479][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:37:05,979][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:37:06,478][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:37:06,978][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:37:07,480][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:37:07,983][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:37:08,486][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:37:08,987][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:37:09,487][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:37:09,987][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:37:10,491][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:37:10,992][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:37:11,489][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:37:11,986][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:37:12,483][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:37:12,982][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:37:13,480][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:37:13,977][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:37:14,477][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:37:14,975][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:37:15,472][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:37:15,968][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:37:16,465][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:37:16,961][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:37:17,462][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:37:17,960][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:37:18,459][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:37:18,956][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:37:19,453][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:37:19,949][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:37:20,450][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:37:20,952][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:37:21,452][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:37:21,953][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:37:22,450][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:37:22,946][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:37:23,442][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:37:23,941][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:37:24,438][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:37:24,935][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:37:25,431][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:37:25,928][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:37:26,427][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:37:26,924][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:37:27,425][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:37:27,925][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:37:28,424][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:37:28,921][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:37:29,420][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:37:29,919][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:37:30,420][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:37:30,920][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:37:31,420][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:37:31,917][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:37:32,414][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:37:32,910][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:37:33,405][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:37:33,900][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:37:34,397][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:37:34,892][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:37:35,387][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10082 tokens. +[2026-03-25 23:37:36,038][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.06%, ΔTime: 00:00:32 +[2026-03-25 23:37:36,800][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:37:36,802][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:37:36,804][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:37:37,508][__main__][INFO] - Iteration 112 took 51s (31.83% Gen, 66.81% Train). Generation: 16s, Training: 34s. Estimated remaining time: 41h 42m 45s. Estimated total time: 43h 19m 33s. Time estimates for 10 more iterations: 8m 39s, 100 more iterations: 1h 26m 39s, 500 more iterations: 7h 13m 15s. +[2026-03-25 23:37:37,511][__main__][INFO] - Starting iteration 112. +[2026-03-25 23:37:37,911][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 11 and human policies 1. +[2026-03-25 23:37:37,911][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:37:53,597][__main__][INFO] - Number of regex retries in iteration 112: 0 +[2026-03-25 23:37:53,598][__main__][INFO] - agents played in iteration 112 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:37:54,379][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:37:54,403][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:37:54,426][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:37:54,449][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:37:54,450][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:37:54,450][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:37:55,068][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:37:55,519][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:37:56,023][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:37:56,521][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:37:57,019][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:37:57,518][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:37:58,017][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:37:58,515][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:37:59,018][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:37:59,519][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:38:00,017][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:38:00,515][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:38:01,017][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:38:01,519][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:38:02,021][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:38:02,521][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:38:03,018][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:38:03,515][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:38:04,014][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:38:04,511][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:38:05,008][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:38:05,506][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:38:06,004][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:38:06,502][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:38:07,000][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:38:07,498][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:38:07,995][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:38:08,492][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:38:08,988][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:38:09,485][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:38:09,981][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:38:10,478][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:38:10,976][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:38:11,480][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:38:11,983][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:38:12,490][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:38:12,987][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:38:13,485][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:38:13,982][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:38:14,479][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:38:14,981][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:38:15,479][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:38:15,977][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:38:16,476][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:38:16,976][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:38:17,474][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:38:17,971][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:38:18,468][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:38:18,964][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:38:19,462][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:38:19,960][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:38:20,459][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:38:20,958][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:38:21,457][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:38:21,956][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:38:22,453][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:38:22,951][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:38:23,447][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:38:23,945][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:38:24,444][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:38:24,939][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:38:25,437][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:38:25,935][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:38:26,431][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:38:26,929][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10061 tokens. +[2026-03-25 23:38:27,588][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 61.95%, ΔTime: 00:00:32 +[2026-03-25 23:38:28,344][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:38:28,346][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:38:28,348][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:38:29,048][__main__][INFO] - Iteration 113 took 51s (30.68% Gen, 67.95% Train). Generation: 15s, Training: 34s. Estimated remaining time: 40h 59m 13s. Estimated total time: 42h 36m 53s. Time estimates for 10 more iterations: 8m 31s, 100 more iterations: 1h 25m 13s, 500 more iterations: 7h 6m 8s. +[2026-03-25 23:38:29,050][__main__][INFO] - Starting iteration 113. +[2026-03-25 23:38:29,452][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 11 and human policies 1. +[2026-03-25 23:38:29,453][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:38:45,147][__main__][INFO] - Number of regex retries in iteration 113: 0 +[2026-03-25 23:38:45,148][__main__][INFO] - agents played in iteration 113 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:38:45,924][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:38:45,948][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:38:45,971][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:38:45,994][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:38:45,995][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:38:45,995][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:38:46,609][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:38:47,061][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:38:47,560][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:38:48,058][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:38:48,558][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:38:49,055][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:38:49,551][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:38:50,047][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:38:50,544][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:38:51,044][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:38:51,545][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:38:52,046][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:38:52,547][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:38:53,047][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:38:53,547][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:38:54,047][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:38:54,566][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:38:55,062][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:38:55,558][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:38:56,055][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:38:56,550][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:38:57,047][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:38:57,545][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:38:58,043][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:38:58,540][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:38:59,037][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:38:59,535][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:39:00,033][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:39:00,531][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:39:01,030][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:39:01,534][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:39:02,037][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:39:02,544][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:39:03,040][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:39:03,537][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:39:04,033][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:39:04,530][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:39:05,025][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:39:05,520][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:39:06,016][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:39:06,512][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:39:07,011][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:39:07,508][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:39:08,008][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:39:08,504][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:39:09,001][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:39:09,499][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:39:09,998][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:39:10,494][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:39:10,990][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:39:11,486][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:39:11,983][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:39:12,480][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:39:12,981][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:39:13,478][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:39:13,978][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:39:14,474][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:39:14,970][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:39:15,469][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:39:15,967][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:39:16,463][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:39:16,964][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:39:17,461][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:39:17,964][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:39:18,463][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10101 tokens. +[2026-03-25 23:39:19,125][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.17%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.02%, ΔTime: 00:00:32 +[2026-03-25 23:39:19,879][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:39:19,881][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:39:19,883][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:39:20,555][__main__][INFO] - Iteration 114 took 51s (30.71% Gen, 67.97% Train). Generation: 15s, Training: 34s. Estimated remaining time: 40h 56m 38s. Estimated total time: 42h 35m 10s. Time estimates for 10 more iterations: 8m 31s, 100 more iterations: 1h 25m 10s, 500 more iterations: 7h 5m 51s. +[2026-03-25 23:39:20,557][__main__][INFO] - Starting iteration 114. +[2026-03-25 23:39:20,958][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 11 and human policies 1. +[2026-03-25 23:39:20,959][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:39:36,797][__main__][INFO] - Number of regex retries in iteration 114: 0 +[2026-03-25 23:39:36,798][__main__][INFO] - agents played in iteration 114 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:39:37,571][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:39:37,594][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:39:37,618][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:39:37,641][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:39:37,641][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:39:37,642][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:39:38,269][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:39:38,720][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:39:39,220][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:39:39,719][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:39:40,218][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:39:40,715][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:39:41,212][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:39:41,707][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:39:42,205][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:39:42,703][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:39:43,200][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:39:43,696][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:39:44,192][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:39:44,693][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:39:45,192][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:39:45,689][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:39:46,185][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:39:46,680][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:39:47,176][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:39:47,672][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:39:48,168][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:39:48,664][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:39:49,160][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:39:49,656][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:39:50,160][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:39:50,684][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:39:51,188][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:39:51,687][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:39:52,185][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:39:52,686][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:39:53,186][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:39:53,682][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:39:54,180][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:39:54,680][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:39:55,178][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:39:55,676][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:39:56,177][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:39:56,677][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:39:57,175][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:39:57,673][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:39:58,170][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:39:58,667][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:39:59,163][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:39:59,659][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:40:00,156][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:40:00,652][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:40:01,149][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:40:01,646][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:40:02,142][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:40:02,638][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:40:03,131][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:40:03,626][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:40:04,122][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:40:04,619][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:40:05,114][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:40:05,610][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:40:06,106][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:40:06,606][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:40:07,103][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:40:07,601][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:40:08,099][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:40:08,598][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:40:09,095][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:40:09,593][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:40:10,089][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10025 tokens. +[2026-03-25 23:40:10,755][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 61.98%, ΔTime: 00:00:32 +[2026-03-25 23:40:11,540][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:40:11,542][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:40:11,544][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:40:12,273][__main__][INFO] - Iteration 115 took 51s (30.87% Gen, 67.71% Train). Generation: 15s, Training: 34s. Estimated remaining time: 41h 6m 23s. Estimated total time: 42h 45m 46s. Time estimates for 10 more iterations: 8m 33s, 100 more iterations: 1h 25m 31s, 500 more iterations: 7h 7m 37s. +[2026-03-25 23:40:12,275][__main__][INFO] - Starting iteration 115. +[2026-03-25 23:40:12,677][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 11 and human policies 1. +[2026-03-25 23:40:12,677][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:40:28,370][__main__][INFO] - Number of regex retries in iteration 115: 0 +[2026-03-25 23:40:28,371][__main__][INFO] - agents played in iteration 115 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:40:29,148][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:40:29,171][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:40:29,194][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:40:29,217][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:40:29,218][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:40:29,219][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:40:29,851][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:40:30,304][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:40:30,807][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:40:31,306][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:40:31,804][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:40:32,303][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:40:32,801][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:40:33,301][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:40:33,801][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:40:34,303][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:40:34,802][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:40:35,302][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:40:35,805][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:40:36,304][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:40:36,805][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:40:37,304][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:40:37,803][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:40:38,303][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:40:38,799][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:40:39,297][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:40:39,795][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:40:40,293][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:40:40,790][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:40:41,289][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:40:41,794][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:40:42,296][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:40:42,798][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:40:43,295][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:40:43,795][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:40:44,291][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:40:44,788][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:40:45,284][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:40:45,780][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:40:46,278][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:40:46,776][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:40:47,274][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:40:47,771][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:40:48,268][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:40:48,769][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:40:49,266][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:40:49,764][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:40:50,264][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:40:50,761][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:40:51,259][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:40:51,758][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:40:52,255][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:40:52,752][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:40:53,249][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:40:53,746][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:40:54,243][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:40:54,743][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:40:55,242][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:40:55,738][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:40:56,237][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:40:56,737][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:40:57,232][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:40:57,729][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:40:58,226][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:40:58,722][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:40:59,218][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:40:59,714][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:41:00,211][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:41:00,708][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:41:01,205][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:41:01,701][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10042 tokens. +[2026-03-25 23:41:02,369][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.00%, Current % of VRAM taken: 60.43%, Block Peak % of device VRAM: 61.89%, ΔTime: 00:00:32 +[2026-03-25 23:41:03,127][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:41:03,129][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:41:03,131][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:41:03,843][__main__][INFO] - Iteration 116 took 51s (30.67% Gen, 67.93% Train). Generation: 15s, Training: 34s. Estimated remaining time: 40h 58m 12s. Estimated total time: 42h 38m 27s. Time estimates for 10 more iterations: 8m 31s, 100 more iterations: 1h 25m 16s, 500 more iterations: 7h 6m 24s. +[2026-03-25 23:41:03,846][__main__][INFO] - Starting iteration 116. +[2026-03-25 23:41:04,245][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 11 and human policies 1. +[2026-03-25 23:41:04,246][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:41:19,474][__main__][INFO] - Number of regex retries in iteration 116: 0 +[2026-03-25 23:41:19,476][__main__][INFO] - agents played in iteration 116 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:41:20,249][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:41:20,273][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:41:20,296][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:41:20,319][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:41:20,319][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:41:20,320][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:41:20,951][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:41:21,408][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:41:21,910][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:41:22,408][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:41:22,905][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:41:23,401][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:41:23,898][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:41:24,398][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:41:24,897][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:41:25,397][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:41:25,896][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:41:26,397][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:41:26,896][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:41:27,397][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:41:27,896][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:41:28,395][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:41:28,914][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:41:29,411][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:41:29,908][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:41:30,407][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:41:30,907][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:41:31,409][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:41:31,909][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:41:32,411][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:41:32,910][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:41:33,406][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:41:33,903][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:41:34,401][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:41:34,899][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:41:35,397][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:41:35,896][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:41:36,392][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:41:36,890][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:41:37,388][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:41:37,890][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:41:38,388][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:41:38,885][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:41:39,384][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:41:39,884][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:41:40,382][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:41:40,880][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:41:41,377][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:41:41,876][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:41:42,374][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:41:42,871][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:41:43,366][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:41:43,861][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:41:44,360][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:41:44,857][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:41:45,356][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:41:45,853][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:41:46,350][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:41:46,846][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:41:47,343][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:41:47,843][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:41:48,343][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:41:48,841][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:41:49,338][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:41:49,835][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:41:50,334][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:41:50,829][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:41:51,326][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:41:51,820][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:41:52,315][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:41:52,812][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9989 tokens. +[2026-03-25 23:41:53,488][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.05%, Current % of VRAM taken: 60.48%, Block Peak % of device VRAM: 61.93%, ΔTime: 00:00:32 +[2026-03-25 23:41:54,243][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:41:54,246][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:41:54,247][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:41:54,958][__main__][INFO] - Iteration 117 took 50s (30.03% Gen, 68.57% Train). Generation: 15s, Training: 34s. Estimated remaining time: 40h 34m 32s. Estimated total time: 42h 15m 38s. Time estimates for 10 more iterations: 8m 27s, 100 more iterations: 1h 24m 31s, 500 more iterations: 7h 2m 36s. +[2026-03-25 23:41:54,960][__main__][INFO] - Starting iteration 117. +[2026-03-25 23:41:55,359][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 11 and human policies 1. +[2026-03-25 23:41:55,360][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:42:10,847][__main__][INFO] - Number of regex retries in iteration 117: 0 +[2026-03-25 23:42:10,848][__main__][INFO] - agents played in iteration 117 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:42:11,630][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:42:11,654][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:42:11,677][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:42:11,700][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:42:11,701][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:42:11,701][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:42:12,326][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:42:12,778][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:42:13,281][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:42:13,778][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:42:14,275][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:42:14,772][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:42:15,271][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:42:15,767][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:42:16,265][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:42:16,761][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:42:17,259][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:42:17,757][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:42:18,254][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:42:18,752][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:42:19,248][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:42:19,744][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:42:20,242][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:42:20,738][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:42:21,240][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:42:21,738][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:42:22,235][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:42:22,736][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:42:23,236][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:42:23,742][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:42:24,244][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:42:24,743][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:42:25,240][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:42:25,736][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:42:26,233][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:42:26,730][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:42:27,226][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:42:27,723][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:42:28,220][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:42:28,716][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:42:29,213][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:42:29,709][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:42:30,206][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:42:30,724][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:42:31,224][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:42:31,721][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:42:32,219][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:42:32,715][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:42:33,211][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:42:33,709][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:42:34,205][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:42:34,701][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:42:35,197][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:42:35,694][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:42:36,191][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:42:36,688][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:42:37,186][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:42:37,685][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:42:38,185][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:42:38,683][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:42:39,182][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:42:39,678][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:42:40,178][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:42:40,676][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:42:41,171][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:42:41,668][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:42:42,165][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:42:42,660][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:42:43,158][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:42:43,656][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:42:44,154][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10112 tokens. +[2026-03-25 23:42:44,808][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 61.90%, ΔTime: 00:00:32 +[2026-03-25 23:42:45,564][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:42:45,566][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:42:45,568][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:42:46,234][__main__][INFO] - Iteration 118 took 50s (30.44% Gen, 68.24% Train). Generation: 15s, Training: 34s. Estimated remaining time: 40h 41m 49s. Estimated total time: 42h 23m 47s. Time estimates for 10 more iterations: 8m 28s, 100 more iterations: 1h 24m 47s, 500 more iterations: 7h 3m 57s. +[2026-03-25 23:42:46,236][__main__][INFO] - Starting iteration 118. +[2026-03-25 23:42:46,635][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 11 and human policies 1. +[2026-03-25 23:42:46,635][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:42:49,319][mllm.models.large_language_model_local][WARNING] - Response Proposal: 5 hats, 5 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:43:02,675][__main__][INFO] - Number of regex retries in iteration 118: 1 +[2026-03-25 23:43:02,676][__main__][INFO] - agents played in iteration 118 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:43:03,448][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:43:03,471][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:43:03,494][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:43:03,518][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:43:03,519][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:43:03,520][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:43:04,143][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:43:04,596][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:43:05,099][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:43:05,599][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:43:06,096][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:43:06,595][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:43:07,095][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:43:07,591][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:43:08,088][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:43:08,587][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:43:09,087][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:43:09,586][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:43:10,084][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:43:10,582][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:43:11,079][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:43:11,585][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:43:12,085][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:43:12,583][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:43:13,084][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:43:13,588][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:43:14,089][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:43:14,596][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:43:15,092][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:43:15,590][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:43:16,087][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:43:16,583][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:43:17,078][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:43:17,573][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:43:18,072][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:43:18,569][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:43:19,067][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:43:19,562][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:43:20,060][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:43:20,558][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:43:21,055][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:43:21,553][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:43:22,050][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:43:22,547][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:43:23,044][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:43:23,540][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:43:24,038][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:43:24,535][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:43:25,036][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:43:25,532][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:43:26,032][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:43:26,528][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:43:27,027][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:43:27,528][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:43:28,027][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:43:28,521][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:43:29,018][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:43:29,515][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:43:30,011][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:43:30,506][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:43:31,003][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:43:31,499][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:43:31,996][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:43:32,492][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:43:32,987][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:43:33,483][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:43:33,980][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:43:34,478][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:43:34,976][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:43:35,471][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:43:35,966][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10054 tokens. +[2026-03-25 23:43:36,644][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.99%, Current % of VRAM taken: 60.42%, Block Peak % of device VRAM: 62.00%, ΔTime: 00:00:32 +[2026-03-25 23:43:37,411][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:43:37,413][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:43:37,415][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:43:38,087][__main__][INFO] - Iteration 119 took 51s (31.18% Gen, 67.52% Train). Generation: 16s, Training: 34s. Estimated remaining time: 41h 9m 47s. Estimated total time: 42h 52m 37s. Time estimates for 10 more iterations: 8m 34s, 100 more iterations: 1h 25m 45s, 500 more iterations: 7h 8m 46s. +[2026-03-25 23:43:38,089][__main__][INFO] - Starting iteration 119. +[2026-03-25 23:43:38,490][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 11 and human policies 1. +[2026-03-25 23:43:38,490][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:43:54,547][__main__][INFO] - Number of regex retries in iteration 119: 0 +[2026-03-25 23:43:54,548][__main__][INFO] - agents played in iteration 119 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:43:55,321][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:43:55,344][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:43:55,367][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:43:55,389][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:43:55,390][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:43:55,391][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:43:56,033][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:43:56,486][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:43:56,987][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:43:57,483][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:43:57,979][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:43:58,478][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:43:58,975][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:43:59,473][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:43:59,970][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:44:00,467][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:44:00,964][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:44:01,460][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:44:01,960][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:44:02,457][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:44:02,954][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:44:03,450][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:44:03,945][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:44:04,441][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:44:04,946][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:44:05,449][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:44:05,954][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:44:06,453][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:44:06,950][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:44:07,447][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:44:07,942][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:44:08,439][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:44:08,936][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:44:09,431][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:44:09,928][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:44:10,424][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:44:10,919][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:44:11,416][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:44:11,912][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:44:12,406][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:44:12,901][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:44:13,397][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:44:13,892][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:44:14,386][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:44:14,883][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:44:15,379][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:44:15,877][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:44:16,373][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:44:16,867][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:44:17,362][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:44:17,859][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:44:18,354][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:44:18,848][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:44:19,342][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:44:19,837][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:44:20,337][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:44:20,836][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:44:21,333][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:44:21,830][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:44:22,327][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:44:22,824][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:44:23,320][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:44:23,816][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:44:24,312][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:44:24,808][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:44:25,304][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:44:25,801][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:44:26,297][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:44:26,794][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:44:27,289][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:44:27,785][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10131 tokens. +[2026-03-25 23:44:28,456][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 61.86%, ΔTime: 00:00:32 +[2026-03-25 23:44:29,226][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:44:29,228][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:44:29,230][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:44:29,939][__main__][INFO] - Iteration 120 took 51s (31.21% Gen, 67.41% Train). Generation: 16s, Training: 34s. Estimated remaining time: 41h 8m 49s. Estimated total time: 42h 52m 30s. Time estimates for 10 more iterations: 8m 34s, 100 more iterations: 1h 25m 45s, 500 more iterations: 7h 8m 45s. +[2026-03-25 23:44:29,941][__main__][INFO] - Starting iteration 120. +[2026-03-25 23:44:30,343][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 11 and human policies 1. +[2026-03-25 23:44:30,343][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:44:32,853][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:44:45,193][__main__][INFO] - Number of regex retries in iteration 120: 1 +[2026-03-25 23:44:45,194][__main__][INFO] - agents played in iteration 120 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:44:45,972][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:44:45,995][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:44:46,018][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:44:46,040][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:44:46,041][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:44:46,042][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:44:46,661][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:44:47,113][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:44:47,615][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:44:48,115][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:44:48,614][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:44:49,111][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:44:49,610][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:44:50,112][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:44:50,610][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:44:51,107][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:44:51,605][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:44:52,102][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:44:52,601][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:44:53,100][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:44:53,598][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:44:54,096][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:44:54,594][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:44:55,094][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:44:55,600][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:44:56,103][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:44:56,603][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:44:57,098][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:44:57,593][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:44:58,090][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:44:58,587][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:44:59,102][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:44:59,604][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:45:00,103][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:45:00,603][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:45:01,104][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:45:01,603][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:45:02,101][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:45:02,601][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:45:03,101][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:45:03,601][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:45:04,100][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:45:04,600][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:45:05,099][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:45:05,598][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:45:06,098][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:45:06,597][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:45:07,094][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:45:07,591][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:45:08,088][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:45:08,586][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:45:09,083][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:45:09,581][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:45:10,079][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:45:10,580][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:45:11,077][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:45:11,574][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:45:12,068][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:45:12,564][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:45:13,063][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:45:13,559][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:45:14,055][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:45:14,550][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:45:15,049][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:45:15,548][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:45:16,044][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:45:16,540][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:45:17,036][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:45:17,532][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:45:18,033][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:45:18,533][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10158 tokens. +[2026-03-25 23:45:19,211][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.61%, Block Peak % of device VRAM: 62.05%, ΔTime: 00:00:32 +[2026-03-25 23:45:19,992][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:45:19,994][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:45:19,995][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:45:21,413][__main__][INFO] - Iteration 121 took 51s (29.08% Gen, 68.14% Train). Generation: 14s, Training: 34s. Estimated remaining time: 40h 48m 59s. Estimated total time: 42h 33m 32s. Time estimates for 10 more iterations: 8m 30s, 100 more iterations: 1h 25m 7s, 500 more iterations: 7h 5m 35s. +[2026-03-25 23:45:21,415][__main__][INFO] - Starting iteration 121. +[2026-03-25 23:45:21,815][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 12 and human policies 1. +[2026-03-25 23:45:21,816][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:45:38,388][__main__][INFO] - Number of regex retries in iteration 121: 0 +[2026-03-25 23:45:38,389][__main__][INFO] - agents played in iteration 121 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:45:39,163][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:45:39,186][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:45:39,209][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:45:39,232][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:45:39,233][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:45:39,233][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:45:39,867][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:45:40,324][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:45:40,825][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:45:41,322][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:45:41,819][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:45:42,319][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:45:42,816][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:45:43,314][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:45:43,812][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:45:44,311][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:45:44,810][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:45:45,312][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:45:45,820][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:45:46,321][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:45:46,822][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:45:47,319][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:45:47,816][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:45:48,315][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:45:48,817][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:45:49,315][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:45:49,814][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:45:50,317][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:45:50,818][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:45:51,315][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:45:51,812][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:45:52,310][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:45:52,807][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:45:53,306][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:45:53,807][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:45:54,304][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:45:54,801][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:45:55,298][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:45:55,795][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:45:56,293][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:45:56,792][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:45:57,289][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:45:57,784][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:45:58,282][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:45:58,778][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:45:59,274][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:45:59,770][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:46:00,266][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:46:00,761][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:46:01,259][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:46:01,758][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:46:02,254][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:46:02,750][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:46:03,245][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:46:03,741][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:46:04,240][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:46:04,739][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:46:05,238][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:46:05,738][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:46:06,236][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:46:06,731][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:46:07,233][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:46:07,730][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:46:08,229][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:46:08,728][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:46:09,227][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:46:09,726][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:46:10,221][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:46:10,719][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:46:11,218][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:46:11,717][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10143 tokens. +[2026-03-25 23:46:12,380][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 61.99%, ΔTime: 00:00:32 +[2026-03-25 23:46:13,145][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:46:13,147][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:46:13,149][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:46:13,827][__main__][INFO] - Iteration 122 took 52s (31.86% Gen, 66.83% Train). Generation: 16s, Training: 34s. Estimated remaining time: 41h 35m 12s. Estimated total time: 43h 20m 37s. Time estimates for 10 more iterations: 8m 40s, 100 more iterations: 1h 26m 41s, 500 more iterations: 7h 13m 26s. +[2026-03-25 23:46:13,829][__main__][INFO] - Starting iteration 122. +[2026-03-25 23:46:14,229][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 12 and human policies 1. +[2026-03-25 23:46:14,229][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:46:22,960][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:46:30,216][__main__][INFO] - Number of regex retries in iteration 122: 1 +[2026-03-25 23:46:30,217][__main__][INFO] - agents played in iteration 122 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:46:30,988][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:46:31,012][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:46:31,035][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:46:31,058][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:46:31,058][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:46:31,059][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:46:31,687][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:46:32,138][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:46:32,639][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:46:33,138][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:46:33,634][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:46:34,130][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:46:34,626][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:46:35,127][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:46:35,631][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:46:36,132][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:46:36,634][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:46:37,137][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:46:37,641][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:46:38,139][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:46:38,637][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:46:39,134][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:46:39,631][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:46:40,128][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:46:40,626][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:46:41,126][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:46:41,623][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:46:42,119][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:46:42,619][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:46:43,115][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:46:43,611][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:46:44,106][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:46:44,605][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:46:45,101][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:46:45,601][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:46:46,098][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:46:46,596][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:46:47,093][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:46:47,592][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:46:48,089][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:46:48,587][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:46:49,086][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:46:49,583][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:46:50,102][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:46:50,603][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:46:51,099][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:46:51,603][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:46:52,099][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:46:52,600][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:46:53,097][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:46:53,595][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:46:54,095][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:46:54,591][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:46:55,091][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:46:55,588][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:46:56,087][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:46:56,583][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:46:57,081][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:46:57,581][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:46:58,078][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:46:58,578][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:46:59,073][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:46:59,570][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:47:00,068][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:47:00,566][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:47:01,063][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:47:01,561][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:47:02,061][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:47:02,560][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:47:03,060][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:47:03,557][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10155 tokens. +[2026-03-25 23:47:04,217][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.22%, ΔTime: 00:00:32 +[2026-03-25 23:47:04,981][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:47:04,983][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:47:04,985][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:47:05,710][__main__][INFO] - Iteration 123 took 51s (31.05% Gen, 67.53% Train). Generation: 15s, Training: 34s. Estimated remaining time: 41h 7m 50s. Estimated total time: 42h 54m 7s. Time estimates for 10 more iterations: 8m 34s, 100 more iterations: 1h 25m 48s, 500 more iterations: 7h 9m 1s. +[2026-03-25 23:47:05,713][__main__][INFO] - Starting iteration 123. +[2026-03-25 23:47:06,113][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 12 and human policies 1. +[2026-03-25 23:47:06,114][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:47:22,473][__main__][INFO] - Number of regex retries in iteration 123: 0 +[2026-03-25 23:47:22,474][__main__][INFO] - agents played in iteration 123 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:47:23,254][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:47:23,277][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:47:23,300][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:47:23,323][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:47:23,323][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:47:23,324][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:47:23,953][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:47:24,412][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:47:24,918][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:47:25,416][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:47:25,915][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:47:26,419][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:47:26,921][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:47:27,425][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:47:27,928][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:47:28,432][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:47:28,933][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:47:29,434][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:47:29,935][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:47:30,437][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:47:30,933][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:47:31,430][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:47:31,928][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:47:32,426][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:47:32,924][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:47:33,424][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:47:33,923][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:47:34,423][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:47:34,921][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:47:35,421][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:47:35,922][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:47:36,421][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:47:36,919][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:47:37,417][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:47:37,916][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:47:38,416][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:47:38,914][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:47:39,411][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:47:39,910][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:47:40,410][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:47:40,910][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:47:41,410][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:47:41,909][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:47:42,411][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:47:42,912][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:47:43,417][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:47:43,917][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:47:44,417][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:47:44,914][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:47:45,412][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:47:45,909][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:47:46,406][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:47:46,906][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:47:47,404][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:47:47,905][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:47:48,403][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:47:48,901][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:47:49,396][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:47:49,892][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:47:50,391][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:47:50,886][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:47:51,382][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:47:51,880][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:47:52,377][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:47:52,875][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:47:53,371][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:47:53,869][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:47:54,366][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:47:54,862][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:47:55,364][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:47:55,861][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10190 tokens. +[2026-03-25 23:47:56,527][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.15%, Current % of VRAM taken: 60.58%, Block Peak % of device VRAM: 62.28%, ΔTime: 00:00:32 +[2026-03-25 23:47:57,293][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:47:57,295][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:47:57,297][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:47:57,976][__main__][INFO] - Iteration 124 took 51s (31.55% Gen, 67.14% Train). Generation: 16s, Training: 34s. Estimated remaining time: 41h 26m 1s. Estimated total time: 43h 13m 10s. Time estimates for 10 more iterations: 8m 38s, 100 more iterations: 1h 26m 26s, 500 more iterations: 7h 12m 11s. +[2026-03-25 23:47:57,978][__main__][INFO] - Starting iteration 124. +[2026-03-25 23:47:58,377][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 12 and human policies 1. +[2026-03-25 23:47:58,378][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:48:14,659][__main__][INFO] - Number of regex retries in iteration 124: 0 +[2026-03-25 23:48:14,660][__main__][INFO] - agents played in iteration 124 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:48:15,437][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:48:15,461][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:48:15,484][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:48:15,507][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:48:15,507][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:48:15,508][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:48:16,164][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:48:16,619][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:48:17,124][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:48:17,622][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:48:18,120][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:48:18,623][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:48:19,126][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:48:19,632][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:48:20,136][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:48:20,639][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:48:21,139][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:48:21,638][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:48:22,143][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:48:22,642][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:48:23,140][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:48:23,640][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:48:24,136][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:48:24,637][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:48:25,133][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:48:25,630][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:48:26,127][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:48:26,625][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:48:27,121][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:48:27,619][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:48:28,117][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:48:28,611][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:48:29,106][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:48:29,602][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:48:30,097][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:48:30,592][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:48:31,090][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:48:31,587][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:48:32,082][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:48:32,580][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:48:33,079][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:48:33,575][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:48:34,072][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:48:34,567][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:48:35,064][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:48:35,562][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:48:36,059][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:48:36,556][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:48:37,056][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:48:37,552][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:48:38,049][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:48:38,546][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:48:39,043][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:48:39,539][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:48:40,036][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:48:40,537][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:48:41,037][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:48:41,535][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:48:42,037][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:48:42,538][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:48:43,036][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:48:43,554][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:48:44,058][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:48:44,553][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:48:45,053][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:48:45,553][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:48:46,050][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:48:46,545][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:48:47,042][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:48:47,539][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:48:48,039][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10222 tokens. +[2026-03-25 23:48:48,712][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.17%, Current % of VRAM taken: 60.60%, Block Peak % of device VRAM: 62.13%, ΔTime: 00:00:32 +[2026-03-25 23:48:49,486][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:48:49,489][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:48:49,491][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:48:50,174][__main__][INFO] - Iteration 125 took 51s (31.43% Gen, 67.24% Train). Generation: 16s, Training: 34s. Estimated remaining time: 41h 21m 50s. Estimated total time: 43h 9m 51s. Time estimates for 10 more iterations: 8m 37s, 100 more iterations: 1h 26m 19s, 500 more iterations: 7h 11m 38s. +[2026-03-25 23:48:50,176][__main__][INFO] - Starting iteration 125. +[2026-03-25 23:48:50,575][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 12 and human policies 1. +[2026-03-25 23:48:50,576][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:48:54,688][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:48:59,164][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:49:06,801][__main__][INFO] - Number of regex retries in iteration 125: 2 +[2026-03-25 23:49:06,802][__main__][INFO] - agents played in iteration 125 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:49:07,584][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:49:07,607][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:49:07,630][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:49:07,653][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:49:07,654][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:49:07,655][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:49:08,289][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:49:08,742][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:49:09,244][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:49:09,741][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:49:10,239][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:49:10,740][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:49:11,245][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:49:11,749][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:49:12,253][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:49:12,749][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:49:13,247][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:49:13,744][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:49:14,245][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:49:14,746][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:49:15,247][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:49:15,746][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:49:16,245][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:49:16,740][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:49:17,237][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:49:17,734][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:49:18,233][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:49:18,729][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:49:19,224][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:49:19,722][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:49:20,221][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:49:20,718][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:49:21,215][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:49:21,711][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:49:22,212][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:49:22,707][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:49:23,203][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:49:23,699][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:49:24,195][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:49:24,692][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:49:25,188][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:49:25,685][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:49:26,183][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:49:26,680][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:49:27,179][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:49:27,678][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:49:28,174][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:49:28,669][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:49:29,164][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:49:29,660][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:49:30,157][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:49:30,656][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:49:31,152][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:49:31,647][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:49:32,144][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:49:32,643][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:49:33,142][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:49:33,641][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:49:34,144][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:49:34,644][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:49:35,142][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:49:35,641][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:49:36,141][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:49:36,636][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:49:37,132][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:49:37,628][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:49:38,124][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:49:38,621][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:49:39,117][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:49:39,614][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:49:40,110][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10173 tokens. +[2026-03-25 23:49:40,779][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.05%, Current % of VRAM taken: 60.48%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-25 23:49:41,546][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:49:41,549][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:49:41,551][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:49:42,291][__main__][INFO] - Iteration 126 took 51s (31.37% Gen, 67.19% Train). Generation: 16s, Training: 34s. Estimated remaining time: 41h 16m 57s. Estimated total time: 43h 5m 51s. Time estimates for 10 more iterations: 8m 37s, 100 more iterations: 1h 26m 11s, 500 more iterations: 7h 10m 58s. +[2026-03-25 23:49:42,294][__main__][INFO] - Starting iteration 126. +[2026-03-25 23:49:42,697][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 12 and human policies 1. +[2026-03-25 23:49:42,697][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:49:53,766][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:49:59,096][__main__][INFO] - Number of regex retries in iteration 126: 1 +[2026-03-25 23:49:59,097][__main__][INFO] - agents played in iteration 126 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:49:59,900][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:49:59,924][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:49:59,947][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:49:59,970][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:49:59,970][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:49:59,971][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:50:00,614][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:50:01,065][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:50:01,572][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:50:02,076][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:50:02,583][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:50:03,089][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:50:03,591][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:50:04,089][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:50:04,592][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:50:05,088][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:50:05,590][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:50:06,087][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:50:06,585][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:50:07,082][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:50:07,578][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:50:08,073][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:50:08,569][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:50:09,069][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:50:09,565][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:50:10,065][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:50:10,564][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:50:11,060][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:50:11,557][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:50:12,054][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:50:12,550][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:50:13,045][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:50:13,540][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:50:14,036][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:50:14,531][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:50:15,027][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:50:15,522][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:50:16,019][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:50:16,516][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:50:17,014][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:50:17,513][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:50:18,008][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:50:18,505][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:50:19,001][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:50:19,497][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:50:19,994][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:50:20,497][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:50:20,996][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:50:21,491][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:50:21,988][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:50:22,485][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:50:22,982][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:50:23,479][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:50:23,978][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:50:24,476][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:50:24,972][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:50:25,469][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:50:25,969][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:50:26,467][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:50:26,963][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:50:27,459][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:50:27,954][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:50:28,450][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:50:28,946][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:50:29,444][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:50:29,942][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:50:30,440][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:50:30,938][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:50:31,435][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:50:31,932][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:50:32,430][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10114 tokens. +[2026-03-25 23:50:33,101][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.12%, ΔTime: 00:00:32 +[2026-03-25 23:50:33,872][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:50:33,874][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:50:33,876][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:50:34,595][__main__][INFO] - Iteration 127 took 51s (31.60% Gen, 67.01% Train). Generation: 16s, Training: 34s. Estimated remaining time: 41h 25m 12s. Estimated total time: 43h 14m 58s. Time estimates for 10 more iterations: 8m 38s, 100 more iterations: 1h 26m 29s, 500 more iterations: 7h 12m 29s. +[2026-03-25 23:50:34,598][__main__][INFO] - Starting iteration 127. +[2026-03-25 23:50:35,004][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 12 and human policies 1. +[2026-03-25 23:50:35,005][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:50:50,677][__main__][INFO] - Number of regex retries in iteration 127: 0 +[2026-03-25 23:50:50,678][__main__][INFO] - agents played in iteration 127 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:50:51,460][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:50:51,483][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:50:51,506][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:50:51,529][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:50:51,530][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:50:51,530][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:50:52,166][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:50:52,624][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:50:53,129][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:50:53,628][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:50:54,124][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:50:54,623][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:50:55,123][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:50:55,621][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:50:56,117][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:50:56,616][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:50:57,117][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:50:57,616][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:50:58,114][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:50:58,612][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:50:59,110][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:50:59,607][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:51:00,109][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:51:00,607][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:51:01,104][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:51:01,602][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:51:02,099][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:51:02,596][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:51:03,091][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:51:03,586][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:51:04,082][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:51:04,580][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:51:05,076][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:51:05,572][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:51:06,069][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:51:06,565][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:51:07,062][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:51:07,559][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:51:08,056][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:51:08,554][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:51:09,050][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:51:09,549][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:51:10,046][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:51:10,542][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:51:11,040][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:51:11,537][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:51:12,034][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:51:12,532][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:51:13,028][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:51:13,525][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:51:14,022][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:51:14,519][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:51:15,016][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:51:15,513][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:51:16,032][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:51:16,529][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:51:17,025][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:51:17,524][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:51:18,024][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:51:18,521][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:51:19,020][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:51:19,519][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:51:20,025][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:51:20,526][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:51:21,024][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:51:21,522][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:51:22,019][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:51:22,518][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:51:23,018][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:51:23,517][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:51:24,016][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10103 tokens. +[2026-03-25 23:51:24,711][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.16%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.11%, ΔTime: 00:00:32 +[2026-03-25 23:51:25,486][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:51:25,488][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:51:25,490][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:51:26,174][__main__][INFO] - Iteration 128 took 51s (30.63% Gen, 68.03% Train). Generation: 15s, Training: 34s. Estimated remaining time: 40h 47m 52s. Estimated total time: 42h 38m 29s. Time estimates for 10 more iterations: 8m 31s, 100 more iterations: 1h 25m 16s, 500 more iterations: 7h 6m 24s. +[2026-03-25 23:51:26,176][__main__][INFO] - Starting iteration 128. +[2026-03-25 23:51:26,578][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 12 and human policies 1. +[2026-03-25 23:51:26,579][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:51:42,742][__main__][INFO] - Number of regex retries in iteration 128: 0 +[2026-03-25 23:51:42,743][__main__][INFO] - agents played in iteration 128 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:51:43,514][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:51:43,538][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:51:43,562][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:51:43,585][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:51:43,586][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:51:43,586][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:51:44,224][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:51:44,682][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:51:45,194][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:51:45,695][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:51:46,198][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:51:46,698][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:51:47,196][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:51:47,693][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:51:48,194][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:51:48,691][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:51:49,193][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:51:49,695][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:51:50,199][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:51:50,695][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:51:51,195][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:51:51,699][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:51:52,201][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:51:52,698][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:51:53,196][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:51:53,693][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:51:54,190][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:51:54,687][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:51:55,185][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:51:55,682][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:51:56,180][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:51:56,678][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:51:57,179][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:51:57,676][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:51:58,173][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:51:58,671][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:51:59,167][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:51:59,664][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:52:00,161][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:52:00,658][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:52:01,157][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:52:01,656][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:52:02,152][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:52:02,649][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:52:03,144][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:52:03,640][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:52:04,136][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:52:04,634][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:52:05,132][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:52:05,630][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:52:06,130][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:52:06,628][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:52:07,127][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:52:07,623][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:52:08,122][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:52:08,620][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:52:09,116][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:52:09,612][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:52:10,109][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:52:10,605][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:52:11,102][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:52:11,599][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:52:12,097][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:52:12,593][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:52:13,089][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:52:13,586][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:52:14,081][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:52:14,577][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:52:15,071][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:52:15,566][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:52:16,061][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10130 tokens. +[2026-03-25 23:52:16,740][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.13%, ΔTime: 00:00:32 +[2026-03-25 23:52:17,513][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:52:17,516][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:52:17,518][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:52:18,200][__main__][INFO] - Iteration 129 took 51s (31.31% Gen, 67.36% Train). Generation: 16s, Training: 34s. Estimated remaining time: 41h 9m 38s. Estimated total time: 43h 1m 8s. Time estimates for 10 more iterations: 8m 36s, 100 more iterations: 1h 26m 2s, 500 more iterations: 7h 10m 11s. +[2026-03-25 23:52:18,202][__main__][INFO] - Starting iteration 129. +[2026-03-25 23:52:18,603][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 12 and human policies 1. +[2026-03-25 23:52:18,603][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:52:30,325][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls + +Given the per-item values, hats are the most valuable for me, followed by books, and balls are the least valuable. Alice values hats the least and balls the most. By claiming all the hats, I maximize my value in this round, as I know Alice will not be bidding for them anyway. Claiming all the hats ensures I get the full benefit of their value, which is 10 points per hat, for a total of 100 points from the hats. I will not bid for books or balls since Alice values them more, and there's no strategic benefit in splitting them. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:52:36,018][__main__][INFO] - Number of regex retries in iteration 129: 1 +[2026-03-25 23:52:36,019][__main__][INFO] - agents played in iteration 129 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:52:36,793][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:52:36,816][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:52:36,840][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:52:36,863][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:52:36,863][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:52:36,864][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:52:37,519][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:52:37,971][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:52:38,475][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:52:38,974][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:52:39,471][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:52:39,970][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:52:40,481][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:52:40,985][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:52:41,486][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:52:41,984][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:52:42,481][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:52:42,978][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:52:43,478][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:52:43,979][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:52:44,478][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:52:44,976][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:52:45,472][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:52:45,969][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:52:46,466][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:52:46,964][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:52:47,461][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:52:47,976][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:52:48,480][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:52:48,977][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:52:49,474][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:52:49,970][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:52:50,471][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:52:50,968][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:52:51,463][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:52:51,962][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:52:52,463][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:52:52,963][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:52:53,460][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:52:53,958][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:52:54,454][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:52:54,951][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:52:55,450][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:52:55,946][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:52:56,445][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:52:56,943][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:52:57,440][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:52:57,937][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:52:58,434][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:52:58,931][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:52:59,432][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:52:59,929][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:53:00,427][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:53:00,926][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:53:01,426][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:53:01,923][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:53:02,421][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:53:02,918][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:53:03,415][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:53:03,911][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:53:04,407][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:53:04,903][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:53:05,400][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:53:05,898][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:53:06,398][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:53:06,896][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:53:07,396][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:53:07,892][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:53:08,394][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:53:08,891][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:53:09,387][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10186 tokens. +[2026-03-25 23:53:10,027][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.13%, ΔTime: 00:00:32 +[2026-03-25 23:53:10,799][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:53:10,801][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:53:10,803][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:53:11,487][__main__][INFO] - Iteration 130 took 52s (32.93% Gen, 65.77% Train). Generation: 17s, Training: 34s. Estimated remaining time: 42h 11m 51s. Estimated total time: 44h 4m 14s. Time estimates for 10 more iterations: 8m 48s, 100 more iterations: 1h 28m 8s, 500 more iterations: 7h 20m 42s. +[2026-03-25 23:53:11,489][__main__][INFO] - Starting iteration 130. +[2026-03-25 23:53:11,889][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 12 and human policies 1. +[2026-03-25 23:53:11,890][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:53:14,936][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:53:19,276][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:53:24,260][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:53:24,320][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:53:28,068][__main__][INFO] - Number of regex retries in iteration 130: 4 +[2026-03-25 23:53:28,069][__main__][INFO] - agents played in iteration 130 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:53:28,841][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:53:28,865][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:53:28,888][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:53:28,911][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:53:28,911][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:53:28,912][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:53:29,525][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:53:29,984][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:53:30,496][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:53:31,000][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:53:31,500][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:53:31,999][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:53:32,499][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:53:32,997][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:53:33,497][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:53:33,998][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:53:34,500][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:53:34,996][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:53:35,495][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:53:35,991][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:53:36,487][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:53:36,987][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:53:37,484][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:53:37,982][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:53:38,479][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:53:38,975][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:53:39,473][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:53:39,970][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:53:40,469][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:53:40,966][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:53:41,461][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:53:41,960][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:53:42,456][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:53:42,952][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:53:43,448][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:53:43,944][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:53:44,441][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:53:44,937][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:53:45,432][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:53:45,931][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:53:46,429][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:53:46,929][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:53:47,430][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:53:47,935][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:53:48,434][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:53:48,935][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:53:49,435][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:53:49,931][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:53:50,431][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:53:50,926][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:53:51,422][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:53:51,919][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:53:52,415][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:53:52,911][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:53:53,407][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:53:53,904][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:53:54,401][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:53:54,899][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:53:55,397][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:53:55,895][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:53:56,392][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:53:56,890][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:53:57,387][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:53:57,884][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:53:58,381][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:53:58,878][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:53:59,377][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:53:59,876][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:54:00,375][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:54:00,874][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:54:01,370][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10234 tokens. +[2026-03-25 23:54:02,021][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-25 23:54:02,783][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:54:02,786][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:54:02,787][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:54:04,015][__main__][INFO] - Iteration 131 took 52s (31.04% Gen, 66.60% Train). Generation: 16s, Training: 34s. Estimated remaining time: 41h 33m 3s. Estimated total time: 43h 26m 19s. Time estimates for 10 more iterations: 8m 41s, 100 more iterations: 1h 26m 52s, 500 more iterations: 7h 14m 23s. +[2026-03-25 23:54:04,018][__main__][INFO] - Starting iteration 131. +[2026-03-25 23:54:04,419][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 13 and human policies 1. +[2026-03-25 23:54:04,420][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:54:07,271][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:54:07,560][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:54:20,599][__main__][INFO] - Number of regex retries in iteration 131: 2 +[2026-03-25 23:54:20,600][__main__][INFO] - agents played in iteration 131 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:54:21,374][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:54:21,398][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:54:21,420][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:54:21,443][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:54:21,444][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:54:21,444][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:54:22,053][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:54:22,505][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:54:23,006][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:54:23,503][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:54:24,000][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:54:24,496][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:54:24,997][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:54:25,494][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:54:25,993][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:54:26,498][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:54:26,994][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:54:27,490][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:54:27,987][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:54:28,487][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:54:28,987][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:54:29,486][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:54:29,986][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:54:30,483][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:54:30,982][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:54:31,480][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:54:31,978][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:54:32,474][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:54:32,970][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:54:33,466][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:54:33,960][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:54:34,456][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:54:34,952][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:54:35,447][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:54:35,943][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:54:36,437][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:54:36,934][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:54:37,430][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:54:37,924][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:54:38,423][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:54:38,920][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:54:39,416][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:54:39,913][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:54:40,413][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:54:40,909][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:54:41,406][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:54:41,902][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:54:42,399][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:54:42,895][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:54:43,391][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:54:43,887][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:54:44,383][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:54:44,879][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:54:45,374][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:54:45,870][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:54:46,365][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:54:46,861][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:54:47,360][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:54:47,855][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:54:48,351][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:54:48,847][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:54:49,346][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:54:49,841][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:54:50,339][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:54:50,837][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:54:51,333][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:54:51,828][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:54:52,325][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:54:52,822][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:54:53,320][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:54:53,815][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10256 tokens. +[2026-03-25 23:54:54,461][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.11%, ΔTime: 00:00:32 +[2026-03-25 23:54:55,222][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:54:55,224][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:54:55,226][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:54:55,897][__main__][INFO] - Iteration 132 took 51s (31.43% Gen, 67.26% Train). Generation: 16s, Training: 34s. Estimated remaining time: 40h 59m 48s. Estimated total time: 42h 53m 55s. Time estimates for 10 more iterations: 8m 34s, 100 more iterations: 1h 25m 47s, 500 more iterations: 7h 8m 59s. +[2026-03-25 23:54:55,899][__main__][INFO] - Starting iteration 132. +[2026-03-25 23:54:56,299][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 13 and human policies 1. +[2026-03-25 23:54:56,300][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:55:00,612][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 books, 0 hats, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:55:12,876][__main__][INFO] - Number of regex retries in iteration 132: 1 +[2026-03-25 23:55:12,877][__main__][INFO] - agents played in iteration 132 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:55:13,648][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:55:13,671][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:55:13,694][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:55:13,717][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:55:13,717][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:55:13,718][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:55:14,323][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:55:14,777][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:55:15,284][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:55:15,781][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:55:16,280][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:55:16,778][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:55:17,276][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:55:17,775][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:55:18,280][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:55:18,784][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:55:19,283][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:55:19,779][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:55:20,280][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:55:20,778][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:55:21,279][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:55:21,779][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:55:22,278][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:55:22,774][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:55:23,271][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:55:23,768][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:55:24,264][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:55:24,760][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:55:25,257][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:55:25,752][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:55:26,248][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:55:26,747][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:55:27,246][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:55:27,745][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:55:28,244][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:55:28,742][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:55:29,247][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:55:29,745][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:55:30,244][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:55:30,740][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:55:31,236][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:55:31,732][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:55:32,230][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:55:32,725][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:55:33,221][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:55:33,718][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:55:34,214][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:55:34,713][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:55:35,208][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:55:35,703][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:55:36,202][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:55:36,700][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:55:37,200][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:55:37,696][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:55:38,194][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:55:38,689][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:55:39,184][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:55:39,680][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:55:40,177][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:55:40,672][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:55:41,167][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:55:41,663][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:55:42,160][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:55:42,655][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:55:43,149][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:55:43,644][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:55:44,137][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:55:44,633][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:55:45,128][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:55:45,622][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:55:46,116][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10185 tokens. +[2026-03-25 23:55:46,771][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.03%, Current % of VRAM taken: 60.45%, Block Peak % of device VRAM: 62.32%, ΔTime: 00:00:32 +[2026-03-25 23:55:47,535][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:55:47,537][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:55:47,539][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:55:48,239][__main__][INFO] - Iteration 133 took 51s (31.92% Gen, 66.73% Train). Generation: 16s, Training: 34s. Estimated remaining time: 41h 22m 0s. Estimated total time: 43h 17m 0s. Time estimates for 10 more iterations: 8m 39s, 100 more iterations: 1h 26m 34s, 500 more iterations: 7h 12m 50s. +[2026-03-25 23:55:48,241][__main__][INFO] - Starting iteration 133. +[2026-03-25 23:55:48,643][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 13 and human policies 1. +[2026-03-25 23:55:48,643][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:56:05,010][__main__][INFO] - Number of regex retries in iteration 133: 0 +[2026-03-25 23:56:05,011][__main__][INFO] - agents played in iteration 133 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:56:05,784][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:56:05,808][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:56:05,830][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:56:05,853][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:56:05,854][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:56:05,854][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:56:06,481][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:56:06,937][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:56:07,441][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:56:07,944][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:56:08,443][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:56:08,943][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:56:09,442][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:56:09,940][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:56:10,440][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:56:10,944][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:56:11,443][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:56:11,941][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:56:12,443][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:56:12,940][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:56:13,441][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:56:13,941][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:56:14,439][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:56:14,939][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:56:15,439][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:56:15,936][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:56:16,436][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:56:16,933][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:56:17,434][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:56:17,930][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:56:18,427][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:56:18,923][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:56:19,420][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:56:19,917][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:56:20,416][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:56:20,912][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:56:21,407][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:56:21,903][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:56:22,401][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:56:22,901][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:56:23,396][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:56:23,892][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:56:24,389][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:56:24,885][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:56:25,384][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:56:25,882][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:56:26,379][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:56:26,877][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:56:27,376][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:56:27,873][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:56:28,370][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:56:28,867][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:56:29,363][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:56:29,860][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:56:30,357][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:56:30,857][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:56:31,356][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:56:31,852][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:56:32,349][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:56:32,846][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:56:33,342][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:56:33,839][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:56:34,339][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:56:34,837][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:56:35,335][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:56:35,832][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:56:36,328][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:56:36,825][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:56:37,322][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:56:37,820][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:56:38,317][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10187 tokens. +[2026-03-25 23:56:38,979][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.14%, ΔTime: 00:00:32 +[2026-03-25 23:56:39,728][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:56:39,730][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:56:39,732][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:56:40,406][__main__][INFO] - Iteration 134 took 51s (31.62% Gen, 67.08% Train). Generation: 16s, Training: 34s. Estimated remaining time: 41h 12m 19s. Estimated total time: 43h 8m 11s. Time estimates for 10 more iterations: 8m 37s, 100 more iterations: 1h 26m 16s, 500 more iterations: 7h 11m 21s. +[2026-03-25 23:56:40,408][__main__][INFO] - Starting iteration 134. +[2026-03-25 23:56:40,817][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 13 and human policies 1. +[2026-03-25 23:56:40,818][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:56:56,498][__main__][INFO] - Number of regex retries in iteration 134: 0 +[2026-03-25 23:56:56,499][__main__][INFO] - agents played in iteration 134 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:56:57,269][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:56:57,292][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:56:57,315][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:56:57,338][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:56:57,338][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:56:57,339][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:56:57,955][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:56:58,410][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:56:58,911][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:56:59,408][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:56:59,904][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:57:00,402][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:57:00,900][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:57:01,399][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:57:01,896][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:57:02,398][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:57:02,896][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:57:03,394][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:57:03,891][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:57:04,389][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:57:04,887][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:57:05,385][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:57:05,885][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:57:06,384][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:57:06,883][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:57:07,381][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:57:07,880][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:57:08,379][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:57:08,877][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:57:09,377][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:57:09,875][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:57:10,377][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:57:10,880][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:57:11,380][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:57:11,880][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:57:12,378][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:57:12,879][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:57:13,377][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:57:13,874][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:57:14,370][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:57:14,865][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:57:15,361][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:57:15,858][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:57:16,354][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:57:16,851][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:57:17,348][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:57:17,843][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:57:18,343][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:57:18,839][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:57:19,337][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:57:19,836][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:57:20,335][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:57:20,836][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:57:21,336][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:57:21,833][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:57:22,329][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:57:22,826][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:57:23,325][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:57:23,822][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:57:24,319][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:57:24,815][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:57:25,311][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:57:25,807][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:57:26,304][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:57:26,804][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:57:27,302][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:57:27,800][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:57:28,298][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:57:28,799][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:57:29,296][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:57:29,793][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10238 tokens. +[2026-03-25 23:57:30,452][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.13%, ΔTime: 00:00:32 +[2026-03-25 23:57:31,227][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:57:31,229][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:57:31,231][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:57:31,924][__main__][INFO] - Iteration 135 took 51s (30.68% Gen, 67.96% Train). Generation: 15s, Training: 34s. Estimated remaining time: 40h 38m 39s. Estimated total time: 42h 35m 23s. Time estimates for 10 more iterations: 8m 31s, 100 more iterations: 1h 25m 10s, 500 more iterations: 7h 5m 53s. +[2026-03-25 23:57:31,927][__main__][INFO] - Starting iteration 135. +[2026-03-25 23:57:32,331][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 13 and human policies 1. +[2026-03-25 23:57:32,331][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:57:37,026][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:57:50,254][__main__][INFO] - Number of regex retries in iteration 135: 1 +[2026-03-25 23:57:50,255][__main__][INFO] - agents played in iteration 135 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:57:51,024][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:57:51,048][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:57:51,071][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:57:51,093][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:57:51,094][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:57:51,095][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:57:51,703][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:57:52,156][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:57:52,657][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:57:53,157][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:57:53,658][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:57:54,157][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:57:54,654][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:57:55,152][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:57:55,648][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:57:56,144][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:57:56,642][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:57:57,137][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:57:57,636][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:57:58,132][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:57:58,628][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:57:59,124][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:57:59,620][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:58:00,119][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:58:00,619][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:58:01,119][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:58:01,617][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:58:02,115][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:58:02,634][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:58:03,138][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:58:03,637][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:58:04,132][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:58:04,629][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:58:05,125][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:58:05,621][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:58:06,116][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:58:06,612][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:58:07,107][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:58:07,604][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:58:08,102][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:58:08,599][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:58:09,097][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:58:09,597][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:58:10,093][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:58:10,590][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:58:11,086][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:58:11,587][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:58:12,085][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:58:12,583][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:58:13,082][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:58:13,581][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:58:14,081][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:58:14,581][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:58:15,079][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:58:15,578][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:58:16,077][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:58:16,575][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:58:17,076][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:58:17,573][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:58:18,072][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:58:18,571][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:58:19,072][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:58:19,569][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:58:20,071][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:58:20,571][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:58:21,069][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:58:21,569][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:58:22,067][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:58:22,566][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:58:23,065][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:58:23,565][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10174 tokens. +[2026-03-25 23:58:24,225][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.15%, Current % of VRAM taken: 60.58%, Block Peak % of device VRAM: 62.10%, ΔTime: 00:00:32 +[2026-03-25 23:58:24,995][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:58:24,997][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:58:25,011][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:58:25,690][__main__][INFO] - Iteration 136 took 53s (33.59% Gen, 65.13% Train). Generation: 17s, Training: 34s. Estimated remaining time: 42h 30m 23s. Estimated total time: 44h 28m 0s. Time estimates for 10 more iterations: 8m 53s, 100 more iterations: 1h 28m 56s, 500 more iterations: 7h 24m 40s. +[2026-03-25 23:58:25,693][__main__][INFO] - Starting iteration 136. +[2026-03-25 23:58:26,096][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 13 and human policies 1. +[2026-03-25 23:58:26,096][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:58:29,254][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-25 23:58:42,703][__main__][INFO] - Number of regex retries in iteration 136: 1 +[2026-03-25 23:58:42,704][__main__][INFO] - agents played in iteration 136 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:58:43,476][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:58:43,499][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:58:43,522][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:58:43,544][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:58:43,545][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:58:43,546][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:58:44,154][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:58:44,606][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:58:45,108][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:58:45,605][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:58:46,103][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:58:46,601][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:58:47,100][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:58:47,600][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:58:48,098][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:58:48,595][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:58:49,090][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:58:49,585][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:58:50,085][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:58:50,581][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:58:51,079][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:58:51,577][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:58:52,073][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:58:52,573][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:58:53,069][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:58:53,565][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:58:54,062][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:58:54,560][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:58:55,059][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:58:55,556][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:58:56,053][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:58:56,551][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:58:57,050][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:58:57,545][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:58:58,041][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:58:58,540][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:58:59,037][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:58:59,532][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:59:00,032][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:59:00,529][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:59:01,025][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:59:01,524][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:59:02,023][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:59:02,520][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:59:03,016][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:59:03,515][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:59:04,010][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:59:04,506][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:59:05,004][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:59:05,500][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:59:05,997][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:59:06,493][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:59:06,989][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:59:07,486][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-25 23:59:07,982][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-25 23:59:08,482][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-25 23:59:08,979][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-25 23:59:09,479][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-25 23:59:09,978][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-25 23:59:10,475][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-25 23:59:10,974][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-25 23:59:11,471][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-25 23:59:11,968][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-25 23:59:12,466][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-25 23:59:12,965][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-25 23:59:13,465][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-25 23:59:13,963][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-25 23:59:14,462][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-25 23:59:14,960][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-25 23:59:15,459][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-25 23:59:15,958][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10176 tokens. +[2026-03-25 23:59:16,605][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.15%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.00%, ΔTime: 00:00:32 +[2026-03-25 23:59:17,372][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-25 23:59:17,374][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-25 23:59:17,376][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-25 23:59:18,119][__main__][INFO] - Iteration 137 took 52s (31.92% Gen, 66.65% Train). Generation: 16s, Training: 34s. Estimated remaining time: 41h 22m 41s. Estimated total time: 43h 21m 11s. Time estimates for 10 more iterations: 8m 40s, 100 more iterations: 1h 26m 42s, 500 more iterations: 7h 13m 31s. +[2026-03-25 23:59:18,121][__main__][INFO] - Starting iteration 137. +[2026-03-25 23:59:18,521][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 13 and human policies 1. +[2026-03-25 23:59:18,522][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-25 23:59:34,927][__main__][INFO] - Number of regex retries in iteration 137: 0 +[2026-03-25 23:59:34,928][__main__][INFO] - agents played in iteration 137 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-25 23:59:35,715][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:59:35,738][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:59:35,761][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:59:35,784][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-25 23:59:35,784][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-25 23:59:35,785][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-25 23:59:36,409][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-25 23:59:36,864][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-25 23:59:37,368][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-25 23:59:37,866][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-25 23:59:38,365][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-25 23:59:38,863][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-25 23:59:39,361][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-25 23:59:39,861][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-25 23:59:40,365][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-25 23:59:40,863][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-25 23:59:41,360][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-25 23:59:41,858][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-25 23:59:42,361][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-25 23:59:42,863][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-25 23:59:43,384][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-25 23:59:43,890][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-25 23:59:44,386][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-25 23:59:44,884][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-25 23:59:45,382][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-25 23:59:45,881][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-25 23:59:46,379][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-25 23:59:46,877][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-25 23:59:47,374][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-25 23:59:47,874][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-25 23:59:48,374][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-25 23:59:48,873][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-25 23:59:49,371][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-25 23:59:49,869][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-25 23:59:50,373][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-25 23:59:50,870][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-25 23:59:51,369][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-25 23:59:51,867][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-25 23:59:52,369][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-25 23:59:52,868][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-25 23:59:53,366][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-25 23:59:53,863][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-25 23:59:54,363][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-25 23:59:54,863][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-25 23:59:55,364][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-25 23:59:55,864][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-25 23:59:56,364][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-25 23:59:56,864][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-25 23:59:57,362][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-25 23:59:57,860][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-25 23:59:58,358][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-25 23:59:58,857][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-25 23:59:59,353][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-25 23:59:59,850][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:00:00,347][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:00:00,847][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:00:01,344][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:00:01,841][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:00:02,340][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:00:02,837][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:00:03,335][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:00:03,832][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:00:04,330][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:00:04,828][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:00:05,325][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:00:05,823][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:00:06,324][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:00:06,822][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:00:07,322][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:00:07,821][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:00:08,320][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10277 tokens. +[2026-03-26 00:00:08,959][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.15%, Current % of VRAM taken: 60.58%, Block Peak % of device VRAM: 62.12%, ΔTime: 00:00:32 +[2026-03-26 00:00:09,716][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:00:09,718][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:00:09,720][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:00:10,466][__main__][INFO] - Iteration 138 took 51s (31.58% Gen, 66.98% Train). Generation: 16s, Training: 34s. Estimated remaining time: 41h 17m 55s. Estimated total time: 43h 17m 17s. Time estimates for 10 more iterations: 8m 39s, 100 more iterations: 1h 26m 34s, 500 more iterations: 7h 12m 52s. +[2026-03-26 00:00:10,468][__main__][INFO] - Starting iteration 138. +[2026-03-26 00:00:10,868][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 13 and human policies 1. +[2026-03-26 00:00:10,869][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:00:14,947][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:00:17,497][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:00:22,273][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:00:28,094][__main__][INFO] - Number of regex retries in iteration 138: 3 +[2026-03-26 00:00:28,095][__main__][INFO] - agents played in iteration 138 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:00:28,881][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:00:28,904][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:00:28,927][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:00:28,949][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:00:28,950][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:00:28,951][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:00:29,562][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:00:30,020][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:00:30,529][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:00:31,031][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:00:31,533][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:00:32,034][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:00:32,535][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:00:33,037][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:00:33,537][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:00:34,036][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:00:34,534][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:00:35,031][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:00:35,529][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:00:36,028][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:00:36,527][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:00:37,029][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:00:37,526][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:00:38,026][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:00:38,525][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:00:39,023][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:00:39,520][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:00:40,022][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:00:40,521][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:00:41,017][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:00:41,514][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:00:42,013][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:00:42,509][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:00:43,005][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:00:43,505][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:00:44,006][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:00:44,505][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:00:45,002][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:00:45,500][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:00:45,996][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:00:46,492][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:00:46,990][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:00:47,488][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:00:47,985][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:00:48,483][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:00:48,983][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:00:49,480][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:00:49,980][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:00:50,480][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:00:50,978][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:00:51,476][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:00:51,973][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:00:52,470][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:00:52,968][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:00:53,465][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:00:53,962][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:00:54,462][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:00:54,963][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:00:55,463][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:00:55,963][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:00:56,464][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:00:56,961][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:00:57,463][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:00:57,962][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:00:58,459][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:00:58,957][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:00:59,455][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:00:59,955][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:01:00,451][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:01:00,951][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:01:01,451][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10348 tokens. +[2026-03-26 00:01:02,092][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.21%, Current % of VRAM taken: 60.63%, Block Peak % of device VRAM: 62.29%, ΔTime: 00:00:32 +[2026-03-26 00:01:02,861][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:01:02,863][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:01:02,865][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:01:03,620][__main__][INFO] - Iteration 139 took 52s (32.65% Gen, 65.91% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 57m 22s. Estimated total time: 43h 57m 36s. Time estimates for 10 more iterations: 8m 47s, 100 more iterations: 1h 27m 55s, 500 more iterations: 7h 19m 36s. +[2026-03-26 00:01:03,622][__main__][INFO] - Starting iteration 139. +[2026-03-26 00:01:04,022][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 13 and human policies 1. +[2026-03-26 00:01:04,023][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:01:20,100][__main__][INFO] - Number of regex retries in iteration 139: 0 +[2026-03-26 00:01:20,101][__main__][INFO] - agents played in iteration 139 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:01:20,891][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:01:20,914][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:01:20,937][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:01:20,960][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:01:20,961][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:01:20,961][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:01:21,595][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:01:22,052][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:01:22,553][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:01:23,057][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:01:23,557][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:01:24,060][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:01:24,558][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:01:25,060][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:01:25,561][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:01:26,058][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:01:26,555][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:01:27,053][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:01:27,550][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:01:28,046][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:01:28,542][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:01:29,039][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:01:29,535][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:01:30,037][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:01:30,538][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:01:31,038][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:01:31,539][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:01:32,039][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:01:32,537][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:01:33,037][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:01:33,535][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:01:34,037][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:01:34,539][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:01:35,036][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:01:35,536][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:01:36,040][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:01:36,540][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:01:37,041][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:01:37,540][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:01:38,039][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:01:38,538][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:01:39,035][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:01:39,537][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:01:40,037][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:01:40,535][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:01:41,036][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:01:41,535][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:01:42,030][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:01:42,526][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:01:43,020][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:01:43,517][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:01:44,013][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:01:44,509][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:01:45,004][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:01:45,501][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:01:45,998][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:01:46,493][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:01:46,989][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:01:47,485][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:01:47,981][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:01:48,477][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:01:48,974][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:01:49,469][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:01:49,964][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:01:50,463][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:01:50,960][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:01:51,457][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:01:51,953][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:01:52,447][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:01:52,944][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:01:53,440][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10279 tokens. +[2026-03-26 00:01:54,106][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 00:01:54,872][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:01:54,874][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:01:54,876][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:01:55,568][__main__][INFO] - Iteration 140 took 51s (31.19% Gen, 67.46% Train). Generation: 16s, Training: 34s. Estimated remaining time: 40h 56m 11s. Estimated total time: 42h 57m 18s. Time estimates for 10 more iterations: 8m 35s, 100 more iterations: 1h 25m 54s, 500 more iterations: 7h 9m 33s. +[2026-03-26 00:01:55,570][__main__][INFO] - Starting iteration 140. +[2026-03-26 00:01:55,972][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 13 and human policies 1. +[2026-03-26 00:01:55,972][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:02:12,611][__main__][INFO] - Number of regex retries in iteration 140: 0 +[2026-03-26 00:02:12,612][__main__][INFO] - agents played in iteration 140 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:02:13,412][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:02:13,435][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:02:13,458][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:02:13,481][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:02:13,481][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:02:13,482][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:02:14,099][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:02:14,557][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:02:15,060][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:02:15,559][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:02:16,059][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:02:16,556][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:02:17,059][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:02:17,559][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:02:18,062][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:02:18,563][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:02:19,063][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:02:19,562][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:02:20,064][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:02:20,565][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:02:21,064][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:02:21,563][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:02:22,063][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:02:22,564][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:02:23,061][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:02:23,560][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:02:24,057][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:02:24,558][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:02:25,058][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:02:25,555][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:02:26,051][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:02:26,550][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:02:27,048][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:02:27,542][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:02:28,038][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:02:28,533][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:02:29,030][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:02:29,525][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:02:30,022][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:02:30,520][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:02:31,018][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:02:31,516][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:02:32,011][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:02:32,510][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:02:33,007][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:02:33,504][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:02:34,000][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:02:34,501][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:02:34,999][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:02:35,499][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:02:35,998][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:02:36,499][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:02:36,996][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:02:37,495][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:02:37,994][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:02:38,493][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:02:38,988][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:02:39,485][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:02:39,982][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:02:40,479][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:02:40,977][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:02:41,474][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:02:41,970][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:02:42,466][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:02:42,963][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:02:43,462][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:02:43,962][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:02:44,459][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:02:44,960][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:02:45,460][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:02:45,960][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10334 tokens. +[2026-03-26 00:02:46,608][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.19%, Current % of VRAM taken: 60.62%, Block Peak % of device VRAM: 62.18%, ΔTime: 00:00:32 +[2026-03-26 00:02:47,370][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:02:47,372][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:02:47,374][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:02:48,744][__main__][INFO] - Iteration 141 took 52s (31.53% Gen, 65.87% Train). Generation: 16s, Training: 34s. Estimated remaining time: 41h 56m 36s. Estimated total time: 43h 58m 37s. Time estimates for 10 more iterations: 8m 47s, 100 more iterations: 1h 27m 57s, 500 more iterations: 7h 19m 46s. +[2026-03-26 00:02:48,746][__main__][INFO] - Starting iteration 141. +[2026-03-26 00:02:49,147][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 14 and human policies 1. +[2026-03-26 00:02:49,148][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:03:05,090][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:03:05,953][__main__][INFO] - Number of regex retries in iteration 141: 1 +[2026-03-26 00:03:05,954][__main__][INFO] - agents played in iteration 141 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:03:06,736][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:03:06,759][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:03:06,782][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:03:06,805][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:03:06,805][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:03:06,806][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:03:07,428][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:03:07,881][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:03:08,385][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:03:08,884][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:03:09,381][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:03:09,880][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:03:10,378][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:03:10,876][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:03:11,372][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:03:11,869][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:03:12,369][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:03:12,867][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:03:13,367][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:03:13,864][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:03:14,362][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:03:14,864][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:03:15,364][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:03:15,864][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:03:16,363][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:03:16,860][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:03:17,357][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:03:17,855][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:03:18,351][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:03:18,849][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:03:19,348][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:03:19,845][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:03:20,347][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:03:20,843][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:03:21,341][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:03:21,839][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:03:22,339][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:03:22,836][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:03:23,334][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:03:23,830][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:03:24,324][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:03:24,819][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:03:25,314][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:03:25,808][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:03:26,305][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:03:26,802][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:03:27,298][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:03:27,798][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:03:28,298][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:03:28,795][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:03:29,291][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:03:29,793][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:03:30,292][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:03:30,789][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:03:31,288][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:03:31,785][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:03:32,282][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:03:32,779][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:03:33,278][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:03:33,776][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:03:34,276][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:03:34,772][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:03:35,269][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:03:35,765][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:03:36,262][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:03:36,759][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:03:37,255][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:03:37,751][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:03:38,247][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:03:38,743][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:03:39,240][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10248 tokens. +[2026-03-26 00:03:39,905][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.04%, ΔTime: 00:00:32 +[2026-03-26 00:03:40,668][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:03:40,671][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:03:40,672][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:03:41,355][__main__][INFO] - Iteration 142 took 52s (32.19% Gen, 66.50% Train). Generation: 16s, Training: 34s. Estimated remaining time: 41h 27m 32s. Estimated total time: 43h 30m 25s. Time estimates for 10 more iterations: 8m 42s, 100 more iterations: 1h 27m 0s, 500 more iterations: 7h 15m 4s. +[2026-03-26 00:03:41,357][__main__][INFO] - Starting iteration 142. +[2026-03-26 00:03:41,759][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 14 and human policies 1. +[2026-03-26 00:03:41,760][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:03:58,866][__main__][INFO] - Number of regex retries in iteration 142: 0 +[2026-03-26 00:03:58,867][__main__][INFO] - agents played in iteration 142 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:03:59,650][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:03:59,673][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:03:59,696][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:03:59,719][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:03:59,719][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:03:59,720][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:04:00,343][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:04:00,796][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:04:01,303][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:04:01,804][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:04:02,304][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:04:02,801][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:04:03,302][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:04:03,803][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:04:04,301][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:04:04,797][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:04:05,293][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:04:05,792][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:04:06,289][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:04:06,784][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:04:07,281][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:04:07,778][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:04:08,275][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:04:08,772][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:04:09,270][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:04:09,767][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:04:10,264][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:04:10,761][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:04:11,260][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:04:11,759][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:04:12,256][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:04:12,751][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:04:13,246][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:04:13,741][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:04:14,236][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:04:14,732][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:04:15,228][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:04:15,723][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:04:16,220][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:04:16,717][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:04:17,214][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:04:17,712][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:04:18,212][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:04:18,709][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:04:19,206][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:04:19,702][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:04:20,200][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:04:20,698][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:04:21,198][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:04:21,695][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:04:22,192][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:04:22,690][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:04:23,190][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:04:23,686][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:04:24,184][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:04:24,683][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:04:25,180][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:04:25,678][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:04:26,175][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:04:26,674][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:04:27,170][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:04:27,666][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:04:28,166][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:04:28,663][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:04:29,160][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:04:29,658][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:04:30,157][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:04:30,652][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:04:31,150][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:04:31,647][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:04:32,143][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10218 tokens. +[2026-03-26 00:04:32,795][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.04%, Current % of VRAM taken: 60.47%, Block Peak % of device VRAM: 62.06%, ΔTime: 00:00:32 +[2026-03-26 00:04:33,558][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:04:33,560][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:04:33,562][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:04:34,247][__main__][INFO] - Iteration 143 took 52s (32.59% Gen, 66.10% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 40m 41s. Estimated total time: 43h 44m 27s. Time estimates for 10 more iterations: 8m 44s, 100 more iterations: 1h 27m 28s, 500 more iterations: 7h 17m 24s. +[2026-03-26 00:04:34,249][__main__][INFO] - Starting iteration 143. +[2026-03-26 00:04:34,649][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 14 and human policies 1. +[2026-03-26 00:04:34,649][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:04:37,905][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:04:50,787][__main__][INFO] - Number of regex retries in iteration 143: 1 +[2026-03-26 00:04:50,788][__main__][INFO] - agents played in iteration 143 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:04:51,567][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:04:51,590][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:04:51,613][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:04:51,636][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:04:51,636][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:04:51,637][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:04:52,256][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:04:52,710][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:04:53,218][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:04:53,716][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:04:54,217][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:04:54,713][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:04:55,213][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:04:55,709][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:04:56,207][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:04:56,705][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:04:57,204][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:04:57,701][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:04:58,198][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:04:58,697][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:04:59,193][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:04:59,691][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:05:00,189][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:05:00,689][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:05:01,187][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:05:01,687][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:05:02,188][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:05:02,684][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:05:03,181][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:05:03,682][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:05:04,181][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:05:04,680][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:05:05,179][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:05:05,678][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:05:06,175][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:05:06,674][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:05:07,174][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:05:07,673][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:05:08,177][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:05:08,677][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:05:09,177][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:05:09,682][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:05:10,182][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:05:10,682][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:05:11,181][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:05:11,677][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:05:12,178][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:05:12,677][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:05:13,177][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:05:13,674][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:05:14,175][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:05:14,672][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:05:15,168][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:05:15,666][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:05:16,167][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:05:16,663][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:05:17,162][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:05:17,660][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:05:18,159][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:05:18,658][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:05:19,156][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:05:19,656][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:05:20,158][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:05:20,655][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:05:21,151][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:05:21,648][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:05:22,145][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:05:22,643][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:05:23,140][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:05:23,639][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:05:24,135][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10292 tokens. +[2026-03-26 00:05:24,793][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.05%, Current % of VRAM taken: 60.47%, Block Peak % of device VRAM: 62.23%, ΔTime: 00:00:32 +[2026-03-26 00:05:25,557][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:05:25,559][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:05:25,561][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:05:26,242][__main__][INFO] - Iteration 144 took 51s (31.28% Gen, 67.40% Train). Generation: 16s, Training: 34s. Estimated remaining time: 40h 55m 3s. Estimated total time: 42h 59m 40s. Time estimates for 10 more iterations: 8m 35s, 100 more iterations: 1h 25m 59s, 500 more iterations: 7h 9m 56s. +[2026-03-26 00:05:26,244][__main__][INFO] - Starting iteration 144. +[2026-03-26 00:05:26,647][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 14 and human policies 1. +[2026-03-26 00:05:26,647][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:05:42,895][__main__][INFO] - Number of regex retries in iteration 144: 0 +[2026-03-26 00:05:42,895][__main__][INFO] - agents played in iteration 144 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:05:43,679][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:05:43,703][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:05:43,730][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:05:43,753][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:05:43,753][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:05:43,754][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:05:44,379][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:05:44,834][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:05:45,341][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:05:45,842][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:05:46,340][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:05:46,842][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:05:47,343][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:05:47,848][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:05:48,348][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:05:48,849][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:05:49,350][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:05:49,848][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:05:50,349][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:05:50,847][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:05:51,348][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:05:51,846][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:05:52,342][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:05:52,843][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:05:53,349][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:05:53,850][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:05:54,348][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:05:54,872][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:05:55,372][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:05:55,871][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:05:56,371][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:05:56,867][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:05:57,364][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:05:57,864][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:05:58,361][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:05:58,859][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:05:59,355][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:05:59,852][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:06:00,348][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:06:00,845][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:06:01,342][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:06:01,836][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:06:02,332][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:06:02,831][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:06:03,331][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:06:03,830][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:06:04,328][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:06:04,821][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:06:05,317][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:06:05,814][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:06:06,310][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:06:06,804][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:06:07,300][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:06:07,794][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:06:08,289][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:06:08,785][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:06:09,283][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:06:09,780][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:06:10,275][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:06:10,772][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:06:11,266][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:06:11,761][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:06:12,257][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:06:12,757][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:06:13,258][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:06:13,756][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:06:14,256][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:06:14,756][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:06:15,254][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:06:15,751][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:06:16,248][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10221 tokens. +[2026-03-26 00:06:16,918][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.32%, ΔTime: 00:00:32 +[2026-03-26 00:06:17,701][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:06:17,704][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:06:17,705][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:06:18,432][__main__][INFO] - Iteration 145 took 51s (31.38% Gen, 67.22% Train). Generation: 16s, Training: 34s. Estimated remaining time: 41h 3m 49s. Estimated total time: 43h 9m 18s. Time estimates for 10 more iterations: 8m 37s, 100 more iterations: 1h 26m 18s, 500 more iterations: 7h 11m 33s. +[2026-03-26 00:06:18,435][__main__][INFO] - Starting iteration 145. +[2026-03-26 00:06:18,873][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 14 and human policies 1. +[2026-03-26 00:06:18,874][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:06:41,490][__main__][INFO] - Number of regex retries in iteration 145: 0 +[2026-03-26 00:06:41,491][__main__][INFO] - agents played in iteration 145 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:06:42,275][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:06:42,299][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:06:42,322][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:06:42,345][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:06:42,345][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:06:42,346][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:06:42,985][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:06:43,436][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:06:43,941][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:06:44,440][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:06:44,943][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:06:45,443][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:06:45,940][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:06:46,441][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:06:46,941][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:06:47,439][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:06:47,938][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:06:48,438][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:06:48,938][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:06:49,433][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:06:49,932][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:06:50,434][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:06:50,928][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:06:51,447][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:06:51,947][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:06:52,447][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:06:52,946][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:06:53,445][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:06:53,941][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:06:54,439][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:06:54,938][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:06:55,437][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:06:55,934][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:06:56,430][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:06:56,926][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:06:57,422][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:06:57,921][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:06:58,419][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:06:58,917][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:06:59,416][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:06:59,914][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:07:00,412][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:07:00,911][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:07:01,409][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:07:01,905][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:07:02,405][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:07:02,904][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:07:03,403][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:07:03,903][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:07:04,404][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:07:04,902][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:07:05,403][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:07:05,902][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:07:06,402][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:07:06,900][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:07:07,399][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:07:07,897][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:07:08,397][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:07:08,893][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:07:09,391][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:07:09,891][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:07:10,386][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:07:10,885][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:07:11,384][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:07:11,883][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:07:12,379][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:07:12,879][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:07:13,380][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:07:13,877][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:07:14,381][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:07:14,883][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10534 tokens. +[2026-03-26 00:07:15,554][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.05%, Current % of VRAM taken: 60.47%, Block Peak % of device VRAM: 62.11%, ΔTime: 00:00:32 +[2026-03-26 00:07:16,314][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:07:16,316][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:07:16,318][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:07:17,014][__main__][INFO] - Iteration 146 took 58s (38.90% Gen, 59.90% Train). Generation: 22s, Training: 34s. Estimated remaining time: 46h 20m 34s. Estimated total time: 48h 27m 2s. Time estimates for 10 more iterations: 9m 41s, 100 more iterations: 1h 36m 54s, 500 more iterations: 8h 4m 30s. +[2026-03-26 00:07:17,016][__main__][INFO] - Starting iteration 146. +[2026-03-26 00:07:17,416][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 14 and human policies 1. +[2026-03-26 00:07:17,417][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:07:34,430][__main__][INFO] - Number of regex retries in iteration 146: 0 +[2026-03-26 00:07:34,431][__main__][INFO] - agents played in iteration 146 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:07:35,213][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:07:35,237][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:07:35,260][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:07:35,282][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:07:35,283][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:07:35,284][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:07:35,919][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:07:36,374][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:07:36,877][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:07:37,380][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:07:37,882][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:07:38,383][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:07:38,884][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:07:39,381][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:07:39,882][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:07:40,387][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:07:40,888][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:07:41,398][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:07:41,897][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:07:42,398][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:07:42,900][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:07:43,398][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:07:43,906][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:07:44,402][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:07:44,901][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:07:45,399][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:07:45,896][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:07:46,392][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:07:46,890][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:07:47,388][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:07:47,885][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:07:48,384][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:07:48,881][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:07:49,379][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:07:49,878][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:07:50,380][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:07:50,876][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:07:51,372][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:07:51,868][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:07:52,368][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:07:52,864][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:07:53,365][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:07:53,865][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:07:54,365][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:07:54,864][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:07:55,363][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:07:55,863][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:07:56,362][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:07:56,861][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:07:57,358][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:07:57,855][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:07:58,352][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:07:58,849][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:07:59,346][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:07:59,843][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:08:00,342][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:08:00,839][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:08:01,338][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:08:01,834][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:08:02,334][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:08:02,833][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:08:03,331][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:08:03,830][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:08:04,329][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:08:04,828][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:08:05,331][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:08:05,832][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:08:06,332][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:08:06,835][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:08:07,339][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:08:07,841][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10392 tokens. +[2026-03-26 00:08:08,511][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.24%, Current % of VRAM taken: 60.67%, Block Peak % of device VRAM: 62.44%, ΔTime: 00:00:32 +[2026-03-26 00:08:09,270][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:08:09,272][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:08:09,274][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:08:10,019][__main__][INFO] - Iteration 147 took 52s (32.35% Gen, 66.24% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 42m 47s. Estimated total time: 43h 50m 9s. Time estimates for 10 more iterations: 8m 46s, 100 more iterations: 1h 27m 40s, 500 more iterations: 7h 18m 21s. +[2026-03-26 00:08:10,021][__main__][INFO] - Starting iteration 147. +[2026-03-26 00:08:10,423][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 14 and human policies 1. +[2026-03-26 00:08:10,424][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:08:20,977][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:08:27,475][__main__][INFO] - Number of regex retries in iteration 147: 1 +[2026-03-26 00:08:27,476][__main__][INFO] - agents played in iteration 147 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:08:28,269][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:08:28,292][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:08:28,315][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:08:28,338][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:08:28,338][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:08:28,339][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:08:28,974][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:08:29,431][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:08:29,932][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:08:30,431][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:08:30,932][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:08:31,434][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:08:31,932][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:08:32,435][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:08:32,933][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:08:33,436][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:08:33,939][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:08:34,444][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:08:34,944][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:08:35,445][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:08:35,945][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:08:36,443][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:08:36,967][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:08:37,467][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:08:37,966][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:08:38,465][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:08:38,963][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:08:39,464][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:08:39,964][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:08:40,461][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:08:40,960][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:08:41,460][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:08:41,960][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:08:42,457][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:08:42,958][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:08:43,455][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:08:43,956][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:08:44,454][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:08:44,957][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:08:45,455][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:08:45,955][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:08:46,456][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:08:46,955][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:08:47,452][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:08:47,954][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:08:48,452][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:08:48,953][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:08:49,455][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:08:49,956][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:08:50,457][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:08:50,960][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:08:51,459][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:08:51,960][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:08:52,459][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:08:52,962][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:08:53,463][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:08:53,963][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:08:54,464][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:08:54,962][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:08:55,463][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:08:55,963][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:08:56,462][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:08:56,963][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:08:57,465][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:08:57,972][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:08:58,471][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:08:58,967][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:08:59,465][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:08:59,966][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:09:00,466][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:09:00,965][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10518 tokens. +[2026-03-26 00:09:01,630][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.17%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.22%, ΔTime: 00:00:32 +[2026-03-26 00:09:02,385][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:09:02,388][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:09:02,389][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:09:03,073][__main__][INFO] - Iteration 148 took 52s (32.39% Gen, 66.31% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 44m 18s. Estimated total time: 43h 52m 32s. Time estimates for 10 more iterations: 8m 46s, 100 more iterations: 1h 27m 45s, 500 more iterations: 7h 18m 45s. +[2026-03-26 00:09:03,076][__main__][INFO] - Starting iteration 148. +[2026-03-26 00:09:03,476][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 14 and human policies 1. +[2026-03-26 00:09:03,477][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:09:21,191][__main__][INFO] - Number of regex retries in iteration 148: 0 +[2026-03-26 00:09:21,192][__main__][INFO] - agents played in iteration 148 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:09:21,977][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:09:22,000][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:09:22,024][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:09:22,047][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:09:22,047][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:09:22,048][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:09:22,685][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:09:23,139][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:09:23,642][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:09:24,141][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:09:24,642][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:09:25,143][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:09:25,642][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:09:26,142][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:09:26,642][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:09:27,147][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:09:27,647][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:09:28,145][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:09:28,663][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:09:29,162][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:09:29,666][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:09:30,165][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:09:30,664][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:09:31,164][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:09:31,662][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:09:32,163][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:09:32,663][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:09:33,164][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:09:33,663][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:09:34,164][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:09:34,663][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:09:35,160][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:09:35,658][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:09:36,155][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:09:36,651][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:09:37,148][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:09:37,648][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:09:38,148][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:09:38,648][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:09:39,147][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:09:39,647][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:09:40,143][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:09:40,640][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:09:41,138][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:09:41,636][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:09:42,133][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:09:42,630][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:09:43,133][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:09:43,634][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:09:44,135][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:09:44,632][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:09:45,132][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:09:45,630][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:09:46,135][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:09:46,635][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:09:47,133][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:09:47,631][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:09:48,132][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:09:48,632][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:09:49,136][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:09:49,637][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:09:50,135][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:09:50,632][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:09:51,131][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:09:51,628][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:09:52,125][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:09:52,623][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:09:53,120][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:09:53,617][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:09:54,115][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:09:54,612][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10515 tokens. +[2026-03-26 00:09:55,281][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.18%, ΔTime: 00:00:32 +[2026-03-26 00:09:56,032][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:09:56,034][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:09:56,035][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:09:56,723][__main__][INFO] - Iteration 149 took 53s (33.27% Gen, 65.44% Train). Generation: 17s, Training: 34s. Estimated remaining time: 42h 13m 15s. Estimated total time: 44h 22m 23s. Time estimates for 10 more iterations: 8m 52s, 100 more iterations: 1h 28m 44s, 500 more iterations: 7h 23m 43s. +[2026-03-26 00:09:56,725][__main__][INFO] - Starting iteration 149. +[2026-03-26 00:09:57,127][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 14 and human policies 1. +[2026-03-26 00:09:57,128][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:10:13,543][__main__][INFO] - Number of regex retries in iteration 149: 0 +[2026-03-26 00:10:13,544][__main__][INFO] - agents played in iteration 149 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:10:14,324][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:10:14,347][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:10:14,370][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:10:14,393][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:10:14,393][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:10:14,394][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:10:15,011][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:10:15,463][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:10:15,969][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:10:16,472][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:10:16,978][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:10:17,481][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:10:17,981][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:10:18,486][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:10:18,986][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:10:19,486][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:10:19,985][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:10:20,484][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:10:20,984][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:10:21,483][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:10:21,983][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:10:22,483][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:10:22,979][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:10:23,479][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:10:23,975][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:10:24,472][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:10:24,973][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:10:25,471][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:10:25,970][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:10:26,468][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:10:26,967][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:10:27,467][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:10:27,967][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:10:28,465][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:10:28,965][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:10:29,465][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:10:29,964][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:10:30,464][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:10:30,963][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:10:31,462][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:10:31,959][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:10:32,457][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:10:32,953][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:10:33,450][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:10:33,945][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:10:34,442][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:10:34,942][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:10:35,439][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:10:35,936][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:10:36,437][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:10:36,936][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:10:37,431][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:10:37,936][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:10:38,437][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:10:38,941][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:10:39,443][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:10:39,942][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:10:40,441][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:10:40,936][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:10:41,434][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:10:41,935][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:10:42,434][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:10:42,929][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:10:43,426][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:10:43,925][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:10:44,424][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:10:44,920][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:10:45,418][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:10:45,918][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:10:46,415][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:10:46,915][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10419 tokens. +[2026-03-26 00:10:47,583][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.20%, Current % of VRAM taken: 60.63%, Block Peak % of device VRAM: 62.25%, ΔTime: 00:00:32 +[2026-03-26 00:10:48,344][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:10:48,346][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:10:48,348][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:10:49,039][__main__][INFO] - Iteration 150 took 51s (31.62% Gen, 67.04% Train). Generation: 16s, Training: 34s. Estimated remaining time: 41h 5m 37s. Estimated total time: 43h 15m 38s. Time estimates for 10 more iterations: 8m 39s, 100 more iterations: 1h 26m 31s, 500 more iterations: 7h 12m 36s. +[2026-03-26 00:10:49,041][__main__][INFO] - Starting iteration 150. +[2026-03-26 00:10:49,442][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 14 and human policies 1. +[2026-03-26 00:10:49,443][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:11:06,393][__main__][INFO] - Number of regex retries in iteration 150: 0 +[2026-03-26 00:11:06,394][__main__][INFO] - agents played in iteration 150 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:11:07,181][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:11:07,205][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:11:07,228][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:11:07,250][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:11:07,251][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:11:07,251][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:11:07,897][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:11:08,351][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:11:08,857][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:11:09,358][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:11:09,859][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:11:10,361][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:11:10,863][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:11:11,360][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:11:11,862][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:11:12,361][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:11:12,862][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:11:13,361][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:11:13,859][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:11:14,357][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:11:14,854][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:11:15,351][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:11:15,849][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:11:16,352][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:11:16,852][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:11:17,352][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:11:17,865][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:11:18,369][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:11:18,868][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:11:19,367][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:11:19,868][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:11:20,368][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:11:20,869][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:11:21,368][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:11:21,864][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:11:22,364][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:11:22,862][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:11:23,359][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:11:23,858][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:11:24,356][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:11:24,853][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:11:25,349][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:11:25,842][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:11:26,339][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:11:26,835][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:11:27,331][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:11:27,827][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:11:28,327][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:11:28,833][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:11:29,332][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:11:29,834][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:11:30,330][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:11:30,830][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:11:31,327][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:11:31,823][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:11:32,323][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:11:32,820][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:11:33,320][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:11:33,816][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:11:34,318][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:11:34,816][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:11:35,316][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:11:35,814][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:11:36,308][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:11:36,807][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:11:37,306][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:11:37,800][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:11:38,298][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:11:38,795][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:11:39,292][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:11:39,788][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10400 tokens. +[2026-03-26 00:11:40,489][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.18%, ΔTime: 00:00:32 +[2026-03-26 00:11:41,250][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:11:41,252][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:11:41,254][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:11:42,717][__main__][INFO] - Iteration 151 took 53s (31.82% Gen, 65.43% Train). Generation: 16s, Training: 34s. Estimated remaining time: 42h 12m 55s. Estimated total time: 44h 23m 49s. Time estimates for 10 more iterations: 8m 52s, 100 more iterations: 1h 28m 47s, 500 more iterations: 7h 23m 58s. +[2026-03-26 00:11:42,721][__main__][INFO] - Starting iteration 151. +[2026-03-26 00:11:43,121][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 15 and human policies 1. +[2026-03-26 00:11:43,121][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:12:03,303][__main__][INFO] - Number of regex retries in iteration 151: 0 +[2026-03-26 00:12:03,305][__main__][INFO] - agents played in iteration 151 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:12:04,089][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:12:04,112][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:12:04,135][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:12:04,158][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:12:04,158][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:12:04,159][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:12:04,796][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:12:05,255][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:12:05,762][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:12:06,265][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:12:06,766][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:12:07,265][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:12:07,766][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:12:08,270][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:12:08,771][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:12:09,272][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:12:09,770][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:12:10,268][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:12:10,769][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:12:11,268][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:12:11,767][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:12:12,267][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:12:12,765][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:12:13,267][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:12:13,766][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:12:14,265][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:12:14,763][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:12:15,263][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:12:15,764][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:12:16,264][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:12:16,763][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:12:17,260][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:12:17,764][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:12:18,262][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:12:18,761][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:12:19,262][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:12:19,764][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:12:20,266][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:12:20,767][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:12:21,266][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:12:21,770][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:12:22,274][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:12:22,781][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:12:23,281][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:12:23,779][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:12:24,281][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:12:24,779][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:12:25,278][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:12:25,774][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:12:26,274][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:12:26,770][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:12:27,270][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:12:27,769][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:12:28,274][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:12:28,773][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:12:29,271][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:12:29,772][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:12:30,268][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:12:30,768][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:12:31,265][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:12:31,765][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:12:32,261][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:12:32,760][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:12:33,260][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:12:33,760][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:12:34,261][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:12:34,760][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:12:35,260][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:12:35,760][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:12:36,260][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:12:36,760][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10514 tokens. +[2026-03-26 00:12:37,438][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.16%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.17%, ΔTime: 00:00:32 +[2026-03-26 00:12:38,199][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:12:38,201][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:12:38,203][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:12:38,932][__main__][INFO] - Iteration 152 took 55s (36.16% Gen, 62.53% Train). Generation: 20s, Training: 34s. Estimated remaining time: 44h 18m 45s. Estimated total time: 46h 30m 35s. Time estimates for 10 more iterations: 9m 18s, 100 more iterations: 1h 33m 1s, 500 more iterations: 7h 45m 5s. +[2026-03-26 00:12:38,934][__main__][INFO] - Starting iteration 152. +[2026-03-26 00:12:39,336][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 15 and human policies 1. +[2026-03-26 00:12:39,337][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:13:04,916][__main__][INFO] - Number of regex retries in iteration 152: 0 +[2026-03-26 00:13:04,917][__main__][INFO] - agents played in iteration 152 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:13:05,701][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:13:05,724][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:13:05,747][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:13:05,770][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:13:05,770][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:13:05,771][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:13:06,397][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:13:06,852][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:13:07,354][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:13:07,856][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:13:08,356][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:13:08,854][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:13:09,350][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:13:09,847][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:13:10,346][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:13:10,848][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:13:11,347][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:13:11,848][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:13:12,352][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:13:12,860][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:13:13,359][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:13:13,859][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:13:14,360][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:13:14,864][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:13:15,365][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:13:15,864][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:13:16,365][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:13:16,865][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:13:17,366][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:13:17,866][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:13:18,365][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:13:18,889][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:13:19,390][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:13:19,891][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:13:20,393][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:13:20,896][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:13:21,398][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:13:21,897][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:13:22,397][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:13:22,897][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:13:23,398][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:13:23,896][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:13:24,393][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:13:24,892][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:13:25,391][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:13:25,889][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:13:26,389][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:13:26,885][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:13:27,383][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:13:27,882][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:13:28,382][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:13:28,881][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:13:29,379][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:13:29,879][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:13:30,375][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:13:30,874][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:13:31,372][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:13:31,871][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:13:32,370][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:13:32,868][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:13:33,367][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:13:33,866][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:13:34,363][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:13:34,861][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:13:35,360][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:13:35,857][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:13:36,354][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:13:36,851][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:13:37,355][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:13:37,857][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:13:38,357][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10465 tokens. +[2026-03-26 00:13:39,026][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 00:13:39,791][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:13:39,793][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:13:39,795][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:13:40,531][__main__][INFO] - Iteration 153 took 1m 1s (41.80% Gen, 56.99% Train). Generation: 25s, Training: 34s. Estimated remaining time: 48h 46m 55s. Estimated total time: 50h 59m 47s. Time estimates for 10 more iterations: 10m 11s, 100 more iterations: 1h 41m 59s, 500 more iterations: 8h 29m 57s. +[2026-03-26 00:13:40,533][__main__][INFO] - Starting iteration 153. +[2026-03-26 00:13:40,934][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 15 and human policies 1. +[2026-03-26 00:13:40,935][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:13:57,557][__main__][INFO] - Number of regex retries in iteration 153: 0 +[2026-03-26 00:13:57,558][__main__][INFO] - agents played in iteration 153 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:13:58,344][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:13:58,367][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:13:58,390][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:13:58,413][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:13:58,414][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:13:58,414][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:13:59,048][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:13:59,503][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:14:00,009][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:14:00,505][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:14:01,004][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:14:01,503][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:14:02,003][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:14:02,501][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:14:02,998][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:14:03,496][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:14:03,993][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:14:04,491][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:14:04,991][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:14:05,489][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:14:05,988][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:14:06,488][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:14:06,986][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:14:07,486][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:14:07,987][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:14:08,491][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:14:08,990][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:14:09,490][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:14:09,987][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:14:10,483][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:14:10,978][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:14:11,477][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:14:11,979][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:14:12,477][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:14:12,979][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:14:13,478][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:14:13,978][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:14:14,478][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:14:14,978][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:14:15,478][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:14:15,978][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:14:16,480][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:14:16,980][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:14:17,480][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:14:17,980][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:14:18,481][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:14:18,982][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:14:19,483][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:14:19,983][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:14:20,486][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:14:20,986][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:14:21,487][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:14:21,986][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:14:22,485][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:14:22,984][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:14:23,481][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:14:23,983][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:14:24,483][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:14:24,983][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:14:25,481][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:14:25,981][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:14:26,479][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:14:26,979][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:14:27,479][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:14:27,978][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:14:28,478][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:14:28,978][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:14:29,480][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:14:29,979][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:14:30,483][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:14:30,985][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10571 tokens. +[2026-03-26 00:14:31,658][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.26%, ΔTime: 00:00:32 +[2026-03-26 00:14:32,419][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:14:32,422][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:14:32,424][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:14:33,111][__main__][INFO] - Iteration 154 took 52s (31.86% Gen, 66.82% Train). Generation: 16s, Training: 34s. Estimated remaining time: 41h 15m 9s. Estimated total time: 43h 28m 54s. Time estimates for 10 more iterations: 8m 41s, 100 more iterations: 1h 26m 57s, 500 more iterations: 7h 14m 49s. +[2026-03-26 00:14:33,114][__main__][INFO] - Starting iteration 154. +[2026-03-26 00:14:33,514][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 15 and human policies 1. +[2026-03-26 00:14:33,515][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:14:50,848][__main__][INFO] - Number of regex retries in iteration 154: 0 +[2026-03-26 00:14:50,849][__main__][INFO] - agents played in iteration 154 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:14:51,632][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:14:51,655][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:14:51,678][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:14:51,700][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:14:51,701][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:14:51,701][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:14:52,337][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:14:52,792][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:14:53,295][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:14:53,793][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:14:54,295][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:14:54,797][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:14:55,299][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:14:55,797][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:14:56,296][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:14:56,800][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:14:57,303][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:14:57,809][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:14:58,331][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:14:58,838][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:14:59,339][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:14:59,841][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:15:00,342][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:15:00,841][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:15:01,339][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:15:01,840][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:15:02,340][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:15:02,844][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:15:03,349][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:15:03,849][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:15:04,346][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:15:04,847][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:15:05,346][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:15:05,847][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:15:06,347][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:15:06,843][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:15:07,343][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:15:07,840][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:15:08,340][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:15:08,838][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:15:09,335][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:15:09,834][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:15:10,331][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:15:10,827][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:15:11,324][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:15:11,819][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:15:12,314][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:15:12,810][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:15:13,308][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:15:13,808][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:15:14,309][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:15:14,808][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:15:15,305][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:15:15,801][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:15:16,301][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:15:16,800][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:15:17,300][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:15:17,796][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:15:18,292][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:15:18,789][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:15:19,286][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:15:19,784][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:15:20,283][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:15:20,784][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:15:21,282][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:15:21,781][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:15:22,282][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:15:22,781][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:15:23,277][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:15:23,777][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:15:24,276][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10470 tokens. +[2026-03-26 00:15:24,942][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.60%, Block Peak % of device VRAM: 62.33%, ΔTime: 00:00:32 +[2026-03-26 00:15:25,704][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:15:25,706][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:15:25,707][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:15:26,392][__main__][INFO] - Iteration 155 took 52s (32.78% Gen, 65.92% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 49m 17s. Estimated total time: 44h 3m 55s. Time estimates for 10 more iterations: 8m 48s, 100 more iterations: 1h 28m 7s, 500 more iterations: 7h 20m 39s. +[2026-03-26 00:15:26,394][__main__][INFO] - Starting iteration 155. +[2026-03-26 00:15:26,793][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 15 and human policies 1. +[2026-03-26 00:15:26,794][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:15:31,022][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:15:36,083][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls + +This proposal is based on the idea that hats are highly valuable to me (value 10) and only slightly valuable to Bob (value 1), so I can potentially grab all the hats for myself without fearing an excessive proposal from Bob that would not benefit him more than me. Books and balls have similar values to both of us, so there's a risk that splitting those might result in a lower overall payoff for me due to the proportional distribution if my proposal is exceeded. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:15:41,984][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:15:43,781][__main__][INFO] - Number of regex retries in iteration 155: 3 +[2026-03-26 00:15:43,782][__main__][INFO] - agents played in iteration 155 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:15:44,563][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:15:44,586][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:15:44,609][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:15:44,631][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:15:44,632][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:15:44,632][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:15:45,250][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:15:45,703][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:15:46,209][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:15:46,709][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:15:47,209][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:15:47,709][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:15:48,211][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:15:48,709][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:15:49,210][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:15:49,713][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:15:50,217][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:15:50,722][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:15:51,225][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:15:51,742][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:15:52,246][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:15:52,746][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:15:53,247][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:15:53,747][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:15:54,246][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:15:54,744][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:15:55,243][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:15:55,743][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:15:56,242][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:15:56,738][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:15:57,235][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:15:57,729][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:15:58,228][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:15:58,728][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:15:59,228][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:15:59,724][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:16:00,222][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:16:00,722][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:16:01,222][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:16:01,723][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:16:02,228][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:16:02,731][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:16:03,234][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:16:03,733][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:16:04,228][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:16:04,726][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:16:05,223][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:16:05,723][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:16:06,223][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:16:06,723][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:16:07,223][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:16:07,722][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:16:08,221][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:16:08,721][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:16:09,220][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:16:09,726][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:16:10,226][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:16:10,726][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:16:11,225][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:16:11,725][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:16:12,225][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:16:12,724][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:16:13,225][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:16:13,726][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:16:14,224][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:16:14,723][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:16:15,223][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:16:15,723][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:16:16,223][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:16:16,721][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:16:17,220][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10609 tokens. +[2026-03-26 00:16:17,907][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.24%, Current % of VRAM taken: 60.66%, Block Peak % of device VRAM: 62.44%, ΔTime: 00:00:32 +[2026-03-26 00:16:18,678][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:16:18,680][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:16:18,682][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:16:19,379][__main__][INFO] - Iteration 156 took 52s (32.30% Gen, 66.37% Train). Generation: 16s, Training: 34s. Estimated remaining time: 41h 33m 49s. Estimated total time: 43h 49m 20s. Time estimates for 10 more iterations: 8m 45s, 100 more iterations: 1h 27m 38s, 500 more iterations: 7h 18m 13s. +[2026-03-26 00:16:19,382][__main__][INFO] - Starting iteration 156. +[2026-03-26 00:16:19,783][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 15 and human policies 1. +[2026-03-26 00:16:19,784][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:16:36,367][__main__][INFO] - Number of regex retries in iteration 156: 0 +[2026-03-26 00:16:36,368][__main__][INFO] - agents played in iteration 156 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:16:37,152][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:16:37,175][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:16:37,199][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:16:37,221][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:16:37,222][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:16:37,223][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:16:37,861][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:16:38,315][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:16:38,819][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:16:39,317][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:16:39,818][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:16:40,322][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:16:40,822][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:16:41,323][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:16:41,825][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:16:42,323][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:16:42,826][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:16:43,325][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:16:43,823][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:16:44,326][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:16:44,827][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:16:45,331][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:16:45,830][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:16:46,329][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:16:46,827][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:16:47,327][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:16:47,824][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:16:48,324][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:16:48,821][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:16:49,323][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:16:49,844][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:16:50,343][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:16:50,844][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:16:51,344][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:16:51,843][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:16:52,343][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:16:52,842][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:16:53,339][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:16:53,840][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:16:54,344][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:16:54,848][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:16:55,354][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:16:55,859][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:16:56,362][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:16:56,861][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:16:57,362][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:16:57,862][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:16:58,359][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:16:58,856][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:16:59,353][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:16:59,853][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:17:00,351][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:17:00,849][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:17:01,348][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:17:01,845][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:17:02,342][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:17:02,840][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:17:03,338][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:17:03,839][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:17:04,339][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:17:04,839][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:17:05,337][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:17:05,834][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:17:06,334][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:17:06,828][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:17:07,327][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:17:07,825][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:17:08,321][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:17:08,819][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:17:09,318][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:17:09,815][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10673 tokens. +[2026-03-26 00:17:10,474][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.18%, ΔTime: 00:00:32 +[2026-03-26 00:17:11,228][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:17:11,230][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:17:11,232][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:17:11,957][__main__][INFO] - Iteration 157 took 52s (31.79% Gen, 66.82% Train). Generation: 16s, Training: 34s. Estimated remaining time: 41h 12m 18s. Estimated total time: 43h 28m 42s. Time estimates for 10 more iterations: 8m 41s, 100 more iterations: 1h 26m 57s, 500 more iterations: 7h 14m 47s. +[2026-03-26 00:17:11,959][__main__][INFO] - Starting iteration 157. +[2026-03-26 00:17:12,362][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 15 and human policies 1. +[2026-03-26 00:17:12,362][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:17:15,875][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:17:29,020][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls +oncé +The reasoning behind this proposal is to maximize my points by taking advantage of my high value for books and Bob's low value for books. Here's a breakdown: + +1. **Hats**: Both of us value hats at 1. Since the value is the same, splitting them equally (10 hats each) is fair and maintains平衡，但重点是我主要追求高价值物品，而鲍勃对此类物品的评价较低。 +2. **Books**: 我对书籍的评价是10，而对方对书籍的评价只有1。如果我能获得所有书籍，那么这将大大增加我的得分。因此，我提议分得所有的书籍。 +3. **Balls**: 我和对方对球的评价都是1，因此平分球类也是合理的，但由于我认为已有更好的策略分配帽子和书籍，直接给出0个球以简化分配。 + +通过这样的策略，我确保了自己在高价值物品（书籍）上的优先获得，同时保持了基本公平分配原则在其他物品上。 did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:17:31,571][__main__][INFO] - Number of regex retries in iteration 157: 2 +[2026-03-26 00:17:31,572][__main__][INFO] - agents played in iteration 157 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:17:32,366][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:17:32,389][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:17:32,412][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:17:32,435][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:17:32,436][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:17:32,436][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:17:33,078][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:17:33,533][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:17:34,041][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:17:34,544][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:17:35,046][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:17:35,544][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:17:36,046][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:17:36,545][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:17:37,049][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:17:37,549][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:17:38,050][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:17:38,547][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:17:39,044][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:17:39,542][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:17:40,040][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:17:40,538][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:17:41,038][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:17:41,538][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:17:42,038][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:17:42,537][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:17:43,037][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:17:43,538][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:17:44,037][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:17:44,533][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:17:45,033][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:17:45,535][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:17:46,037][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:17:46,544][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:17:47,048][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:17:47,557][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:17:48,062][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:17:48,564][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:17:49,063][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:17:49,564][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:17:50,066][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:17:50,567][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:17:51,067][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:17:51,567][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:17:52,067][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:17:52,565][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:17:53,066][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:17:53,566][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:17:54,065][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:17:54,563][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:17:55,063][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:17:55,563][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:17:56,062][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:17:56,563][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:17:57,062][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:17:57,558][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:17:58,057][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:17:58,554][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:17:59,109][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:17:59,609][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:18:00,107][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:18:00,605][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:18:01,104][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:18:01,603][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:18:02,104][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:18:02,603][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:18:03,104][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:18:03,603][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:18:04,103][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:18:04,602][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:18:05,102][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10695 tokens. +[2026-03-26 00:18:05,769][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.19%, Current % of VRAM taken: 60.62%, Block Peak % of device VRAM: 62.39%, ΔTime: 00:00:32 +[2026-03-26 00:18:06,526][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:18:06,528][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:18:06,530][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:18:07,216][__main__][INFO] - Iteration 158 took 54s (35.02% Gen, 63.73% Train). Generation: 19s, Training: 34s. Estimated remaining time: 43h 25m 26s. Estimated total time: 45h 42m 44s. Time estimates for 10 more iterations: 9m 8s, 100 more iterations: 1h 31m 25s, 500 more iterations: 7h 37m 7s. +[2026-03-26 00:18:07,218][__main__][INFO] - Starting iteration 158. +[2026-03-26 00:18:07,619][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 15 and human policies 1. +[2026-03-26 00:18:07,619][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:18:11,284][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:18:24,806][__main__][INFO] - Number of regex retries in iteration 158: 1 +[2026-03-26 00:18:24,808][__main__][INFO] - agents played in iteration 158 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:18:25,593][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:18:25,617][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:18:25,640][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:18:25,662][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:18:25,663][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:18:25,663][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:18:26,306][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:18:26,762][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:18:27,266][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:18:27,765][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:18:28,266][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:18:28,767][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:18:29,272][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:18:29,772][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:18:30,274][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:18:30,769][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:18:31,267][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:18:31,766][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:18:32,268][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:18:32,769][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:18:33,269][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:18:33,773][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:18:34,270][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:18:34,770][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:18:35,271][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:18:35,768][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:18:36,266][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:18:36,764][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:18:37,264][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:18:37,761][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:18:38,262][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:18:38,769][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:18:39,274][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:18:39,775][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:18:40,272][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:18:40,768][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:18:41,269][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:18:41,768][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:18:42,268][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:18:42,765][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:18:43,263][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:18:43,764][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:18:44,261][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:18:44,759][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:18:45,260][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:18:45,759][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:18:46,259][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:18:46,761][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:18:47,259][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:18:47,756][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:18:48,256][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:18:48,757][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:18:49,253][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:18:49,755][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:18:50,252][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:18:50,753][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:18:51,254][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:18:51,753][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:18:52,253][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:18:52,753][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:18:53,252][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:18:53,751][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:18:54,251][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:18:54,751][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:18:55,248][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:18:55,750][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:18:56,251][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:18:56,750][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:18:57,249][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:18:57,749][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:18:58,249][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10598 tokens. +[2026-03-26 00:18:58,919][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.19%, Current % of VRAM taken: 60.61%, Block Peak % of device VRAM: 62.21%, ΔTime: 00:00:32 +[2026-03-26 00:18:59,685][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:18:59,687][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:18:59,689][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:19:00,378][__main__][INFO] - Iteration 159 took 52s (32.58% Gen, 66.11% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 39m 49s. Estimated total time: 43h 58m 1s. Time estimates for 10 more iterations: 8m 47s, 100 more iterations: 1h 27m 56s, 500 more iterations: 7h 19m 40s. +[2026-03-26 00:19:00,380][__main__][INFO] - Starting iteration 159. +[2026-03-26 00:19:00,781][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 15 and human policies 1. +[2026-03-26 00:19:00,782][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:19:16,497][__main__][INFO] - Number of regex retries in iteration 159: 0 +[2026-03-26 00:19:16,499][__main__][INFO] - agents played in iteration 159 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:19:17,278][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:19:17,301][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:19:17,324][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:19:17,347][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:19:17,348][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:19:17,348][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:19:17,983][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:19:18,437][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:19:18,944][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:19:19,447][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:19:19,945][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:19:20,447][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:19:20,949][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:19:21,446][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:19:21,946][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:19:22,447][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:19:22,945][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:19:23,444][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:19:23,944][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:19:24,444][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:19:24,944][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:19:25,443][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:19:25,944][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:19:26,445][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:19:26,944][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:19:27,442][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:19:27,942][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:19:28,443][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:19:28,942][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:19:29,443][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:19:29,942][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:19:30,445][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:19:30,948][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:19:31,455][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:19:31,956][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:19:32,457][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:19:32,957][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:19:33,456][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:19:33,956][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:19:34,456][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:19:34,952][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:19:35,451][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:19:35,947][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:19:36,448][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:19:36,947][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:19:37,447][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:19:37,945][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:19:38,445][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:19:38,944][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:19:39,443][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:19:39,942][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:19:40,441][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:19:40,938][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:19:41,437][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:19:41,933][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:19:42,432][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:19:42,933][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:19:43,429][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:19:43,928][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:19:44,427][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:19:44,925][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:19:45,424][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:19:45,924][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:19:46,422][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:19:46,919][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:19:47,416][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:19:47,913][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:19:48,411][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:19:48,908][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:19:49,405][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:19:49,902][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10595 tokens. +[2026-03-26 00:19:50,569][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 00:19:51,339][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:19:51,341][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:19:51,343][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:19:52,005][__main__][INFO] - Iteration 160 took 51s (30.68% Gen, 68.02% Train). Generation: 15s, Training: 34s. Estimated remaining time: 40h 22m 9s. Estimated total time: 42h 41m 12s. Time estimates for 10 more iterations: 8m 32s, 100 more iterations: 1h 25m 22s, 500 more iterations: 7h 6m 52s. +[2026-03-26 00:19:52,007][__main__][INFO] - Starting iteration 160. +[2026-03-26 00:19:52,416][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 15 and human policies 1. +[2026-03-26 00:19:52,417][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:19:57,010][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:20:09,167][__main__][INFO] - Number of regex retries in iteration 160: 1 +[2026-03-26 00:20:09,167][__main__][INFO] - agents played in iteration 160 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:20:09,957][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:20:09,980][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:20:10,003][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:20:10,026][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:20:10,027][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:20:10,027][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:20:10,657][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:20:11,108][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:20:11,612][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:20:12,111][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:20:12,608][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:20:13,105][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:20:13,604][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:20:14,101][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:20:14,603][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:20:15,104][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:20:15,606][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:20:16,104][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:20:16,604][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:20:17,104][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:20:17,610][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:20:18,112][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:20:18,614][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:20:19,114][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:20:19,614][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:20:20,117][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:20:20,621][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:20:21,126][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:20:21,626][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:20:22,126][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:20:22,628][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:20:23,129][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:20:23,650][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:20:24,147][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:20:24,648][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:20:25,147][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:20:25,646][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:20:26,144][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:20:26,644][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:20:27,139][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:20:27,637][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:20:28,132][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:20:28,628][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:20:29,126][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:20:29,623][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:20:30,121][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:20:30,618][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:20:31,119][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:20:31,618][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:20:32,118][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:20:32,617][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:20:33,116][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:20:33,615][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:20:34,114][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:20:34,614][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:20:35,112][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:20:35,611][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:20:36,111][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:20:36,611][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:20:37,108][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:20:37,608][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:20:38,106][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:20:38,606][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:20:39,105][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:20:39,604][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:20:40,104][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:20:40,601][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:20:41,100][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:20:41,597][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:20:42,093][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:20:42,589][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10571 tokens. +[2026-03-26 00:20:43,248][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.20%, ΔTime: 00:00:32 +[2026-03-26 00:20:44,006][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:20:44,008][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:20:44,010][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:20:45,392][__main__][INFO] - Iteration 161 took 52s (31.62% Gen, 65.77% Train). Generation: 16s, Training: 34s. Estimated remaining time: 41h 48m 53s. Estimated total time: 44h 8m 50s. Time estimates for 10 more iterations: 8m 49s, 100 more iterations: 1h 28m 17s, 500 more iterations: 7h 21m 28s. +[2026-03-26 00:20:45,394][__main__][INFO] - Starting iteration 161. +[2026-03-26 00:20:45,812][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 16 and human policies 1. +[2026-03-26 00:20:45,813][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:21:03,305][__main__][INFO] - Number of regex retries in iteration 161: 0 +[2026-03-26 00:21:03,306][__main__][INFO] - agents played in iteration 161 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:21:04,097][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:21:04,120][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:21:04,143][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:21:04,166][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:21:04,166][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:21:04,167][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:21:04,810][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:21:05,265][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:21:05,774][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:21:06,277][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:21:06,774][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:21:07,272][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:21:07,775][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:21:08,279][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:21:08,783][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:21:09,285][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:21:09,785][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:21:10,287][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:21:10,792][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:21:11,298][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:21:11,805][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:21:12,304][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:21:12,803][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:21:13,303][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:21:13,803][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:21:14,302][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:21:14,803][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:21:15,304][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:21:15,805][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:21:16,304][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:21:16,804][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:21:17,301][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:21:17,799][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:21:18,300][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:21:18,799][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:21:19,297][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:21:19,794][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:21:20,296][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:21:20,795][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:21:21,296][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:21:21,796][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:21:22,293][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:21:22,797][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:21:23,298][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:21:23,797][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:21:24,293][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:21:24,792][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:21:25,313][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:21:25,816][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:21:26,319][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:21:26,821][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:21:27,322][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:21:27,824][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:21:28,326][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:21:28,827][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:21:29,327][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:21:29,826][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:21:30,326][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:21:30,826][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:21:31,327][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:21:31,829][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:21:32,330][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:21:32,831][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:21:33,328][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:21:33,828][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:21:34,324][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:21:34,824][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:21:35,325][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:21:35,824][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:21:36,323][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:21:36,818][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10701 tokens. +[2026-03-26 00:21:37,493][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.39%, ΔTime: 00:00:32 +[2026-03-26 00:21:38,262][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:21:38,264][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:21:38,266][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:21:39,071][__main__][INFO] - Iteration 162 took 53s (32.85% Gen, 65.64% Train). Generation: 17s, Training: 34s. Estimated remaining time: 42h 2m 7s. Estimated total time: 44h 22m 57s. Time estimates for 10 more iterations: 8m 52s, 100 more iterations: 1h 28m 45s, 500 more iterations: 7h 23m 49s. +[2026-03-26 00:21:39,073][__main__][INFO] - Starting iteration 162. +[2026-03-26 00:21:39,472][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 16 and human policies 1. +[2026-03-26 00:21:39,473][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:21:43,067][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:21:55,566][__main__][INFO] - Number of regex retries in iteration 162: 1 +[2026-03-26 00:21:55,567][__main__][INFO] - agents played in iteration 162 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:21:56,346][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:21:56,369][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:21:56,392][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:21:56,415][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:21:56,415][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:21:56,416][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:21:57,076][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:21:57,537][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:21:58,038][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:21:58,538][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:21:59,040][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:21:59,538][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:22:00,040][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:22:00,538][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:22:01,042][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:22:01,553][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:22:02,058][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:22:02,563][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:22:03,066][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:22:03,566][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:22:04,069][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:22:04,596][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:22:05,098][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:22:05,600][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:22:06,101][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:22:06,603][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:22:07,103][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:22:07,604][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:22:08,104][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:22:08,605][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:22:09,105][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:22:09,603][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:22:10,101][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:22:10,599][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:22:11,098][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:22:11,597][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:22:12,093][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:22:12,593][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:22:13,089][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:22:13,589][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:22:14,088][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:22:14,588][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:22:15,088][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:22:15,588][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:22:16,087][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:22:16,587][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:22:17,086][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:22:17,584][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:22:18,087][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:22:18,584][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:22:19,085][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:22:19,584][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:22:20,085][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:22:20,585][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:22:21,089][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:22:21,589][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:22:22,089][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:22:22,589][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:22:23,092][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:22:23,592][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:22:24,091][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:22:24,591][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:22:25,092][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:22:25,593][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:22:26,089][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:22:26,589][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:22:27,086][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:22:27,585][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:22:28,086][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:22:28,586][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:22:29,087][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10725 tokens. +[2026-03-26 00:22:29,774][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.24%, Current % of VRAM taken: 60.66%, Block Peak % of device VRAM: 62.33%, ΔTime: 00:00:32 +[2026-03-26 00:22:30,549][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:22:30,551][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:22:30,553][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:22:31,214][__main__][INFO] - Iteration 163 took 51s (31.10% Gen, 67.62% Train). Generation: 16s, Training: 34s. Estimated remaining time: 40h 45m 24s. Estimated total time: 43h 7m 6s. Time estimates for 10 more iterations: 8m 37s, 100 more iterations: 1h 26m 14s, 500 more iterations: 7h 11m 11s. +[2026-03-26 00:22:31,216][__main__][INFO] - Starting iteration 163. +[2026-03-26 00:22:31,623][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 16 and human policies 1. +[2026-03-26 00:22:31,624][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:22:47,916][__main__][INFO] - Number of regex retries in iteration 163: 0 +[2026-03-26 00:22:47,917][__main__][INFO] - agents played in iteration 163 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:22:48,703][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:22:48,727][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:22:48,750][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:22:48,773][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:22:48,774][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:22:48,774][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:22:49,397][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:22:49,854][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:22:50,362][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:22:50,863][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:22:51,363][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:22:51,863][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:22:52,367][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:22:52,868][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:22:53,369][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:22:53,877][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:22:54,380][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:22:54,883][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:22:55,385][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:22:55,889][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:22:56,390][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:22:56,894][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:22:57,399][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:22:57,899][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:22:58,399][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:22:58,900][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:22:59,401][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:22:59,898][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:23:00,399][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:23:00,899][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:23:01,399][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:23:01,898][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:23:02,399][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:23:02,899][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:23:03,399][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:23:03,898][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:23:04,399][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:23:04,900][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:23:05,400][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:23:05,917][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:23:06,422][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:23:06,922][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:23:07,421][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:23:07,922][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:23:08,423][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:23:08,924][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:23:09,425][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:23:09,927][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:23:10,429][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:23:10,933][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:23:11,434][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:23:11,934][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:23:12,435][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:23:12,934][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:23:13,433][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:23:13,936][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:23:14,436][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:23:14,937][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:23:15,438][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:23:15,939][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:23:16,438][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:23:16,938][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:23:17,436][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:23:17,936][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:23:18,437][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:23:18,937][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:23:19,434][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:23:19,934][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:23:20,432][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:23:20,932][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:23:21,429][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10741 tokens. +[2026-03-26 00:23:22,112][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.15%, Current % of VRAM taken: 60.58%, Block Peak % of device VRAM: 62.36%, ΔTime: 00:00:32 +[2026-03-26 00:23:22,877][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:23:22,879][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:23:22,881][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:23:23,608][__main__][INFO] - Iteration 164 took 51s (31.34% Gen, 67.26% Train). Generation: 16s, Training: 34s. Estimated remaining time: 40h 56m 43s. Estimated total time: 43h 19m 18s. Time estimates for 10 more iterations: 8m 39s, 100 more iterations: 1h 26m 38s, 500 more iterations: 7h 13m 13s. +[2026-03-26 00:23:23,611][__main__][INFO] - Starting iteration 164. +[2026-03-26 00:23:24,012][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 16 and human policies 1. +[2026-03-26 00:23:24,013][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:23:27,719][mllm.models.large_language_model_local][WARNING] - Response Proposer: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:23:40,495][__main__][INFO] - Number of regex retries in iteration 164: 1 +[2026-03-26 00:23:40,496][__main__][INFO] - agents played in iteration 164 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:23:41,276][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:23:41,299][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:23:41,323][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:23:41,346][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:23:41,346][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:23:41,347][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:23:41,965][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:23:42,425][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:23:42,935][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:23:43,436][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:23:43,938][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:23:44,440][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:23:44,940][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:23:45,440][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:23:45,940][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:23:46,439][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:23:46,939][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:23:47,444][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:23:47,947][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:23:48,449][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:23:48,949][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:23:49,450][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:23:49,950][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:23:50,452][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:23:50,953][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:23:51,453][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:23:51,952][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:23:52,451][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:23:52,951][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:23:53,451][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:23:53,951][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:23:54,450][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:23:54,947][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:23:55,446][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:23:55,945][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:23:56,444][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:23:56,943][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:23:57,439][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:23:57,936][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:23:58,436][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:23:58,937][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:23:59,437][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:23:59,937][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:24:00,438][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:24:00,938][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:24:01,439][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:24:01,938][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:24:02,439][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:24:02,940][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:24:03,437][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:24:03,936][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:24:04,436][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:24:04,936][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:24:05,434][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:24:05,933][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:24:06,432][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:24:06,931][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:24:07,431][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:24:07,931][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:24:08,431][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:24:08,932][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:24:09,428][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:24:09,927][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:24:10,427][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:24:10,926][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:24:11,426][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:24:11,926][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:24:12,426][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:24:12,930][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:24:13,447][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:24:13,948][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10683 tokens. +[2026-03-26 00:24:14,622][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.23%, Current % of VRAM taken: 60.66%, Block Peak % of device VRAM: 62.31%, ΔTime: 00:00:32 +[2026-03-26 00:24:15,381][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:24:15,383][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:24:15,385][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:24:16,077][__main__][INFO] - Iteration 165 took 52s (31.66% Gen, 67.01% Train). Generation: 16s, Training: 34s. Estimated remaining time: 40h 59m 51s. Estimated total time: 43h 23m 18s. Time estimates for 10 more iterations: 8m 40s, 100 more iterations: 1h 26m 46s, 500 more iterations: 7h 13m 53s. +[2026-03-26 00:24:16,080][__main__][INFO] - Starting iteration 165. +[2026-03-26 00:24:16,477][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 16 and human policies 1. +[2026-03-26 00:24:16,477][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:24:20,261][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:24:24,601][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:24:32,746][__main__][INFO] - Number of regex retries in iteration 165: 2 +[2026-03-26 00:24:32,747][__main__][INFO] - agents played in iteration 165 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:24:33,526][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:24:33,550][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:24:33,573][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:24:33,595][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:24:33,596][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:24:33,597][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:24:34,220][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:24:34,680][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:24:35,186][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:24:35,683][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:24:36,181][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:24:36,683][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:24:37,182][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:24:37,682][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:24:38,180][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:24:38,684][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:24:39,204][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:24:39,709][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:24:40,214][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:24:40,713][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:24:41,214][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:24:41,716][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:24:42,218][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:24:42,720][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:24:43,217][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:24:43,716][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:24:44,212][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:24:44,711][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:24:45,209][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:24:45,710][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:24:46,209][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:24:46,709][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:24:47,209][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:24:47,708][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:24:48,208][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:24:48,707][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:24:49,206][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:24:49,707][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:24:50,207][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:24:50,708][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:24:51,208][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:24:51,710][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:24:52,210][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:24:52,710][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:24:53,209][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:24:53,709][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:24:54,210][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:24:54,709][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:24:55,210][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:24:55,709][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:24:56,209][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:24:56,708][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:24:57,208][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:24:57,706][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:24:58,206][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:24:58,706][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:24:59,207][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:24:59,707][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:25:00,208][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:25:00,709][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:25:01,209][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:25:01,709][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:25:02,209][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:25:02,706][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:25:03,204][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:25:03,702][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:25:04,203][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:25:04,702][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:25:05,200][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:25:05,698][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:25:06,196][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10671 tokens. +[2026-03-26 00:25:06,874][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.28%, ΔTime: 00:00:32 +[2026-03-26 00:25:07,637][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:25:07,639][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:25:07,641][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:25:08,321][__main__][INFO] - Iteration 166 took 51s (31.38% Gen, 67.30% Train). Generation: 16s, Training: 34s. Estimated remaining time: 40h 47m 55s. Estimated total time: 43h 12m 15s. Time estimates for 10 more iterations: 8m 38s, 100 more iterations: 1h 26m 24s, 500 more iterations: 7h 12m 2s. +[2026-03-26 00:25:08,324][__main__][INFO] - Starting iteration 166. +[2026-03-26 00:25:08,722][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 16 and human policies 1. +[2026-03-26 00:25:08,723][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:25:24,571][__main__][INFO] - Number of regex retries in iteration 166: 0 +[2026-03-26 00:25:24,572][__main__][INFO] - agents played in iteration 166 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:25:25,363][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:25:25,386][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:25:25,410][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:25:25,433][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:25:25,433][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:25:25,434][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:25:26,068][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:25:26,528][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:25:27,043][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:25:27,550][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:25:28,047][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:25:28,547][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:25:29,046][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:25:29,547][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:25:30,045][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:25:30,546][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:25:31,046][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:25:31,546][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:25:32,045][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:25:32,544][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:25:33,045][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:25:33,543][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:25:34,042][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:25:34,543][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:25:35,043][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:25:35,545][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:25:36,047][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:25:36,547][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:25:37,046][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:25:37,546][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:25:38,050][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:25:38,551][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:25:39,050][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:25:39,550][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:25:40,050][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:25:40,550][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:25:41,052][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:25:41,551][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:25:42,051][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:25:42,551][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:25:43,050][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:25:43,550][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:25:44,046][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:25:44,546][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:25:45,045][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:25:45,543][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:25:46,042][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:25:46,563][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:25:47,068][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:25:47,568][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:25:48,068][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:25:48,570][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:25:49,066][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:25:49,566][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:25:50,068][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:25:50,568][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:25:51,067][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:25:51,566][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:25:52,067][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:25:52,565][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:25:53,066][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:25:53,565][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:25:54,063][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:25:54,562][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:25:55,059][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:25:55,559][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:25:56,058][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:25:56,558][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:25:57,057][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:25:57,553][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:25:58,052][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10667 tokens. +[2026-03-26 00:25:58,730][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.20%, Current % of VRAM taken: 60.63%, Block Peak % of device VRAM: 62.46%, ΔTime: 00:00:32 +[2026-03-26 00:25:59,349][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:25:59,351][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:25:59,353][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:26:00,126][__main__][INFO] - Iteration 167 took 51s (30.83% Gen, 67.66% Train). Generation: 15s, Training: 34s. Estimated remaining time: 40h 24m 59s. Estimated total time: 42h 50m 11s. Time estimates for 10 more iterations: 8m 34s, 100 more iterations: 1h 25m 40s, 500 more iterations: 7h 8m 21s. +[2026-03-26 00:26:00,130][__main__][INFO] - Starting iteration 167. +[2026-03-26 00:26:00,529][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 16 and human policies 1. +[2026-03-26 00:26:00,529][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:26:21,149][__main__][INFO] - Number of regex retries in iteration 167: 0 +[2026-03-26 00:26:21,150][__main__][INFO] - agents played in iteration 167 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:26:21,931][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:26:21,954][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:26:21,978][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:26:22,001][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:26:22,001][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:26:22,002][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:26:22,635][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:26:23,093][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:26:23,598][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:26:24,099][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:26:24,602][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:26:25,101][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:26:25,602][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:26:26,106][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:26:26,604][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:26:27,113][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:26:27,619][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:26:28,122][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:26:28,622][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:26:29,124][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:26:29,624][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:26:30,126][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:26:30,628][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:26:31,128][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:26:31,627][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:26:32,128][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:26:32,630][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:26:33,135][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:26:33,638][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:26:34,139][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:26:34,639][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:26:35,139][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:26:35,640][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:26:36,139][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:26:36,640][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:26:37,139][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:26:37,639][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:26:38,138][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:26:38,638][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:26:39,137][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:26:39,638][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:26:40,137][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:26:40,637][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:26:41,138][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:26:41,639][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:26:42,139][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:26:42,639][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:26:43,138][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:26:43,639][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:26:44,139][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:26:44,640][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:26:45,139][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:26:45,639][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:26:46,140][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:26:46,640][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:26:47,139][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:26:47,640][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:26:48,139][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:26:48,639][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:26:49,137][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:26:49,637][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:26:50,136][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:26:50,636][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:26:51,134][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:26:51,629][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:26:52,128][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:26:52,627][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:26:53,126][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:26:53,626][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:26:54,125][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:26:54,622][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10745 tokens. +[2026-03-26 00:26:55,278][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.28%, ΔTime: 00:00:32 +[2026-03-26 00:26:56,012][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:26:56,014][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:26:56,016][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:26:56,728][__main__][INFO] - Iteration 168 took 56s (36.69% Gen, 62.04% Train). Generation: 20s, Training: 34s. Estimated remaining time: 44h 23m 50s. Estimated total time: 46h 49m 59s. Time estimates for 10 more iterations: 9m 21s, 100 more iterations: 1h 33m 39s, 500 more iterations: 7h 48m 19s. +[2026-03-26 00:26:56,730][__main__][INFO] - Starting iteration 168. +[2026-03-26 00:26:57,130][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 16 and human policies 1. +[2026-03-26 00:26:57,131][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:27:12,541][__main__][INFO] - Number of regex retries in iteration 168: 0 +[2026-03-26 00:27:12,542][__main__][INFO] - agents played in iteration 168 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:27:13,318][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:27:13,341][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:27:13,364][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:27:13,387][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:27:13,387][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:27:13,388][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:27:14,007][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:27:14,459][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:27:14,966][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:27:15,465][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:27:15,963][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:27:16,462][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:27:16,963][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:27:17,461][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:27:17,959][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:27:18,461][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:27:18,962][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:27:19,466][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:27:19,966][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:27:20,473][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:27:20,981][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:27:21,486][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:27:21,995][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:27:22,498][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:27:22,999][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:27:23,501][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:27:24,004][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:27:24,508][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:27:25,012][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:27:25,514][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:27:26,015][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:27:26,516][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:27:27,018][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:27:27,519][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:27:28,020][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:27:28,520][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:27:29,020][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:27:29,521][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:27:30,022][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:27:30,520][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:27:31,018][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:27:31,518][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:27:32,018][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:27:32,518][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:27:33,018][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:27:33,515][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:27:34,015][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:27:34,516][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:27:35,015][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:27:35,514][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:27:36,014][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:27:36,515][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:27:37,015][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:27:37,515][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:27:38,012][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:27:38,515][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:27:39,014][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:27:39,515][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:27:40,017][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:27:40,519][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:27:41,019][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:27:41,520][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:27:42,021][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:27:42,520][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:27:43,021][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:27:43,521][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:27:44,021][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:27:44,520][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:27:45,020][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:27:45,521][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:27:46,021][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10827 tokens. +[2026-03-26 00:27:46,675][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.21%, Current % of VRAM taken: 60.63%, Block Peak % of device VRAM: 62.25%, ΔTime: 00:00:32 +[2026-03-26 00:27:47,446][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:27:47,449][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:27:47,451][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:27:48,155][__main__][INFO] - Iteration 169 took 51s (30.20% Gen, 68.42% Train). Generation: 15s, Training: 34s. Estimated remaining time: 40h 4m 15s. Estimated total time: 42h 31m 15s. Time estimates for 10 more iterations: 8m 30s, 100 more iterations: 1h 25m 2s, 500 more iterations: 7h 5m 12s. +[2026-03-26 00:27:48,157][__main__][INFO] - Starting iteration 169. +[2026-03-26 00:27:48,561][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 16 and human policies 1. +[2026-03-26 00:27:48,561][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:27:51,910][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:27:51,964][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:27:57,964][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:28:04,288][__main__][INFO] - Number of regex retries in iteration 169: 3 +[2026-03-26 00:28:04,289][__main__][INFO] - agents played in iteration 169 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:28:05,087][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:28:05,110][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:28:05,134][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:28:05,157][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:28:05,157][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:28:05,158][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:28:05,786][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:28:06,240][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:28:06,747][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:28:07,250][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:28:07,750][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:28:08,252][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:28:08,753][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:28:09,253][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:28:09,758][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:28:10,258][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:28:10,756][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:28:11,258][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:28:11,757][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:28:12,262][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:28:12,763][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:28:13,264][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:28:13,762][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:28:14,266][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:28:14,768][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:28:15,268][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:28:15,768][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:28:16,268][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:28:16,769][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:28:17,272][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:28:17,774][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:28:18,279][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:28:18,781][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:28:19,281][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:28:19,781][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:28:20,283][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:28:20,783][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:28:21,283][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:28:21,783][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:28:22,283][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:28:22,782][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:28:23,281][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:28:23,779][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:28:24,278][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:28:24,777][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:28:25,274][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:28:25,771][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:28:26,271][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:28:26,766][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:28:27,266][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:28:27,766][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:28:28,267][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:28:28,767][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:28:29,267][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:28:29,767][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:28:30,266][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:28:30,763][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:28:31,263][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:28:31,763][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:28:32,262][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:28:32,762][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:28:33,261][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:28:33,760][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:28:34,260][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:28:34,760][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:28:35,259][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:28:35,759][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:28:36,259][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:28:36,760][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:28:37,257][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:28:37,758][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10772 tokens. +[2026-03-26 00:28:38,416][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.23%, Current % of VRAM taken: 60.66%, Block Peak % of device VRAM: 62.34%, ΔTime: 00:00:32 +[2026-03-26 00:28:39,216][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:28:39,219][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:28:39,220][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:28:39,928][__main__][INFO] - Iteration 170 took 51s (30.62% Gen, 68.00% Train). Generation: 15s, Training: 34s. Estimated remaining time: 40h 20m 34s. Estimated total time: 42h 48m 25s. Time estimates for 10 more iterations: 8m 33s, 100 more iterations: 1h 25m 36s, 500 more iterations: 7h 8m 4s. +[2026-03-26 00:28:39,931][__main__][INFO] - Starting iteration 170. +[2026-03-26 00:28:40,332][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 16 and human policies 1. +[2026-03-26 00:28:40,333][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:28:50,650][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:28:57,212][__main__][INFO] - Number of regex retries in iteration 170: 1 +[2026-03-26 00:28:57,213][__main__][INFO] - agents played in iteration 170 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:28:58,006][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:28:58,029][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:28:58,052][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:28:58,075][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:28:58,076][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:28:58,076][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:28:58,699][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:28:59,155][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:28:59,659][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:29:00,160][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:29:00,661][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:29:01,161][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:29:01,663][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:29:02,164][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:29:02,663][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:29:03,164][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:29:03,664][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:29:04,166][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:29:04,666][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:29:05,165][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:29:05,672][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:29:06,174][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:29:06,676][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:29:07,179][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:29:07,681][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:29:08,189][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:29:08,692][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:29:09,194][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:29:09,696][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:29:10,196][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:29:10,701][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:29:11,205][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:29:11,705][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:29:12,207][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:29:12,705][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:29:13,205][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:29:13,703][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:29:14,205][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:29:14,705][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:29:15,205][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:29:15,704][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:29:16,200][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:29:16,698][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:29:17,200][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:29:17,700][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:29:18,201][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:29:18,697][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:29:19,200][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:29:19,699][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:29:20,197][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:29:20,696][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:29:21,197][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:29:21,697][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:29:22,193][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:29:22,694][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:29:23,196][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:29:23,696][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:29:24,196][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:29:24,695][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:29:25,195][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:29:25,695][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:29:26,192][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:29:26,693][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:29:27,194][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:29:27,696][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:29:28,192][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:29:28,692][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:29:29,192][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:29:29,693][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:29:30,194][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:29:30,693][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10692 tokens. +[2026-03-26 00:29:31,360][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.30%, ΔTime: 00:00:32 +[2026-03-26 00:29:32,150][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:29:32,152][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:29:32,154][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:29:33,682][__main__][INFO] - Iteration 171 took 53s (31.64% Gen, 65.49% Train). Generation: 16s, Training: 34s. Estimated remaining time: 41h 58m 45s. Estimated total time: 44h 27m 30s. Time estimates for 10 more iterations: 8m 53s, 100 more iterations: 1h 28m 55s, 500 more iterations: 7h 24m 35s. +[2026-03-26 00:29:33,684][__main__][INFO] - Starting iteration 171. +[2026-03-26 00:29:34,090][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 17 and human policies 1. +[2026-03-26 00:29:34,091][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:29:41,903][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:29:51,492][__main__][INFO] - Number of regex retries in iteration 171: 1 +[2026-03-26 00:29:51,492][__main__][INFO] - agents played in iteration 171 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:29:52,275][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:29:52,298][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:29:52,321][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:29:52,344][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:29:52,344][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:29:52,345][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:29:52,964][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:29:53,419][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:29:53,918][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:29:54,420][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:29:54,921][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:29:55,421][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:29:55,920][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:29:56,420][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:29:56,920][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:29:57,424][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:29:57,924][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:29:58,427][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:29:58,924][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:29:59,424][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:29:59,922][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:30:00,419][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:30:00,921][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:30:01,421][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:30:01,918][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:30:02,419][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:30:02,915][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:30:03,416][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:30:03,915][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:30:04,414][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:30:04,913][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:30:05,412][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:30:05,910][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:30:06,410][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:30:06,910][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:30:07,409][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:30:07,908][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:30:08,407][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:30:08,907][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:30:09,403][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:30:09,904][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:30:10,404][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:30:10,903][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:30:11,403][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:30:11,903][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:30:12,403][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:30:12,902][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:30:13,405][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:30:13,924][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:30:14,426][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:30:14,927][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:30:15,428][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:30:15,929][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:30:16,431][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:30:16,931][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:30:17,431][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:30:17,930][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:30:18,429][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:30:18,930][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:30:19,429][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:30:19,930][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:30:20,431][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:30:20,929][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:30:21,429][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:30:21,927][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:30:22,426][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:30:22,923][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:30:23,424][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:30:23,923][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:30:24,423][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:30:24,923][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10726 tokens. +[2026-03-26 00:30:25,575][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.24%, Current % of VRAM taken: 60.66%, Block Peak % of device VRAM: 62.34%, ΔTime: 00:00:32 +[2026-03-26 00:30:26,315][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:30:26,317][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:30:26,318][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:30:27,024][__main__][INFO] - Iteration 172 took 52s (32.87% Gen, 65.79% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 37m 4s. Estimated total time: 44h 6m 43s. Time estimates for 10 more iterations: 8m 49s, 100 more iterations: 1h 28m 13s, 500 more iterations: 7h 21m 7s. +[2026-03-26 00:30:27,029][__main__][INFO] - Starting iteration 172. +[2026-03-26 00:30:27,429][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 17 and human policies 1. +[2026-03-26 00:30:27,429][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:30:44,038][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 1, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:30:44,879][__main__][INFO] - Number of regex retries in iteration 172: 1 +[2026-03-26 00:30:44,879][__main__][INFO] - agents played in iteration 172 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:30:45,659][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:30:45,683][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:30:45,706][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:30:45,729][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:30:45,729][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:30:45,730][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:30:46,353][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:30:46,809][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:30:47,314][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:30:47,816][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:30:48,322][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:30:48,825][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:30:49,335][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:30:49,834][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:30:50,339][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:30:50,838][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:30:51,339][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:30:51,836][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:30:52,336][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:30:52,837][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:30:53,336][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:30:53,834][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:30:54,334][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:30:54,831][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:30:55,332][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:30:55,836][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:30:56,336][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:30:56,833][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:30:57,330][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:30:57,830][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:30:58,329][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:30:58,828][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:30:59,328][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:30:59,827][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:31:00,327][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:31:00,827][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:31:01,327][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:31:01,826][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:31:02,327][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:31:02,827][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:31:03,326][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:31:03,825][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:31:04,325][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:31:04,824][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:31:05,324][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:31:05,822][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:31:06,321][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:31:06,821][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:31:07,320][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:31:07,819][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:31:08,318][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:31:08,814][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:31:09,315][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:31:09,815][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:31:10,314][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:31:10,813][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:31:11,311][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:31:11,811][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:31:12,308][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:31:12,809][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:31:13,308][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:31:13,809][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:31:14,309][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:31:14,806][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:31:15,304][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:31:15,805][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:31:16,302][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:31:16,803][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:31:17,301][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:31:17,803][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:31:18,306][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10670 tokens. +[2026-03-26 00:31:18,966][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.20%, Current % of VRAM taken: 60.63%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 00:31:19,706][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:31:19,708][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:31:19,710][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:31:20,419][__main__][INFO] - Iteration 173 took 52s (32.93% Gen, 65.73% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 39m 0s. Estimated total time: 44h 9m 32s. Time estimates for 10 more iterations: 8m 49s, 100 more iterations: 1h 28m 19s, 500 more iterations: 7h 21m 35s. +[2026-03-26 00:31:20,421][__main__][INFO] - Starting iteration 173. +[2026-03-26 00:31:20,819][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 17 and human policies 1. +[2026-03-26 00:31:20,820][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:31:28,719][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:31:36,086][__main__][INFO] - Number of regex retries in iteration 173: 1 +[2026-03-26 00:31:36,087][__main__][INFO] - agents played in iteration 173 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:31:36,870][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:31:36,894][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:31:36,916][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:31:36,939][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:31:36,940][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:31:36,940][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:31:37,568][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:31:38,026][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:31:38,532][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:31:39,034][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:31:39,536][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:31:40,040][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:31:40,548][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:31:41,057][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:31:41,557][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:31:42,054][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:31:42,553][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:31:43,050][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:31:43,550][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:31:44,048][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:31:44,546][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:31:45,049][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:31:45,550][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:31:46,049][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:31:46,550][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:31:47,049][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:31:47,552][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:31:48,054][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:31:48,554][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:31:49,051][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:31:49,549][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:31:50,050][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:31:50,551][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:31:51,051][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:31:51,551][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:31:52,050][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:31:52,548][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:31:53,047][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:31:53,545][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:31:54,045][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:31:54,545][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:31:55,046][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:31:55,542][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:31:56,042][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:31:56,542][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:31:57,039][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:31:57,539][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:31:58,039][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:31:58,540][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:31:59,043][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:31:59,545][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:32:00,045][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:32:00,549][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:32:01,050][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:32:01,550][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:32:02,050][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:32:02,550][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:32:03,049][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:32:03,549][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:32:04,050][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:32:04,550][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:32:05,050][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:32:05,549][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:32:06,046][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:32:06,547][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:32:07,046][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:32:07,546][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:32:08,046][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:32:08,548][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:32:09,047][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:32:09,547][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10729 tokens. +[2026-03-26 00:32:10,204][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.25%, Current % of VRAM taken: 60.68%, Block Peak % of device VRAM: 62.40%, ΔTime: 00:00:32 +[2026-03-26 00:32:10,943][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:32:10,945][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:32:10,947][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:32:11,647][__main__][INFO] - Iteration 174 took 50s (30.04% Gen, 68.58% Train). Generation: 15s, Training: 34s. Estimated remaining time: 39h 50m 2s. Estimated total time: 42h 21m 25s. Time estimates for 10 more iterations: 8m 28s, 100 more iterations: 1h 24m 42s, 500 more iterations: 7h 3m 34s. +[2026-03-26 00:32:11,650][__main__][INFO] - Starting iteration 174. +[2026-03-26 00:32:12,051][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 17 and human policies 1. +[2026-03-26 00:32:12,051][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:32:27,648][__main__][INFO] - Number of regex retries in iteration 174: 0 +[2026-03-26 00:32:27,648][__main__][INFO] - agents played in iteration 174 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:32:28,439][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:32:28,463][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:32:28,486][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:32:28,508][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:32:28,509][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:32:28,509][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:32:29,136][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:32:29,591][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:32:30,095][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:32:30,594][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:32:31,091][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:32:31,596][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:32:32,100][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:32:32,605][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:32:33,110][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:32:33,617][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:32:34,120][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:32:34,641][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:32:35,143][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:32:35,647][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:32:36,150][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:32:36,651][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:32:37,152][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:32:37,657][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:32:38,157][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:32:38,657][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:32:39,156][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:32:39,656][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:32:40,160][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:32:40,662][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:32:41,162][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:32:41,662][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:32:42,161][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:32:42,661][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:32:43,162][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:32:43,661][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:32:44,162][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:32:44,663][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:32:45,163][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:32:45,664][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:32:46,164][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:32:46,662][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:32:47,161][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:32:47,661][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:32:48,164][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:32:48,662][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:32:49,163][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:32:49,663][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:32:50,165][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:32:50,664][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:32:51,164][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:32:51,665][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:32:52,161][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:32:52,662][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:32:53,162][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:32:53,664][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:32:54,166][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:32:54,667][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:32:55,170][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:32:55,670][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:32:56,169][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:32:56,671][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:32:57,173][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:32:57,673][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:32:58,174][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:32:58,675][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:32:59,173][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:32:59,673][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:33:00,173][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:33:00,672][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:33:01,173][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10720 tokens. +[2026-03-26 00:33:01,826][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.27%, Current % of VRAM taken: 60.69%, Block Peak % of device VRAM: 62.48%, ΔTime: 00:00:32 +[2026-03-26 00:33:02,566][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:33:02,569][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:33:02,570][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:33:03,419][__main__][INFO] - Iteration 175 took 51s (30.36% Gen, 67.98% Train). Generation: 15s, Training: 34s. Estimated remaining time: 40h 16m 12s. Estimated total time: 42h 48m 27s. Time estimates for 10 more iterations: 8m 33s, 100 more iterations: 1h 25m 36s, 500 more iterations: 7h 8m 4s. +[2026-03-26 00:33:03,421][__main__][INFO] - Starting iteration 175. +[2026-03-26 00:33:03,820][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 17 and human policies 1. +[2026-03-26 00:33:03,821][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:33:16,525][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 1 y book, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:33:19,874][__main__][INFO] - Number of regex retries in iteration 175: 1 +[2026-03-26 00:33:19,875][__main__][INFO] - agents played in iteration 175 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:33:20,656][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:33:20,679][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:33:20,702][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:33:20,725][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:33:20,725][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:33:20,726][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:33:21,353][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:33:21,810][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:33:22,322][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:33:22,830][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:33:23,334][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:33:23,836][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:33:24,338][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:33:24,839][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:33:25,341][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:33:25,840][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:33:26,342][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:33:26,841][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:33:27,341][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:33:27,841][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:33:28,341][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:33:28,841][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:33:29,346][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:33:29,846][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:33:30,343][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:33:30,843][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:33:31,341][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:33:31,845][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:33:32,347][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:33:32,844][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:33:33,343][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:33:33,846][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:33:34,347][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:33:34,848][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:33:35,350][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:33:35,851][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:33:36,351][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:33:36,851][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:33:37,352][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:33:37,854][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:33:38,355][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:33:38,856][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:33:39,357][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:33:39,859][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:33:40,359][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:33:40,860][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:33:41,360][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:33:41,861][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:33:42,361][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:33:42,861][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:33:43,361][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:33:43,861][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:33:44,361][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:33:44,862][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:33:45,363][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:33:45,863][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:33:46,363][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:33:46,861][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:33:47,360][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:33:47,861][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:33:48,361][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:33:48,861][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:33:49,361][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:33:49,861][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:33:50,363][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:33:50,864][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:33:51,363][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:33:51,863][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:33:52,363][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:33:52,862][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:33:53,363][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10803 tokens. +[2026-03-26 00:33:54,019][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.23%, Current % of VRAM taken: 60.65%, Block Peak % of device VRAM: 62.31%, ΔTime: 00:00:32 +[2026-03-26 00:33:54,762][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:33:54,764][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:33:54,766][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:33:55,593][__main__][INFO] - Iteration 176 took 51s (31.01% Gen, 67.39% Train). Generation: 16s, Training: 34s. Estimated remaining time: 40h 35m 32s. Estimated total time: 43h 8m 39s. Time estimates for 10 more iterations: 8m 37s, 100 more iterations: 1h 26m 17s, 500 more iterations: 7h 11m 26s. +[2026-03-26 00:33:55,595][__main__][INFO] - Starting iteration 176. +[2026-03-26 00:33:55,993][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 17 and human policies 1. +[2026-03-26 00:33:55,994][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:34:12,352][__main__][INFO] - Number of regex retries in iteration 176: 0 +[2026-03-26 00:34:12,353][__main__][INFO] - agents played in iteration 176 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:34:13,142][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:34:13,165][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:34:13,188][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:34:13,211][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:34:13,212][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:34:13,212][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:34:13,838][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:34:14,291][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:34:14,798][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:34:15,302][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:34:15,805][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:34:16,316][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:34:16,819][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:34:17,319][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:34:17,820][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:34:18,321][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:34:18,821][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:34:19,319][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:34:19,815][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:34:20,313][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:34:20,812][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:34:21,309][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:34:21,810][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:34:22,311][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:34:22,808][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:34:23,308][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:34:23,808][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:34:24,305][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:34:24,807][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:34:25,306][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:34:25,805][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:34:26,304][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:34:26,803][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:34:27,300][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:34:27,798][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:34:28,295][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:34:28,790][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:34:29,286][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:34:29,785][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:34:30,290][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:34:30,806][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:34:31,307][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:34:31,807][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:34:32,306][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:34:32,805][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:34:33,306][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:34:33,807][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:34:34,306][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:34:34,805][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:34:35,306][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:34:35,807][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:34:36,307][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:34:36,808][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:34:37,307][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:34:37,806][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:34:38,302][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:34:38,798][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:34:39,293][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:34:39,791][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:34:40,287][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:34:40,784][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:34:41,283][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:34:41,779][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:34:42,280][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:34:42,779][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:34:43,281][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:34:43,780][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:34:44,281][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:34:44,782][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:34:45,281][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:34:45,782][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10731 tokens. +[2026-03-26 00:34:46,432][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.28%, Current % of VRAM taken: 60.70%, Block Peak % of device VRAM: 62.47%, ΔTime: 00:00:32 +[2026-03-26 00:34:47,242][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:34:47,244][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:34:47,246][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:34:48,070][__main__][INFO] - Iteration 177 took 52s (31.41% Gen, 67.00% Train). Generation: 16s, Training: 34s. Estimated remaining time: 40h 49m 53s. Estimated total time: 43h 23m 53s. Time estimates for 10 more iterations: 8m 40s, 100 more iterations: 1h 26m 47s, 500 more iterations: 7h 13m 58s. +[2026-03-26 00:34:48,073][__main__][INFO] - Starting iteration 177. +[2026-03-26 00:34:48,471][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 17 and human policies 1. +[2026-03-26 00:34:48,472][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:35:05,295][__main__][INFO] - Number of regex retries in iteration 177: 0 +[2026-03-26 00:35:05,296][__main__][INFO] - agents played in iteration 177 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:35:06,085][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.59%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:35:06,108][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.59%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:35:06,131][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.59%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:35:06,154][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.59%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:35:06,154][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:35:06,155][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:35:06,782][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:35:07,235][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:35:07,744][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:35:08,245][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:35:08,743][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:35:09,244][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:35:09,746][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:35:10,247][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:35:10,745][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:35:11,247][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:35:11,748][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:35:12,248][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:35:12,748][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:35:13,248][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:35:13,749][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:35:14,250][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:35:14,751][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:35:15,254][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:35:15,755][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:35:16,255][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:35:16,757][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:35:17,259][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:35:17,762][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:35:18,263][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:35:18,764][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:35:19,262][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:35:19,759][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:35:20,254][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:35:20,756][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:35:21,255][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:35:21,752][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:35:22,251][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:35:22,750][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:35:23,250][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:35:23,750][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:35:24,249][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:35:24,749][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:35:25,246][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:35:25,746][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:35:26,246][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:35:26,745][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:35:27,244][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:35:27,741][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:35:28,241][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:35:28,740][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:35:29,240][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:35:29,737][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:35:30,233][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:35:30,733][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:35:31,236][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:35:31,738][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:35:32,240][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:35:32,740][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:35:33,241][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:35:33,759][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:35:34,261][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:35:34,762][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:35:35,262][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:35:35,761][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:35:36,264][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:35:36,766][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:35:37,268][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:35:37,769][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:35:38,270][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:35:38,770][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10737 tokens. +[2026-03-26 00:35:39,423][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.20%, Current % of VRAM taken: 60.62%, Block Peak % of device VRAM: 62.25%, ΔTime: 00:00:32 +[2026-03-26 00:35:40,167][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:35:40,170][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:35:40,171][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:35:40,937][__main__][INFO] - Iteration 178 took 52s (32.07% Gen, 66.47% Train). Generation: 16s, Training: 34s. Estimated remaining time: 41h 8m 27s. Estimated total time: 43h 43m 20s. Time estimates for 10 more iterations: 8m 44s, 100 more iterations: 1h 27m 26s, 500 more iterations: 7h 17m 13s. +[2026-03-26 00:35:40,940][__main__][INFO] - Starting iteration 178. +[2026-03-26 00:35:41,339][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 17 and human policies 1. +[2026-03-26 00:35:41,340][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:35:58,430][__main__][INFO] - Number of regex retries in iteration 178: 0 +[2026-03-26 00:35:58,431][__main__][INFO] - agents played in iteration 178 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:35:59,217][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:35:59,240][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:35:59,263][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:35:59,286][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:35:59,287][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:35:59,287][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:35:59,906][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:36:00,363][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:36:00,868][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:36:01,367][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:36:01,867][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:36:02,367][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:36:02,866][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:36:03,364][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:36:03,866][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:36:04,366][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:36:04,869][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:36:05,373][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:36:05,875][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:36:06,375][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:36:06,876][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:36:07,379][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:36:07,882][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:36:08,381][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:36:08,881][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:36:09,383][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:36:09,882][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:36:10,386][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:36:10,886][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:36:11,386][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:36:11,884][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:36:12,383][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:36:12,882][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:36:13,383][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:36:13,884][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:36:14,382][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:36:14,880][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:36:15,379][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:36:15,878][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:36:16,381][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:36:16,880][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:36:17,381][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:36:17,881][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:36:18,382][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:36:18,883][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:36:19,384][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:36:19,885][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:36:20,385][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:36:20,886][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:36:21,385][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:36:21,883][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:36:22,382][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:36:22,883][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:36:23,384][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:36:23,884][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:36:24,383][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:36:24,885][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:36:25,384][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:36:25,884][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:36:26,382][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:36:26,883][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:36:27,382][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:36:27,882][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:36:28,383][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:36:28,884][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:36:29,386][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:36:29,887][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:36:30,388][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:36:30,889][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:36:31,390][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:36:31,892][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10792 tokens. +[2026-03-26 00:36:32,546][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.20%, Current % of VRAM taken: 60.63%, Block Peak % of device VRAM: 62.29%, ΔTime: 00:00:32 +[2026-03-26 00:36:33,290][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:36:33,293][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:36:33,295][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:36:34,091][__main__][INFO] - Iteration 179 took 52s (32.40% Gen, 66.09% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 21m 51s. Estimated total time: 43h 57m 37s. Time estimates for 10 more iterations: 8m 47s, 100 more iterations: 1h 27m 55s, 500 more iterations: 7h 19m 36s. +[2026-03-26 00:36:34,093][__main__][INFO] - Starting iteration 179. +[2026-03-26 00:36:34,496][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 17 and human policies 1. +[2026-03-26 00:36:34,496][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:36:50,573][__main__][INFO] - Number of regex retries in iteration 179: 0 +[2026-03-26 00:36:50,574][__main__][INFO] - agents played in iteration 179 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:36:51,352][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:36:51,376][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:36:51,399][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:36:51,422][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:36:51,422][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:36:51,423][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:36:52,063][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:36:52,519][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:36:53,021][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:36:53,520][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:36:54,019][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:36:54,522][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:36:55,024][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:36:55,524][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:36:56,021][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:36:56,522][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:36:57,023][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:36:57,522][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:36:58,022][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:36:58,522][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:36:59,023][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:36:59,523][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:37:00,024][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:37:00,526][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:37:01,026][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:37:01,527][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:37:02,031][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:37:02,531][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:37:03,034][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:37:03,537][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:37:04,039][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:37:04,539][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:37:05,038][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:37:05,538][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:37:06,035][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:37:06,532][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:37:07,029][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:37:07,528][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:37:08,028][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:37:08,528][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:37:09,026][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:37:09,524][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:37:10,022][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:37:10,520][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:37:11,021][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:37:11,520][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:37:12,018][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:37:12,518][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:37:13,017][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:37:13,517][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:37:14,017][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:37:14,518][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:37:15,018][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:37:15,518][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:37:16,019][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:37:16,517][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:37:17,019][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:37:17,519][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:37:18,023][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:37:18,524][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:37:19,023][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:37:19,520][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:37:20,023][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:37:20,525][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:37:21,026][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:37:21,526][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:37:22,027][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:37:22,527][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:37:23,027][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:37:23,527][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:37:24,027][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10692 tokens. +[2026-03-26 00:37:24,698][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.23%, Current % of VRAM taken: 60.66%, Block Peak % of device VRAM: 62.37%, ΔTime: 00:00:32 +[2026-03-26 00:37:25,445][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:37:25,447][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:37:25,449][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:37:26,132][__main__][INFO] - Iteration 180 took 51s (31.14% Gen, 67.54% Train). Generation: 16s, Training: 34s. Estimated remaining time: 40h 25m 12s. Estimated total time: 43h 1m 49s. Time estimates for 10 more iterations: 8m 36s, 100 more iterations: 1h 26m 3s, 500 more iterations: 7h 10m 18s. +[2026-03-26 00:37:26,134][__main__][INFO] - Starting iteration 180. +[2026-03-26 00:37:26,538][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 17 and human policies 1. +[2026-03-26 00:37:26,539][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:37:34,520][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:37:43,189][__main__][INFO] - Number of regex retries in iteration 180: 1 +[2026-03-26 00:37:43,190][__main__][INFO] - agents played in iteration 180 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:37:43,961][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:37:43,985][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:37:44,008][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:37:44,031][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:37:44,031][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:37:44,032][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:37:44,664][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:37:45,121][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:37:45,622][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:37:46,122][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:37:46,621][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:37:47,124][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:37:47,624][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:37:48,121][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:37:48,621][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:37:49,121][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:37:49,622][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:37:50,127][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:37:50,628][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:37:51,127][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:37:51,624][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:37:52,122][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:37:52,623][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:37:53,124][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:37:53,625][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:37:54,125][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:37:54,625][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:37:55,126][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:37:55,627][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:37:56,129][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:37:56,629][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:37:57,133][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:37:57,653][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:37:58,155][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:37:58,658][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:37:59,161][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:37:59,663][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:38:00,165][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:38:00,667][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:38:01,166][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:38:01,666][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:38:02,165][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:38:02,664][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:38:03,164][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:38:03,664][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:38:04,162][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:38:04,663][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:38:05,162][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:38:05,660][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:38:06,161][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:38:06,662][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:38:07,162][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:38:07,662][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:38:08,162][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:38:08,661][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:38:09,160][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:38:09,659][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:38:10,156][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:38:10,657][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:38:11,154][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:38:11,651][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:38:12,148][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:38:12,649][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:38:13,147][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:38:13,646][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:38:14,146][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:38:14,646][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:38:15,146][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:38:15,646][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:38:16,143][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:38:16,643][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10696 tokens. +[2026-03-26 00:38:17,304][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.25%, Current % of VRAM taken: 60.67%, Block Peak % of device VRAM: 62.49%, ΔTime: 00:00:32 +[2026-03-26 00:38:18,066][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:38:18,069][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:38:18,073][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:38:19,489][__main__][INFO] - Iteration 181 took 52s (31.45% Gen, 65.88% Train). Generation: 16s, Training: 34s. Estimated remaining time: 41h 30m 4s. Estimated total time: 44h 7m 35s. Time estimates for 10 more iterations: 8m 49s, 100 more iterations: 1h 28m 15s, 500 more iterations: 7h 21m 15s. +[2026-03-26 00:38:19,492][__main__][INFO] - Starting iteration 181. +[2026-03-26 00:38:19,894][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 18 and human policies 1. +[2026-03-26 00:38:19,894][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:38:23,356][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:38:34,020][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:38:35,585][__main__][INFO] - Number of regex retries in iteration 181: 2 +[2026-03-26 00:38:35,586][__main__][INFO] - agents played in iteration 181 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:38:36,355][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.60%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:38:36,378][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.60%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:38:36,401][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.60%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:38:36,424][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.60%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:38:36,424][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:38:36,425][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:38:37,055][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:38:37,511][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:38:38,013][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:38:38,514][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:38:39,017][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:38:39,519][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:38:40,020][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:38:40,521][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:38:41,021][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:38:41,524][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:38:42,024][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:38:42,524][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:38:43,023][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:38:43,522][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:38:44,023][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:38:44,523][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:38:45,023][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:38:45,542][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:38:46,047][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:38:46,548][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:38:47,049][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:38:47,550][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:38:48,054][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:38:48,555][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:38:49,058][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:38:49,558][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:38:50,061][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:38:50,561][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:38:51,061][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:38:51,561][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:38:52,061][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:38:52,562][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:38:53,063][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:38:53,563][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:38:54,064][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:38:54,560][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:38:55,060][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:38:55,561][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:38:56,060][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:38:56,560][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:38:57,060][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:38:57,556][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:38:58,053][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:38:58,553][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:38:59,048][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:38:59,545][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:39:00,046][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:39:00,546][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:39:01,047][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:39:01,546][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:39:02,044][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:39:02,543][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:39:03,041][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:39:03,540][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:39:04,039][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:39:04,538][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:39:05,038][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:39:05,535][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:39:06,035][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:39:06,535][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:39:07,035][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:39:07,534][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:39:08,033][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:39:08,532][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:39:09,028][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10767 tokens. +[2026-03-26 00:39:09,686][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.17%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.30%, ΔTime: 00:00:32 +[2026-03-26 00:39:10,461][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:39:10,463][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:39:10,465][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:39:11,166][__main__][INFO] - Iteration 182 took 51s (30.60% Gen, 68.03% Train). Generation: 15s, Training: 34s. Estimated remaining time: 40h 5m 15s. Estimated total time: 42h 43m 37s. Time estimates for 10 more iterations: 8m 32s, 100 more iterations: 1h 25m 27s, 500 more iterations: 7h 7m 16s. +[2026-03-26 00:39:11,168][__main__][INFO] - Starting iteration 182. +[2026-03-26 00:39:11,566][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 18 and human policies 1. +[2026-03-26 00:39:11,566][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:39:27,324][__main__][INFO] - Number of regex retries in iteration 182: 0 +[2026-03-26 00:39:27,325][__main__][INFO] - agents played in iteration 182 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:39:28,100][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:39:28,123][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:39:28,146][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:39:28,169][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:39:28,169][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:39:28,170][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:39:28,801][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:39:29,257][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:39:29,763][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:39:30,264][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:39:30,765][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:39:31,265][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:39:31,766][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:39:32,268][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:39:32,769][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:39:33,269][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:39:33,770][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:39:34,270][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:39:34,770][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:39:35,271][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:39:35,772][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:39:36,273][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:39:36,774][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:39:37,274][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:39:37,770][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:39:38,268][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:39:38,767][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:39:39,270][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:39:39,772][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:39:40,274][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:39:40,772][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:39:41,272][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:39:41,775][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:39:42,276][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:39:42,776][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:39:43,278][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:39:43,777][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:39:44,280][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:39:44,777][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:39:45,279][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:39:45,778][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:39:46,278][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:39:46,778][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:39:47,277][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:39:47,777][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:39:48,274][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:39:48,774][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:39:49,274][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:39:49,772][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:39:50,274][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:39:50,774][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:39:51,274][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:39:51,775][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:39:52,275][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:39:52,775][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:39:53,275][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:39:53,773][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:39:54,273][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:39:54,773][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:39:55,272][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:39:55,770][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:39:56,269][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:39:56,765][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:39:57,267][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:39:57,769][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:39:58,269][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:39:58,769][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:39:59,269][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:39:59,769][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:40:00,271][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:40:00,771][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10773 tokens. +[2026-03-26 00:40:01,432][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.26%, Current % of VRAM taken: 60.69%, Block Peak % of device VRAM: 62.50%, ΔTime: 00:00:32 +[2026-03-26 00:40:02,160][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:40:02,163][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:40:02,164][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:40:02,934][__main__][INFO] - Iteration 183 took 51s (30.68% Gen, 67.82% Train). Generation: 15s, Training: 34s. Estimated remaining time: 40h 9m 12s. Estimated total time: 42h 48m 27s. Time estimates for 10 more iterations: 8m 33s, 100 more iterations: 1h 25m 36s, 500 more iterations: 7h 8m 4s. +[2026-03-26 00:40:02,937][__main__][INFO] - Starting iteration 183. +[2026-03-26 00:40:03,341][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 18 and human policies 1. +[2026-03-26 00:40:03,342][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:40:19,541][__main__][INFO] - Number of regex retries in iteration 183: 0 +[2026-03-26 00:40:19,542][__main__][INFO] - agents played in iteration 183 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:40:20,322][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:40:20,346][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:40:20,368][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:40:20,391][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:40:20,391][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:40:20,392][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:40:21,017][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:40:21,470][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:40:21,976][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:40:22,537][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:40:23,041][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:40:23,542][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:40:24,043][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:40:24,542][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:40:25,041][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:40:25,540][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:40:26,039][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:40:26,538][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:40:27,036][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:40:27,534][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:40:28,031][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:40:28,529][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:40:29,028][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:40:29,530][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:40:30,031][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:40:30,536][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:40:31,036][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:40:31,539][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:40:32,041][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:40:32,542][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:40:33,042][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:40:33,541][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:40:34,039][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:40:34,537][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:40:35,033][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:40:35,529][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:40:36,026][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:40:36,524][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:40:37,023][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:40:37,523][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:40:38,023][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:40:38,521][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:40:39,020][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:40:39,520][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:40:40,020][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:40:40,520][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:40:41,021][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:40:41,521][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:40:42,020][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:40:42,546][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:40:43,048][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:40:43,548][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:40:44,048][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:40:44,548][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:40:45,047][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:40:45,551][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:40:46,054][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:40:46,553][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:40:47,052][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:40:47,553][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:40:48,053][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:40:48,553][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:40:49,053][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:40:49,553][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:40:50,053][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:40:50,550][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:40:51,049][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:40:51,548][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:40:52,043][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:40:52,541][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:40:53,040][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10809 tokens. +[2026-03-26 00:40:53,702][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.20%, Current % of VRAM taken: 60.63%, Block Peak % of device VRAM: 62.39%, ΔTime: 00:00:32 +[2026-03-26 00:40:54,440][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:40:54,442][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:40:54,444][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:40:55,160][__main__][INFO] - Iteration 184 took 51s (31.26% Gen, 67.35% Train). Generation: 16s, Training: 34s. Estimated remaining time: 40h 30m 53s. Estimated total time: 43h 11m 0s. Time estimates for 10 more iterations: 8m 38s, 100 more iterations: 1h 26m 22s, 500 more iterations: 7h 11m 50s. +[2026-03-26 00:40:55,162][__main__][INFO] - Starting iteration 184. +[2026-03-26 00:40:55,563][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 18 and human policies 1. +[2026-03-26 00:40:55,564][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:40:59,437][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:41:10,938][__main__][INFO] - Number of regex retries in iteration 184: 1 +[2026-03-26 00:41:10,939][__main__][INFO] - agents played in iteration 184 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:41:11,713][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:41:11,736][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:41:11,759][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:41:11,781][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:41:11,782][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:41:11,783][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:41:12,414][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:41:12,872][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:41:13,378][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:41:13,880][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:41:14,377][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:41:14,878][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:41:15,381][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:41:15,882][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:41:16,383][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:41:16,880][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:41:17,381][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:41:17,880][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:41:18,379][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:41:18,879][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:41:19,381][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:41:19,883][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:41:20,390][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:41:20,890][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:41:21,390][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:41:21,893][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:41:22,397][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:41:22,899][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:41:23,403][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:41:23,907][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:41:24,409][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:41:24,908][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:41:25,408][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:41:25,907][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:41:26,407][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:41:26,907][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:41:27,406][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:41:27,907][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:41:28,406][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:41:28,907][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:41:29,407][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:41:29,907][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:41:30,407][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:41:30,906][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:41:31,407][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:41:31,907][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:41:32,407][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:41:32,907][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:41:33,406][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:41:33,902][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:41:34,399][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:41:34,899][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:41:35,398][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:41:35,898][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:41:36,398][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:41:36,897][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:41:37,398][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:41:37,898][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:41:38,402][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:41:38,902][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:41:39,402][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:41:39,902][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:41:40,401][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:41:40,900][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:41:41,399][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:41:41,897][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:41:42,400][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:41:42,900][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:41:43,398][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:41:43,898][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:41:44,394][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10769 tokens. +[2026-03-26 00:41:45,050][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.23%, ΔTime: 00:00:32 +[2026-03-26 00:41:45,794][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:41:45,797][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:41:45,799][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:41:46,637][__main__][INFO] - Iteration 185 took 51s (30.10% Gen, 68.25% Train). Generation: 15s, Training: 34s. Estimated remaining time: 39h 52m 47s. Estimated total time: 42h 33m 45s. Time estimates for 10 more iterations: 8m 30s, 100 more iterations: 1h 25m 7s, 500 more iterations: 7h 5m 37s. +[2026-03-26 00:41:46,640][__main__][INFO] - Starting iteration 185. +[2026-03-26 00:41:47,040][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 18 and human policies 1. +[2026-03-26 00:41:47,040][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:42:02,785][__main__][INFO] - Number of regex retries in iteration 185: 0 +[2026-03-26 00:42:02,786][__main__][INFO] - agents played in iteration 185 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:42:03,552][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:42:03,576][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:42:03,598][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:42:03,621][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:42:03,622][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:42:03,622][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:42:04,258][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:42:04,715][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:42:05,222][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:42:05,724][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:42:06,228][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:42:06,730][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:42:07,233][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:42:07,734][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:42:08,234][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:42:08,735][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:42:09,236][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:42:09,735][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:42:10,234][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:42:10,737][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:42:11,239][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:42:11,739][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:42:12,245][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:42:12,750][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:42:13,253][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:42:13,754][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:42:14,258][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:42:14,762][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:42:15,266][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:42:15,769][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:42:16,272][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:42:16,775][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:42:17,278][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:42:17,779][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:42:18,282][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:42:18,783][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:42:19,286][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:42:19,787][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:42:20,291][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:42:20,794][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:42:21,294][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:42:21,796][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:42:22,298][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:42:22,800][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:42:23,303][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:42:23,806][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:42:24,307][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:42:24,809][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:42:25,310][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:42:25,811][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:42:26,313][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:42:26,816][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:42:27,316][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:42:27,820][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:42:28,323][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:42:28,825][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:42:29,323][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:42:29,824][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:42:30,323][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:42:30,824][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:42:31,327][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:42:31,829][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:42:32,327][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:42:32,827][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:42:33,329][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:42:33,829][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:42:34,330][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:42:34,830][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:42:35,331][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:42:35,829][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:42:36,330][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10796 tokens. +[2026-03-26 00:42:37,013][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.26%, Current % of VRAM taken: 60.68%, Block Peak % of device VRAM: 62.41%, ΔTime: 00:00:32 +[2026-03-26 00:42:37,756][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:42:37,758][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:42:37,760][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:42:38,463][__main__][INFO] - Iteration 186 took 51s (30.62% Gen, 68.01% Train). Generation: 15s, Training: 34s. Estimated remaining time: 40h 9m 25s. Estimated total time: 42h 51m 14s. Time estimates for 10 more iterations: 8m 34s, 100 more iterations: 1h 25m 42s, 500 more iterations: 7h 8m 32s. +[2026-03-26 00:42:38,466][__main__][INFO] - Starting iteration 186. +[2026-03-26 00:42:38,866][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 18 and human policies 1. +[2026-03-26 00:42:38,867][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:42:41,889][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:42:41,968][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:42:54,573][__main__][INFO] - Number of regex retries in iteration 186: 2 +[2026-03-26 00:42:54,574][__main__][INFO] - agents played in iteration 186 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:42:55,342][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:42:55,365][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:42:55,388][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:42:55,411][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:42:55,411][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:42:55,412][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:42:56,056][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:42:56,515][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:42:57,020][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:42:57,522][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:42:58,025][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:42:58,527][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:42:59,028][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:42:59,528][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:43:00,029][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:43:00,530][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:43:01,031][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:43:01,531][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:43:02,031][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:43:02,530][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:43:03,029][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:43:03,529][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:43:04,030][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:43:04,527][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:43:05,027][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:43:05,531][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:43:06,035][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:43:06,537][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:43:07,035][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:43:07,538][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:43:08,041][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:43:08,544][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:43:09,047][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:43:09,551][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:43:10,055][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:43:10,555][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:43:11,058][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:43:11,560][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:43:12,060][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:43:12,562][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:43:13,063][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:43:13,564][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:43:14,066][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:43:14,566][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:43:15,067][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:43:15,569][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:43:16,071][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:43:16,573][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:43:17,074][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:43:17,574][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:43:18,077][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:43:18,580][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:43:19,081][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:43:19,583][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:43:20,084][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:43:20,583][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:43:21,083][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:43:21,585][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:43:22,086][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:43:22,584][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:43:23,082][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:43:23,580][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:43:24,079][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:43:24,582][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:43:25,083][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:43:25,584][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:43:26,086][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:43:26,586][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:43:27,087][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:43:27,588][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:43:28,086][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10726 tokens. +[2026-03-26 00:43:28,768][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.17%, Current % of VRAM taken: 60.60%, Block Peak % of device VRAM: 62.28%, ΔTime: 00:00:32 +[2026-03-26 00:43:29,520][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:43:29,523][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:43:29,526][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:43:30,240][__main__][INFO] - Iteration 187 took 51s (30.57% Gen, 68.03% Train). Generation: 15s, Training: 34s. Estimated remaining time: 40h 6m 0s. Estimated total time: 42h 48m 42s. Time estimates for 10 more iterations: 8m 33s, 100 more iterations: 1h 25m 37s, 500 more iterations: 7h 8m 7s. +[2026-03-26 00:43:30,242][__main__][INFO] - Starting iteration 187. +[2026-03-26 00:43:30,640][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 18 and human policies 1. +[2026-03-26 00:43:30,641][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:43:47,716][__main__][INFO] - Number of regex retries in iteration 187: 0 +[2026-03-26 00:43:47,717][__main__][INFO] - agents played in iteration 187 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:43:48,486][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:43:48,510][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:43:48,532][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:43:48,555][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:43:48,555][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:43:48,556][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:43:49,210][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:43:49,666][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:43:50,172][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:43:50,673][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:43:51,175][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:43:51,672][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:43:52,172][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:43:52,672][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:43:53,173][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:43:53,673][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:43:54,173][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:43:54,672][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:43:55,170][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:43:55,669][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:43:56,165][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:43:56,660][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:43:57,159][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:43:57,656][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:43:58,159][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:43:58,662][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:43:59,169][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:43:59,667][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:44:00,166][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:44:00,668][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:44:01,168][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:44:01,667][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:44:02,167][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:44:02,669][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:44:03,170][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:44:03,671][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:44:04,168][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:44:04,669][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:44:05,166][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:44:05,666][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:44:06,163][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:44:06,663][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:44:07,160][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:44:07,660][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:44:08,159][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:44:08,658][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:44:09,155][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:44:09,659][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:44:10,163][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:44:10,666][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:44:11,167][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:44:11,667][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:44:12,164][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:44:12,662][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:44:13,162][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:44:13,660][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:44:14,160][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:44:14,661][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:44:15,163][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:44:15,662][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:44:16,160][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:44:16,660][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:44:17,159][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:44:17,661][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:44:18,159][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:44:18,660][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:44:19,160][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:44:19,661][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:44:20,163][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:44:20,661][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:44:21,162][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10625 tokens. +[2026-03-26 00:44:21,846][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.17%, Current % of VRAM taken: 60.60%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 00:44:22,589][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:44:22,591][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:44:22,593][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:44:23,337][__main__][INFO] - Iteration 188 took 52s (32.40% Gen, 66.18% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 11m 18s. Estimated total time: 43h 54m 53s. Time estimates for 10 more iterations: 8m 46s, 100 more iterations: 1h 27m 49s, 500 more iterations: 7h 19m 8s. +[2026-03-26 00:44:23,340][__main__][INFO] - Starting iteration 188. +[2026-03-26 00:44:23,741][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 18 and human policies 1. +[2026-03-26 00:44:23,742][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:44:27,335][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:44:40,376][__main__][INFO] - Number of regex retries in iteration 188: 1 +[2026-03-26 00:44:40,377][__main__][INFO] - agents played in iteration 188 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:44:41,160][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:44:41,183][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:44:41,206][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:44:41,229][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:44:41,230][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:44:41,230][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:44:41,881][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:44:42,339][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:44:42,842][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:44:43,349][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:44:43,851][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:44:44,351][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:44:44,848][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:44:45,349][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:44:45,850][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:44:46,356][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:44:46,857][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:44:47,359][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:44:47,863][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:44:48,363][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:44:48,866][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:44:49,367][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:44:49,868][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:44:50,368][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:44:50,868][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:44:51,371][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:44:51,875][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:44:52,375][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:44:52,872][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:44:53,373][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:44:53,877][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:44:54,378][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:44:54,880][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:44:55,384][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:44:55,886][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:44:56,390][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:44:56,891][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:44:57,389][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:44:57,890][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:44:58,391][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:44:58,893][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:44:59,398][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:44:59,900][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:45:00,400][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:45:00,901][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:45:01,402][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:45:01,907][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:45:02,404][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:45:02,903][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:45:03,403][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:45:03,901][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:45:04,401][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:45:04,900][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:45:05,401][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:45:05,899][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:45:06,400][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:45:06,901][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:45:07,400][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:45:07,903][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:45:08,405][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:45:08,906][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:45:09,406][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:45:09,908][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:45:10,409][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:45:10,911][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:45:11,413][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:45:11,914][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:45:12,414][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:45:12,913][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:45:13,413][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:45:13,913][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10694 tokens. +[2026-03-26 00:45:14,598][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.20%, Current % of VRAM taken: 60.63%, Block Peak % of device VRAM: 62.39%, ΔTime: 00:00:32 +[2026-03-26 00:45:15,340][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:45:15,342][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:45:15,344][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:45:16,046][__main__][INFO] - Iteration 189 took 52s (31.80% Gen, 66.85% Train). Generation: 16s, Training: 34s. Estimated remaining time: 40h 50m 48s. Estimated total time: 43h 35m 15s. Time estimates for 10 more iterations: 8m 43s, 100 more iterations: 1h 27m 10s, 500 more iterations: 7h 15m 52s. +[2026-03-26 00:45:16,048][__main__][INFO] - Starting iteration 189. +[2026-03-26 00:45:16,451][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 18 and human policies 1. +[2026-03-26 00:45:16,452][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:45:32,750][__main__][INFO] - Number of regex retries in iteration 189: 0 +[2026-03-26 00:45:32,752][__main__][INFO] - agents played in iteration 189 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:45:33,523][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:45:33,546][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:45:33,569][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:45:33,592][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:45:33,592][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:45:33,593][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:45:34,243][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:45:34,697][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:45:35,203][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:45:35,705][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:45:36,202][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:45:36,704][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:45:37,203][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:45:37,701][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:45:38,202][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:45:38,699][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:45:39,205][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:45:39,708][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:45:40,209][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:45:40,711][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:45:41,212][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:45:41,714][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:45:42,212][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:45:42,712][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:45:43,212][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:45:43,711][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:45:44,212][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:45:44,717][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:45:45,223][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:45:45,732][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:45:46,233][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:45:46,736][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:45:47,239][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:45:47,743][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:45:48,246][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:45:48,749][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:45:49,249][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:45:49,750][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:45:50,249][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:45:50,747][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:45:51,246][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:45:51,746][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:45:52,243][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:45:52,744][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:45:53,244][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:45:53,740][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:45:54,238][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:45:54,740][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:45:55,240][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:45:55,739][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:45:56,236][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:45:56,735][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:45:57,234][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:45:57,734][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:45:58,233][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:45:58,730][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:45:59,231][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:45:59,732][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:46:00,230][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:46:00,730][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:46:01,232][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:46:01,734][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:46:02,231][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:46:02,731][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:46:03,227][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:46:03,730][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:46:04,229][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:46:04,729][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:46:05,225][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:46:05,723][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:46:06,222][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10653 tokens. +[2026-03-26 00:46:06,913][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.01%, Current % of VRAM taken: 60.44%, Block Peak % of device VRAM: 62.34%, ΔTime: 00:00:32 +[2026-03-26 00:46:07,667][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:46:07,669][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:46:07,671][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:46:08,421][__main__][INFO] - Iteration 190 took 51s (31.36% Gen, 67.19% Train). Generation: 16s, Training: 34s. Estimated remaining time: 40h 33m 15s. Estimated total time: 43h 18m 35s. Time estimates for 10 more iterations: 8m 39s, 100 more iterations: 1h 26m 37s, 500 more iterations: 7h 13m 5s. +[2026-03-26 00:46:08,424][__main__][INFO] - Starting iteration 190. +[2026-03-26 00:46:08,881][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 18 and human policies 1. +[2026-03-26 00:46:08,882][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:46:26,021][__main__][INFO] - Number of regex retries in iteration 190: 0 +[2026-03-26 00:46:26,022][__main__][INFO] - agents played in iteration 190 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:46:26,793][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:46:26,816][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:46:26,839][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:46:26,862][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:46:26,863][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:46:26,863][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:46:27,519][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:46:27,977][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:46:28,475][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:46:28,972][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:46:29,472][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:46:29,973][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:46:30,473][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:46:30,973][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:46:31,470][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:46:31,970][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:46:32,468][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:46:32,968][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:46:33,467][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:46:33,966][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:46:34,466][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:46:34,965][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:46:35,465][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:46:35,963][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:46:36,461][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:46:36,960][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:46:37,459][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:46:37,962][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:46:38,468][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:46:38,970][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:46:39,470][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:46:39,975][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:46:40,503][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:46:41,006][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:46:41,510][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:46:42,015][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:46:42,516][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:46:43,020][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:46:43,520][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:46:44,017][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:46:44,515][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:46:45,012][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:46:45,512][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:46:46,011][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:46:46,510][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:46:47,009][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:46:47,509][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:46:48,010][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:46:48,511][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:46:49,014][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:46:49,515][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:46:50,017][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:46:50,517][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:46:51,020][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:46:51,517][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:46:52,016][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:46:52,514][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:46:53,013][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:46:53,510][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:46:54,008][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:46:54,507][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:46:55,007][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:46:55,507][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:46:56,006][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:46:56,506][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:46:57,007][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:46:57,510][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:46:58,010][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:46:58,513][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:46:59,017][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:46:59,522][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10740 tokens. +[2026-03-26 00:47:00,222][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.24%, Current % of VRAM taken: 60.67%, Block Peak % of device VRAM: 62.34%, ΔTime: 00:00:32 +[2026-03-26 00:47:00,978][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:47:00,981][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:47:00,983][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:47:02,278][__main__][INFO] - Iteration 191 took 53s (32.10% Gen, 65.47% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 43m 39s. Estimated total time: 44h 29m 53s. Time estimates for 10 more iterations: 8m 53s, 100 more iterations: 1h 28m 59s, 500 more iterations: 7h 24m 58s. +[2026-03-26 00:47:02,281][__main__][INFO] - Starting iteration 191. +[2026-03-26 00:47:02,679][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 19 and human policies 1. +[2026-03-26 00:47:02,680][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:47:07,285][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:47:07,364][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:47:16,874][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:47:21,178][__main__][INFO] - Number of regex retries in iteration 191: 3 +[2026-03-26 00:47:21,179][__main__][INFO] - agents played in iteration 191 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:47:21,955][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.59%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:47:21,978][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.59%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:47:22,002][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.59%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:47:22,025][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.59%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:47:22,026][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:47:22,026][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:47:22,687][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:47:23,141][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:47:23,645][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:47:24,144][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:47:24,645][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:47:25,143][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:47:25,640][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:47:26,141][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:47:26,641][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:47:27,141][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:47:27,642][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:47:28,141][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:47:28,640][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:47:29,139][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:47:29,638][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:47:30,136][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:47:30,636][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:47:31,143][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:47:31,647][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:47:32,151][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:47:32,649][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:47:33,151][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:47:33,654][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:47:34,156][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:47:34,658][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:47:35,160][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:47:35,663][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:47:36,165][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:47:36,666][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:47:37,166][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:47:37,665][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:47:38,167][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:47:38,667][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:47:39,166][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:47:39,662][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:47:40,159][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:47:40,655][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:47:41,152][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:47:41,652][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:47:42,148][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:47:42,647][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:47:43,147][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:47:43,646][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:47:44,145][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:47:44,645][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:47:45,149][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:47:45,649][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:47:46,150][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:47:46,650][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:47:47,146][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:47:47,649][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:47:48,150][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:47:48,650][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:47:49,152][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:47:49,655][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:47:50,159][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:47:50,658][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:47:51,156][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:47:51,657][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:47:52,156][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:47:52,655][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:47:53,154][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:47:53,651][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:47:54,150][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:47:54,647][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10693 tokens. +[2026-03-26 00:47:55,339][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.16%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.26%, ΔTime: 00:00:32 +[2026-03-26 00:47:56,084][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:47:56,087][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:47:56,089][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:47:56,811][__main__][INFO] - Iteration 192 took 54s (34.17% Gen, 64.49% Train). Generation: 18s, Training: 34s. Estimated remaining time: 42h 19m 28s. Estimated total time: 45h 6m 37s. Time estimates for 10 more iterations: 9m 1s, 100 more iterations: 1h 30m 13s, 500 more iterations: 7h 31m 6s. +[2026-03-26 00:47:56,813][__main__][INFO] - Starting iteration 192. +[2026-03-26 00:47:57,495][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 19 and human policies 1. +[2026-03-26 00:47:57,496][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:48:09,285][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:48:13,482][__main__][INFO] - Number of regex retries in iteration 192: 1 +[2026-03-26 00:48:13,483][__main__][INFO] - agents played in iteration 192 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:48:14,253][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:48:14,277][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:48:14,300][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:48:14,323][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:48:14,323][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:48:14,324][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:48:14,967][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:48:15,423][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:48:15,927][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:48:16,427][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:48:16,929][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:48:17,430][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:48:17,928][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:48:18,427][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:48:18,926][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:48:19,427][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:48:19,924][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:48:20,426][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:48:20,926][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:48:21,424][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:48:21,923][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:48:22,422][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:48:22,921][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:48:23,419][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:48:23,918][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:48:24,423][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:48:24,927][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:48:25,429][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:48:25,927][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:48:26,428][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:48:26,929][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:48:27,428][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:48:27,930][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:48:28,430][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:48:28,929][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:48:29,428][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:48:29,927][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:48:30,426][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:48:30,924][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:48:31,425][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:48:31,922][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:48:32,421][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:48:32,924][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:48:33,426][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:48:33,926][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:48:34,429][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:48:34,928][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:48:35,426][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:48:35,926][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:48:36,424][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:48:36,924][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:48:37,423][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:48:37,922][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:48:38,421][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:48:38,918][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:48:39,419][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:48:39,919][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:48:40,419][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:48:40,924][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:48:41,433][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:48:41,935][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:48:42,436][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:48:42,953][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:48:43,456][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:48:43,953][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:48:44,455][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:48:44,957][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:48:45,453][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:48:45,952][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:48:46,450][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:48:46,950][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10675 tokens. +[2026-03-26 00:48:47,634][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.21%, Current % of VRAM taken: 60.64%, Block Peak % of device VRAM: 62.23%, ΔTime: 00:00:32 +[2026-03-26 00:48:48,381][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:48:48,383][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:48:48,385][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:48:49,096][__main__][INFO] - Iteration 193 took 51s (30.81% Gen, 67.27% Train). Generation: 15s, Training: 34s. Estimated remaining time: 40h 26m 5s. Estimated total time: 43h 14m 5s. Time estimates for 10 more iterations: 8m 38s, 100 more iterations: 1h 26m 28s, 500 more iterations: 7h 12m 20s. +[2026-03-26 00:48:49,098][__main__][INFO] - Starting iteration 193. +[2026-03-26 00:48:49,497][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 19 and human policies 1. +[2026-03-26 00:48:49,498][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:49:06,896][__main__][INFO] - Number of regex retries in iteration 193: 0 +[2026-03-26 00:49:06,897][__main__][INFO] - agents played in iteration 193 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:49:07,669][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:49:07,693][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:49:07,716][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:49:07,739][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:49:07,740][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:49:07,740][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:49:08,399][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:49:08,852][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:49:09,359][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:49:09,856][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:49:10,354][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:49:10,855][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:49:11,351][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:49:11,850][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:49:12,346][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:49:12,850][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:49:13,369][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:49:13,872][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:49:14,371][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:49:14,872][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:49:15,371][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:49:15,871][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:49:16,368][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:49:16,868][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:49:17,367][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:49:17,872][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:49:18,380][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:49:18,886][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:49:19,385][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:49:19,889][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:49:20,395][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:49:20,898][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:49:21,401][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:49:21,904][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:49:22,404][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:49:22,908][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:49:23,411][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:49:23,911][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:49:24,411][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:49:24,907][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:49:25,405][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:49:25,905][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:49:26,403][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:49:26,904][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:49:27,405][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:49:27,905][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:49:28,405][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:49:28,905][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:49:29,405][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:49:29,901][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:49:30,403][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:49:30,902][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:49:31,402][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:49:31,903][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:49:32,408][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:49:32,914][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:49:33,422][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:49:33,922][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:49:34,422][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:49:34,925][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:49:35,426][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:49:35,925][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:49:36,425][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:49:36,925][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:49:37,426][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:49:37,930][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:49:38,432][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:49:38,929][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:49:39,429][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:49:39,928][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:49:40,425][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10609 tokens. +[2026-03-26 00:49:41,101][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.39%, ΔTime: 00:00:32 +[2026-03-26 00:49:41,844][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:49:41,847][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:49:41,848][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:49:42,682][__main__][INFO] - Iteration 194 took 53s (32.71% Gen, 65.72% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 30m 20s. Estimated total time: 44h 19m 14s. Time estimates for 10 more iterations: 8m 51s, 100 more iterations: 1h 28m 38s, 500 more iterations: 7h 23m 12s. +[2026-03-26 00:49:42,684][__main__][INFO] - Starting iteration 194. +[2026-03-26 00:49:43,085][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 19 and human policies 1. +[2026-03-26 00:49:43,085][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:49:59,957][__main__][INFO] - Number of regex retries in iteration 194: 0 +[2026-03-26 00:49:59,958][__main__][INFO] - agents played in iteration 194 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:50:00,746][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:50:00,769][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:50:00,792][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:50:00,815][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:50:00,816][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:50:00,816][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:50:01,473][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:50:01,925][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:50:02,434][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:50:02,934][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:50:03,435][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:50:03,930][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:50:04,430][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:50:04,930][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:50:05,430][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:50:05,928][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:50:06,427][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:50:06,927][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:50:07,426][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:50:07,922][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:50:08,422][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:50:08,923][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:50:09,424][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:50:09,923][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:50:10,423][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:50:10,923][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:50:11,425][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:50:11,931][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:50:12,433][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:50:12,932][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:50:13,435][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:50:13,938][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:50:14,437][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:50:14,938][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:50:15,440][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:50:15,940][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:50:16,441][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:50:16,941][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:50:17,438][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:50:17,935][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:50:18,437][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:50:18,936][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:50:19,435][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:50:19,938][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:50:20,439][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:50:20,935][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:50:21,434][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:50:21,939][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:50:22,440][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:50:22,942][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:50:23,450][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:50:23,957][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:50:24,462][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:50:24,964][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:50:25,468][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:50:25,973][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:50:26,474][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:50:26,976][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:50:27,478][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:50:27,976][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:50:28,476][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:50:28,976][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:50:29,476][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:50:29,973][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:50:30,469][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:50:30,969][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:50:31,470][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:50:31,971][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:50:32,468][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:50:32,967][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:50:33,467][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10610 tokens. +[2026-03-26 00:50:34,175][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.22%, Current % of VRAM taken: 60.64%, Block Peak % of device VRAM: 62.25%, ΔTime: 00:00:32 +[2026-03-26 00:50:34,975][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:50:34,977][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:50:34,979][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:50:35,743][__main__][INFO] - Iteration 195 took 52s (32.04% Gen, 66.50% Train). Generation: 16s, Training: 35s. Estimated remaining time: 41h 3m 11s. Estimated total time: 43h 52m 58s. Time estimates for 10 more iterations: 8m 46s, 100 more iterations: 1h 27m 45s, 500 more iterations: 7h 18m 49s. +[2026-03-26 00:50:35,746][__main__][INFO] - Starting iteration 195. +[2026-03-26 00:50:36,152][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 19 and human policies 1. +[2026-03-26 00:50:36,153][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:51:00,672][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:51:02,661][__main__][INFO] - Number of regex retries in iteration 195: 1 +[2026-03-26 00:51:02,662][__main__][INFO] - agents played in iteration 195 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:51:03,455][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:51:03,478][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:51:03,501][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:51:03,524][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:51:03,524][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:51:03,525][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:51:04,160][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:51:04,615][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:51:05,117][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:51:05,619][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:51:06,117][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:51:06,618][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:51:07,116][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:51:07,616][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:51:08,115][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:51:08,614][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:51:09,113][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:51:09,613][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:51:10,113][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:51:10,614][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:51:11,114][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:51:11,610][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:51:12,111][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:51:12,610][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:51:13,112][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:51:13,612][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:51:14,118][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:51:14,626][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:51:15,133][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:51:15,641][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:51:16,143][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:51:16,649][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:51:17,146][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:51:17,648][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:51:18,150][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:51:18,651][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:51:19,149][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:51:19,651][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:51:20,152][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:51:20,655][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:51:21,159][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:51:21,661][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:51:22,158][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:51:22,659][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:51:23,161][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:51:23,661][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:51:24,162][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:51:24,661][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:51:25,160][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:51:25,660][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:51:26,159][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:51:26,659][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:51:27,155][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:51:27,658][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:51:28,155][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:51:28,656][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:51:29,157][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:51:29,657][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:51:30,160][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:51:30,661][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:51:31,162][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:51:31,662][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:51:32,162][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:51:32,658][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:51:33,161][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:51:33,662][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:51:34,162][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:51:34,661][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:51:35,160][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:51:35,657][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:51:36,159][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10665 tokens. +[2026-03-26 00:51:36,832][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.17%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.30%, ΔTime: 00:00:32 +[2026-03-26 00:51:37,567][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:51:37,569][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:51:37,570][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:51:38,307][__main__][INFO] - Iteration 196 took 1m 2s (42.65% Gen, 56.16% Train). Generation: 26s, Training: 34s. Estimated remaining time: 48h 56m 59s. Estimated total time: 51h 47m 48s. Time estimates for 10 more iterations: 10m 21s, 100 more iterations: 1h 43m 35s, 500 more iterations: 8h 37m 58s. +[2026-03-26 00:51:38,309][__main__][INFO] - Starting iteration 196. +[2026-03-26 00:51:38,713][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 19 and human policies 1. +[2026-03-26 00:51:38,713][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:51:44,920][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:51:57,837][__main__][INFO] - Number of regex retries in iteration 196: 1 +[2026-03-26 00:51:57,838][__main__][INFO] - agents played in iteration 196 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:51:58,614][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:51:58,637][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:51:58,660][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:51:58,682][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:51:58,683][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:51:58,683][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:51:59,330][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:51:59,783][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:52:00,288][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:52:00,789][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:52:01,288][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:52:01,786][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:52:02,287][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:52:02,785][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:52:03,286][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:52:03,789][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:52:04,291][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:52:04,790][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:52:05,287][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:52:05,785][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:52:06,286][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:52:06,805][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:52:07,309][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:52:07,806][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:52:08,308][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:52:08,814][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:52:09,314][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:52:09,816][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:52:10,321][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:52:10,822][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:52:11,318][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:52:11,817][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:52:12,317][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:52:12,819][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:52:13,323][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:52:13,824][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:52:14,327][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:52:14,831][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:52:15,334][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:52:15,838][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:52:16,339][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:52:16,836][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:52:17,336][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:52:17,838][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:52:18,341][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:52:18,841][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:52:19,341][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:52:19,838][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:52:20,337][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:52:20,837][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:52:21,334][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:52:21,834][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:52:22,334][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:52:22,833][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:52:23,330][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:52:23,830][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:52:24,331][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:52:24,830][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:52:25,331][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:52:25,830][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:52:26,331][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:52:26,835][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:52:27,336][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:52:27,837][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:52:28,334][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:52:28,831][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:52:29,330][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:52:29,832][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:52:30,330][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:52:30,829][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:52:31,329][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10545 tokens. +[2026-03-26 00:52:32,014][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.21%, Current % of VRAM taken: 60.64%, Block Peak % of device VRAM: 62.39%, ΔTime: 00:00:32 +[2026-03-26 00:52:32,765][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:52:32,767][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:52:32,769][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:52:33,423][__main__][INFO] - Iteration 197 took 54s (34.96% Gen, 63.84% Train). Generation: 19s, Training: 34s. Estimated remaining time: 42h 43m 50s. Estimated total time: 45h 35m 35s. Time estimates for 10 more iterations: 9m 7s, 100 more iterations: 1h 31m 11s, 500 more iterations: 7h 35m 55s. +[2026-03-26 00:52:33,425][__main__][INFO] - Starting iteration 197. +[2026-03-26 00:52:33,826][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 19 and human policies 1. +[2026-03-26 00:52:33,827][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:52:37,824][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:52:51,240][__main__][INFO] - Number of regex retries in iteration 197: 1 +[2026-03-26 00:52:51,241][__main__][INFO] - agents played in iteration 197 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:52:52,011][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:52:52,034][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:52:52,057][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:52:52,081][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:52:52,082][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:52:52,083][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:52:52,715][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:52:53,172][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:52:53,675][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:52:54,174][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:52:54,677][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:52:55,180][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:52:55,682][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:52:56,182][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:52:56,684][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:52:57,185][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:52:57,685][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:52:58,185][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:52:58,686][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:52:59,182][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:52:59,683][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:53:00,183][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:53:00,686][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:53:01,192][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:53:01,697][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:53:02,194][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:53:02,697][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:53:03,201][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:53:03,704][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:53:04,210][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:53:04,711][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:53:05,210][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:53:05,713][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:53:06,215][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:53:06,713][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:53:07,212][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:53:07,714][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:53:08,216][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:53:08,712][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:53:09,212][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:53:09,711][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:53:10,210][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:53:10,707][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:53:11,207][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:53:11,705][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:53:12,205][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:53:12,700][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:53:13,196][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:53:13,696][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:53:14,195][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:53:14,693][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:53:15,193][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:53:15,690][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:53:16,190][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:53:16,688][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:53:17,188][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:53:17,688][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:53:18,189][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:53:18,688][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:53:19,190][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:53:19,692][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:53:20,192][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:53:20,692][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:53:21,188][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:53:21,685][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:53:22,182][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:53:22,683][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:53:23,182][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:53:23,681][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:53:24,177][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:53:24,673][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10445 tokens. +[2026-03-26 00:53:25,340][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.28%, ΔTime: 00:00:32 +[2026-03-26 00:53:26,079][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:53:26,081][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:53:26,082][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:53:26,784][__main__][INFO] - Iteration 198 took 52s (32.88% Gen, 65.79% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 15m 15s. Estimated total time: 44h 7m 54s. Time estimates for 10 more iterations: 8m 49s, 100 more iterations: 1h 28m 15s, 500 more iterations: 7h 21m 19s. +[2026-03-26 00:53:26,786][__main__][INFO] - Starting iteration 198. +[2026-03-26 00:53:27,185][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 19 and human policies 1. +[2026-03-26 00:53:27,185][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:53:39,628][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:53:44,195][__main__][INFO] - Number of regex retries in iteration 198: 1 +[2026-03-26 00:53:44,196][__main__][INFO] - agents played in iteration 198 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:53:44,964][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:53:44,987][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:53:45,010][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:53:45,033][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:53:45,034][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:53:45,034][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:53:45,670][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:53:46,127][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:53:46,630][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:53:47,132][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:53:47,634][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:53:48,135][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:53:48,636][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:53:49,137][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:53:49,638][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:53:50,140][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:53:50,641][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:53:51,138][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:53:51,635][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:53:52,133][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:53:52,633][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:53:53,132][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:53:53,633][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:53:54,136][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:53:54,638][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:53:55,145][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:53:55,648][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:53:56,154][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:53:56,652][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:53:57,159][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:53:57,665][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:53:58,169][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:53:58,670][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:53:59,169][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:53:59,671][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:54:00,174][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:54:00,676][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:54:01,180][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:54:01,684][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:54:02,184][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:54:02,681][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:54:03,178][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:54:03,674][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:54:04,173][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:54:04,669][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:54:05,166][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:54:05,664][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:54:06,164][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:54:06,664][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:54:07,164][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:54:07,665][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:54:08,165][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:54:08,666][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:54:09,166][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:54:09,666][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:54:10,161][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:54:10,661][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:54:11,161][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:54:11,660][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:54:12,161][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:54:12,658][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:54:13,157][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:54:13,660][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:54:14,160][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:54:14,660][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:54:15,161][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:54:15,662][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:54:16,163][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:54:16,666][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:54:17,165][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:54:17,665][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10576 tokens. +[2026-03-26 00:54:18,338][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.22%, Current % of VRAM taken: 60.65%, Block Peak % of device VRAM: 62.25%, ΔTime: 00:00:32 +[2026-03-26 00:54:19,078][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:54:19,082][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:54:19,084][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:54:19,785][__main__][INFO] - Iteration 199 took 52s (32.34% Gen, 66.33% Train). Generation: 17s, Training: 34s. Estimated remaining time: 40h 56m 31s. Estimated total time: 43h 50m 3s. Time estimates for 10 more iterations: 8m 46s, 100 more iterations: 1h 27m 40s, 500 more iterations: 7h 18m 20s. +[2026-03-26 00:54:19,787][__main__][INFO] - Starting iteration 199. +[2026-03-26 00:54:20,191][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 19 and human policies 1. +[2026-03-26 00:54:20,192][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:54:36,998][__main__][INFO] - Number of regex retries in iteration 199: 0 +[2026-03-26 00:54:36,999][__main__][INFO] - agents played in iteration 199 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:54:37,769][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:54:37,792][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:54:37,815][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:54:37,838][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:54:37,839][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:54:37,839][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:54:38,478][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:54:38,929][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:54:39,436][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:54:39,935][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:54:40,436][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:54:40,937][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:54:41,437][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:54:41,934][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:54:42,435][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:54:42,933][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:54:43,429][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:54:43,929][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:54:44,425][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:54:44,924][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:54:45,424][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:54:45,929][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:54:46,435][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:54:46,933][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:54:47,430][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:54:47,927][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:54:48,425][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:54:48,922][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:54:49,421][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:54:49,924][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:54:50,427][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:54:50,932][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:54:51,430][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:54:51,937][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:54:52,440][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:54:52,942][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:54:53,445][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:54:53,945][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:54:54,448][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:54:54,949][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:54:55,449][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:54:55,945][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:54:56,442][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:54:56,939][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:54:57,440][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:54:57,939][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:54:58,436][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:54:58,933][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:54:59,433][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:54:59,930][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:55:00,429][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:55:00,926][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:55:01,423][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:55:01,923][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:55:02,420][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:55:02,920][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:55:03,418][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:55:03,915][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:55:04,414][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:55:04,911][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:55:05,408][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:55:05,908][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:55:06,405][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:55:06,909][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:55:07,410][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:55:07,926][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:55:08,423][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:55:08,924][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:55:09,424][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:55:09,924][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:55:10,426][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10418 tokens. +[2026-03-26 00:55:11,091][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.25%, Current % of VRAM taken: 60.68%, Block Peak % of device VRAM: 62.22%, ΔTime: 00:00:32 +[2026-03-26 00:55:11,896][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:55:11,898][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:55:11,900][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:55:12,635][__main__][INFO] - Iteration 200 took 52s (32.05% Gen, 66.55% Train). Generation: 16s, Training: 34s. Estimated remaining time: 40h 47m 50s. Estimated total time: 43h 42m 14s. Time estimates for 10 more iterations: 8m 44s, 100 more iterations: 1h 27m 24s, 500 more iterations: 7h 17m 2s. +[2026-03-26 00:55:12,637][__main__][INFO] - Starting iteration 200. +[2026-03-26 00:55:13,037][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 19 and human policies 1. +[2026-03-26 00:55:13,038][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:55:17,236][mllm.models.large_language_model_local][WARNING] - Response Proposition: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:55:25,081][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 11 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 00:55:31,575][__main__][INFO] - Number of regex retries in iteration 200: 2 +[2026-03-26 00:55:31,576][__main__][INFO] - agents played in iteration 200 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:55:32,346][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:55:32,369][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:55:32,392][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:55:32,414][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:55:32,415][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:55:32,416][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:55:33,048][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:55:33,503][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:55:34,003][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:55:34,505][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:55:35,001][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:55:35,498][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:55:36,004][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:55:36,505][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:55:37,005][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:55:37,505][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:55:38,008][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:55:38,504][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:55:39,003][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:55:39,502][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:55:40,003][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:55:40,502][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:55:41,002][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:55:41,502][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:55:42,022][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:55:42,521][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:55:43,020][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:55:43,526][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:55:44,028][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:55:44,528][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:55:45,035][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:55:45,537][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:55:46,039][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:55:46,542][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:55:47,044][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:55:47,546][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:55:48,048][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:55:48,547][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:55:49,049][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:55:49,546][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:55:50,045][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:55:50,544][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:55:51,043][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:55:51,542][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:55:52,039][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:55:52,541][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:55:53,040][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:55:53,540][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:55:54,040][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:55:54,535][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:55:55,033][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:55:55,533][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:55:56,029][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:55:56,526][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:55:57,026][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:55:57,523][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:55:58,023][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:55:58,521][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:55:59,021][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:55:59,522][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:56:00,023][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:56:00,524][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:56:01,025][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:56:01,525][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:56:02,025][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:56:02,521][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:56:03,023][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:56:03,526][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:56:04,026][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:56:04,527][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:56:05,026][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10450 tokens. +[2026-03-26 00:56:05,693][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.15%, Current % of VRAM taken: 60.58%, Block Peak % of device VRAM: 62.23%, ΔTime: 00:00:32 +[2026-03-26 00:56:06,434][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:56:06,436][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:56:06,438][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:56:07,831][__main__][INFO] - Iteration 201 took 54s (33.83% Gen, 63.62% Train). Generation: 18s, Training: 34s. Estimated remaining time: 42h 44m 25s. Estimated total time: 45h 39m 44s. Time estimates for 10 more iterations: 9m 7s, 100 more iterations: 1h 31m 19s, 500 more iterations: 7h 36m 37s. +[2026-03-26 00:56:07,833][__main__][INFO] - Starting iteration 201. +[2026-03-26 00:56:08,233][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 20 and human policies 1. +[2026-03-26 00:56:08,234][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:56:24,676][__main__][INFO] - Number of regex retries in iteration 201: 0 +[2026-03-26 00:56:24,677][__main__][INFO] - agents played in iteration 201 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:56:25,443][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:56:25,467][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:56:25,490][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:56:25,513][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:56:25,514][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:56:25,515][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:56:26,152][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:56:26,614][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:56:27,121][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:56:27,632][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:56:28,133][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:56:28,633][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:56:29,132][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:56:29,629][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:56:30,131][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:56:30,628][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:56:31,127][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:56:31,625][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:56:32,122][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:56:32,620][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:56:33,117][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:56:33,614][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:56:34,111][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:56:34,608][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:56:35,107][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:56:35,603][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:56:36,101][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:56:36,599][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:56:37,101][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:56:37,602][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:56:38,107][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:56:38,606][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:56:39,110][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:56:39,610][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:56:40,111][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:56:40,613][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:56:41,115][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:56:41,614][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:56:42,116][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:56:42,618][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:56:43,120][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:56:43,621][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:56:44,124][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:56:44,625][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:56:45,127][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:56:45,627][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:56:46,129][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:56:46,627][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:56:47,128][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:56:47,630][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:56:48,132][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:56:48,632][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:56:49,132][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:56:49,630][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:56:50,131][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:56:50,635][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:56:51,139][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:56:51,637][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:56:52,139][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:56:52,639][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:56:53,137][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:56:53,639][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:56:54,141][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:56:54,641][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:56:55,142][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:56:55,643][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:56:56,141][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:56:56,644][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:56:57,145][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:56:57,644][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:56:58,146][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10478 tokens. +[2026-03-26 00:56:58,820][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.23%, Current % of VRAM taken: 60.66%, Block Peak % of device VRAM: 62.30%, ΔTime: 00:00:32 +[2026-03-26 00:56:59,793][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:56:59,795][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:56:59,797][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:57:00,583][__main__][INFO] - Iteration 202 took 52s (31.41% Gen, 67.09% Train). Generation: 16s, Training: 35s. Estimated remaining time: 40h 41m 18s. Estimated total time: 43h 37m 30s. Time estimates for 10 more iterations: 8m 43s, 100 more iterations: 1h 27m 15s, 500 more iterations: 7h 16m 15s. +[2026-03-26 00:57:00,585][__main__][INFO] - Starting iteration 202. +[2026-03-26 00:57:00,986][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 20 and human policies 1. +[2026-03-26 00:57:00,987][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:57:18,695][__main__][INFO] - Number of regex retries in iteration 202: 0 +[2026-03-26 00:57:18,696][__main__][INFO] - agents played in iteration 202 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:57:19,470][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:57:19,493][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:57:19,516][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:57:19,539][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:57:19,540][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:57:19,540][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:57:20,182][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:57:20,637][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:57:21,140][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:57:21,639][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:57:22,134][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:57:22,632][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:57:23,129][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:57:23,630][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:57:24,128][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:57:24,626][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:57:25,125][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:57:25,622][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:57:26,120][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:57:26,621][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:57:27,120][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:57:27,620][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:57:28,119][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:57:28,615][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:57:29,112][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:57:29,607][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:57:30,107][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:57:30,607][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:57:31,107][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:57:31,606][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:57:32,104][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:57:32,608][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:57:33,108][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:57:33,610][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:57:34,111][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:57:34,611][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:57:35,110][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:57:35,608][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:57:36,104][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:57:36,604][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:57:37,104][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:57:37,603][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:57:38,104][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:57:38,605][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:57:39,106][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:57:39,609][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:57:40,106][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:57:40,603][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:57:41,100][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:57:41,598][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:57:42,095][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:57:42,597][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:57:43,094][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:57:43,591][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:57:44,091][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:57:44,587][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:57:45,085][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:57:45,585][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:57:46,083][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:57:46,581][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:57:47,087][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:57:47,584][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:57:48,080][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:57:48,577][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:57:49,076][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:57:49,574][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:57:50,075][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:57:50,573][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:57:51,073][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:57:51,572][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:57:52,071][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10428 tokens. +[2026-03-26 00:57:52,748][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.16%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.21%, ΔTime: 00:00:32 +[2026-03-26 00:57:53,469][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:57:53,472][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:57:53,473][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:57:54,816][__main__][INFO] - Iteration 203 took 53s (32.90% Gen, 64.61% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 54m 23s. Estimated total time: 44h 51m 29s. Time estimates for 10 more iterations: 8m 58s, 100 more iterations: 1h 29m 42s, 500 more iterations: 7h 28m 34s. +[2026-03-26 00:57:54,818][__main__][INFO] - Starting iteration 203. +[2026-03-26 00:57:55,221][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 20 and human policies 1. +[2026-03-26 00:57:55,222][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:58:12,339][__main__][INFO] - Number of regex retries in iteration 203: 0 +[2026-03-26 00:58:12,340][__main__][INFO] - agents played in iteration 203 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:58:13,108][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:58:13,132][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:58:13,155][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:58:13,178][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:58:13,178][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:58:13,179][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:58:13,812][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:58:14,263][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:58:14,768][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:58:15,266][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:58:15,767][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:58:16,264][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:58:16,761][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:58:17,257][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:58:17,758][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:58:18,257][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:58:18,756][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:58:19,254][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:58:19,752][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:58:20,252][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:58:20,749][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:58:21,246][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:58:21,745][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:58:22,248][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:58:22,747][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:58:23,272][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:58:23,775][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:58:24,285][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:58:24,786][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:58:25,290][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:58:25,793][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:58:26,294][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:58:26,796][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:58:27,297][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:58:27,802][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:58:28,304][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:58:28,803][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:58:29,300][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:58:29,800][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:58:30,299][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:58:30,797][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:58:31,294][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:58:31,794][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:58:32,292][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:58:32,793][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:58:33,292][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:58:33,788][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:58:34,288][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:58:34,785][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:58:35,283][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:58:35,782][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:58:36,281][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:58:36,779][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:58:37,280][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:58:37,780][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:58:38,276][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:58:38,772][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:58:39,272][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:58:39,770][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:58:40,267][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:58:40,767][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:58:41,263][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:58:41,762][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:58:42,262][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:58:42,763][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:58:43,263][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:58:43,763][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:58:44,262][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:58:44,760][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:58:45,259][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:58:45,755][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10357 tokens. +[2026-03-26 00:58:46,424][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.16%, Current % of VRAM taken: 60.58%, Block Peak % of device VRAM: 62.31%, ΔTime: 00:00:32 +[2026-03-26 00:58:47,174][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:58:47,176][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:58:47,177][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:58:47,988][__main__][INFO] - Iteration 204 took 52s (32.44% Gen, 66.02% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 0m 24s. Estimated total time: 43h 58m 23s. Time estimates for 10 more iterations: 8m 47s, 100 more iterations: 1h 27m 56s, 500 more iterations: 7h 19m 43s. +[2026-03-26 00:58:47,990][__main__][INFO] - Starting iteration 204. +[2026-03-26 00:58:48,390][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 20 and human policies 1. +[2026-03-26 00:58:48,391][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:59:05,135][__main__][INFO] - Number of regex retries in iteration 204: 0 +[2026-03-26 00:59:05,136][__main__][INFO] - agents played in iteration 204 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:59:05,910][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:59:05,933][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:59:05,956][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:59:05,979][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:59:05,980][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:59:05,980][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 00:59:06,617][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 00:59:07,070][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 00:59:07,574][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 00:59:08,071][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 00:59:08,573][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 00:59:09,068][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 00:59:09,564][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 00:59:10,060][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 00:59:10,560][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 00:59:11,064][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 00:59:11,561][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 00:59:12,061][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 00:59:12,560][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 00:59:13,059][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 00:59:13,555][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 00:59:14,054][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 00:59:14,555][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 00:59:15,050][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 00:59:15,549][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 00:59:16,049][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 00:59:16,551][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 00:59:17,052][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 00:59:17,554][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 00:59:18,052][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 00:59:18,555][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 00:59:19,055][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 00:59:19,558][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 00:59:20,060][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 00:59:20,580][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 00:59:21,087][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 00:59:21,589][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 00:59:22,088][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 00:59:22,588][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 00:59:23,086][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 00:59:23,588][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 00:59:24,086][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 00:59:24,587][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 00:59:25,086][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 00:59:25,585][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 00:59:26,084][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 00:59:26,583][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 00:59:27,082][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 00:59:27,581][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 00:59:28,082][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 00:59:28,583][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 00:59:29,081][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 00:59:29,578][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 00:59:30,079][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 00:59:30,579][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 00:59:31,075][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 00:59:31,574][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 00:59:32,071][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 00:59:32,570][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 00:59:33,071][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 00:59:33,570][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 00:59:34,070][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 00:59:34,572][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 00:59:35,069][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 00:59:35,568][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 00:59:36,065][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 00:59:36,563][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 00:59:37,061][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 00:59:37,558][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 00:59:38,054][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 00:59:38,550][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10341 tokens. +[2026-03-26 00:59:39,224][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.23%, ΔTime: 00:00:32 +[2026-03-26 00:59:39,945][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 00:59:39,947][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 00:59:39,949][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 00:59:40,674][__main__][INFO] - Iteration 205 took 52s (32.03% Gen, 66.58% Train). Generation: 16s, Training: 34s. Estimated remaining time: 40h 35m 20s. Estimated total time: 43h 34m 13s. Time estimates for 10 more iterations: 8m 42s, 100 more iterations: 1h 27m 8s, 500 more iterations: 7h 15m 42s. +[2026-03-26 00:59:40,676][__main__][INFO] - Starting iteration 205. +[2026-03-26 00:59:41,084][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 20 and human policies 1. +[2026-03-26 00:59:41,085][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 00:59:58,677][__main__][INFO] - Number of regex retries in iteration 205: 0 +[2026-03-26 00:59:58,678][__main__][INFO] - agents played in iteration 205 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 00:59:59,448][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:59:59,471][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:59:59,494][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:59:59,517][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 00:59:59,517][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 00:59:59,518][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:00:00,156][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:00:00,608][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:00:01,115][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:00:01,613][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:00:02,116][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:00:02,619][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:00:03,116][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:00:03,618][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:00:04,117][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:00:04,615][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:00:05,113][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:00:05,615][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:00:06,113][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:00:06,614][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:00:07,114][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:00:07,616][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:00:08,119][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:00:08,618][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:00:09,116][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:00:09,615][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:00:10,115][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:00:10,613][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:00:11,116][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:00:11,616][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:00:12,119][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:00:12,624][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:00:13,123][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:00:13,624][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:00:14,125][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:00:14,625][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:00:15,126][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:00:15,628][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:00:16,127][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:00:16,625][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:00:17,125][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:00:17,624][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:00:18,122][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:00:18,622][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:00:19,119][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:00:19,619][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:00:20,119][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:00:20,621][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:00:21,119][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:00:21,621][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:00:22,119][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:00:22,620][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:00:23,119][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:00:23,620][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:00:24,121][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:00:24,618][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:00:25,116][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:00:25,613][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:00:26,110][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:00:26,609][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:00:27,108][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:00:27,608][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:00:28,105][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:00:28,603][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:00:29,100][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:00:29,597][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:00:30,093][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:00:30,592][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:00:31,090][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:00:31,586][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:00:32,085][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10369 tokens. +[2026-03-26 01:00:32,751][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.03%, ΔTime: 00:00:32 +[2026-03-26 01:00:33,816][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:00:33,819][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:00:33,820][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:00:34,536][__main__][INFO] - Iteration 206 took 53s (32.91% Gen, 65.75% Train). Generation: 17s, Training: 35s. Estimated remaining time: 41h 32m 52s. Estimated total time: 44h 32m 38s. Time estimates for 10 more iterations: 8m 54s, 100 more iterations: 1h 29m 5s, 500 more iterations: 7h 25m 26s. +[2026-03-26 01:00:34,539][__main__][INFO] - Starting iteration 206. +[2026-03-26 01:00:34,946][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 20 and human policies 1. +[2026-03-26 01:00:34,947][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:00:52,486][__main__][INFO] - Number of regex retries in iteration 206: 0 +[2026-03-26 01:00:52,487][__main__][INFO] - agents played in iteration 206 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:00:53,264][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:00:53,287][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:00:53,310][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:00:53,332][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:00:53,333][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:00:53,334][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:00:53,972][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:00:54,425][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:00:54,929][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:00:55,427][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:00:55,923][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:00:56,419][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:00:56,917][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:00:57,413][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:00:57,908][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:00:58,405][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:00:58,901][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:00:59,402][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:00:59,898][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:01:00,396][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:01:00,892][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:01:01,389][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:01:01,886][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:01:02,382][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:01:02,878][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:01:03,374][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:01:03,874][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:01:04,371][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:01:04,871][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:01:05,370][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:01:05,868][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:01:06,367][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:01:06,873][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:01:07,373][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:01:07,873][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:01:08,377][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:01:08,880][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:01:09,381][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:01:09,879][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:01:10,375][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:01:10,872][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:01:11,370][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:01:11,869][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:01:12,365][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:01:12,863][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:01:13,361][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:01:13,861][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:01:14,357][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:01:14,878][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:01:15,375][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:01:15,876][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:01:16,371][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:01:16,869][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:01:17,371][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:01:17,873][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:01:18,374][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:01:18,875][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:01:19,375][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:01:19,875][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:01:20,376][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:01:20,876][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:01:21,379][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:01:21,879][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:01:22,382][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:01:22,882][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:01:23,380][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:01:23,881][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:01:24,381][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:01:24,882][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:01:25,382][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:01:25,884][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10264 tokens. +[2026-03-26 01:01:26,558][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.23%, Current % of VRAM taken: 60.66%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 01:01:27,420][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:01:27,423][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:01:27,424][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:01:28,422][__main__][INFO] - Iteration 207 took 53s (32.80% Gen, 65.33% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 33m 11s. Estimated total time: 44h 33m 51s. Time estimates for 10 more iterations: 8m 54s, 100 more iterations: 1h 29m 7s, 500 more iterations: 7h 25m 38s. +[2026-03-26 01:01:28,424][__main__][INFO] - Starting iteration 207. +[2026-03-26 01:01:28,824][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 20 and human policies 1. +[2026-03-26 01:01:28,824][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:01:46,571][__main__][INFO] - Number of regex retries in iteration 207: 0 +[2026-03-26 01:01:46,572][__main__][INFO] - agents played in iteration 207 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:01:47,342][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:01:47,365][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:01:47,388][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:01:47,411][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:01:47,412][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:01:47,412][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:01:48,049][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:01:48,504][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:01:49,007][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:01:49,507][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:01:50,005][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:01:50,504][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:01:51,000][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:01:51,499][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:01:51,996][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:01:52,498][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:01:52,995][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:01:53,499][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:01:54,000][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:01:54,499][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:01:55,001][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:01:55,501][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:01:56,003][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:01:56,503][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:01:57,023][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:01:57,521][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:01:58,025][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:01:58,529][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:01:59,028][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:01:59,529][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:02:00,028][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:02:00,535][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:02:01,037][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:02:01,542][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:02:02,041][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:02:02,544][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:02:03,046][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:02:03,545][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:02:04,045][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:02:04,545][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:02:05,046][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:02:05,545][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:02:06,045][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:02:06,545][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:02:07,043][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:02:07,543][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:02:08,042][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:02:08,539][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:02:09,040][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:02:09,539][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:02:10,037][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:02:10,537][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:02:11,037][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:02:11,537][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:02:12,037][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:02:12,533][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:02:13,031][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:02:13,527][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:02:14,024][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:02:14,520][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:02:15,019][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:02:15,519][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:02:16,017][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:02:16,517][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:02:17,017][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:02:17,521][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:02:18,021][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:02:18,522][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:02:19,022][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:02:19,522][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:02:20,022][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10430 tokens. +[2026-03-26 01:02:20,698][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.61%, Block Peak % of device VRAM: 62.41%, ΔTime: 00:00:32 +[2026-03-26 01:02:21,454][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:02:21,456][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:02:21,458][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:02:22,187][__main__][INFO] - Iteration 208 took 53s (33.26% Gen, 65.37% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 26m 39s. Estimated total time: 44h 28m 12s. Time estimates for 10 more iterations: 8m 53s, 100 more iterations: 1h 28m 56s, 500 more iterations: 7h 24m 42s. +[2026-03-26 01:02:22,190][__main__][INFO] - Starting iteration 208. +[2026-03-26 01:02:22,592][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 20 and human policies 1. +[2026-03-26 01:02:22,593][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:02:39,597][__main__][INFO] - Number of regex retries in iteration 208: 0 +[2026-03-26 01:02:39,598][__main__][INFO] - agents played in iteration 208 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:02:40,364][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:02:40,387][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:02:40,410][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:02:40,432][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:02:40,433][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:02:40,434][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:02:41,078][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:02:41,528][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:02:42,031][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:02:42,531][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:02:43,031][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:02:43,528][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:02:44,026][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:02:44,523][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:02:45,020][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:02:45,526][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:02:46,025][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:02:46,525][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:02:47,023][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:02:47,524][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:02:48,027][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:02:48,526][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:02:49,026][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:02:49,525][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:02:50,022][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:02:50,522][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:02:51,018][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:02:51,515][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:02:52,015][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:02:52,517][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:02:53,019][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:02:53,522][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:02:54,023][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:02:54,548][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:02:55,052][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:02:55,555][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:02:56,058][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:02:56,557][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:02:57,060][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:02:57,558][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:02:58,059][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:02:58,555][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:02:59,052][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:02:59,549][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:03:00,046][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:03:00,545][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:03:01,043][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:03:01,541][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:03:02,042][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:03:02,538][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:03:03,041][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:03:03,541][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:03:04,040][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:03:04,540][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:03:05,041][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:03:05,540][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:03:06,042][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:03:06,542][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:03:07,039][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:03:07,537][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:03:08,034][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:03:08,536][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:03:09,034][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:03:09,535][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:03:10,034][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:03:10,531][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:03:11,030][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:03:11,530][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:03:12,027][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:03:12,525][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:03:13,025][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10367 tokens. +[2026-03-26 01:03:13,699][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.20%, Current % of VRAM taken: 60.62%, Block Peak % of device VRAM: 62.29%, ΔTime: 00:00:32 +[2026-03-26 01:03:14,463][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:03:14,466][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:03:14,467][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:03:15,209][__main__][INFO] - Iteration 209 took 52s (32.32% Gen, 66.27% Train). Generation: 17s, Training: 34s. Estimated remaining time: 40h 48m 27s. Estimated total time: 43h 50m 54s. Time estimates for 10 more iterations: 8m 46s, 100 more iterations: 1h 27m 41s, 500 more iterations: 7h 18m 29s. +[2026-03-26 01:03:15,211][__main__][INFO] - Starting iteration 209. +[2026-03-26 01:03:15,611][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 20 and human policies 1. +[2026-03-26 01:03:15,612][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:03:33,229][__main__][INFO] - Number of regex retries in iteration 209: 0 +[2026-03-26 01:03:33,230][__main__][INFO] - agents played in iteration 209 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:03:33,998][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:03:34,022][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:03:34,045][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:03:34,067][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:03:34,068][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:03:34,068][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:03:34,706][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:03:35,158][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:03:35,664][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:03:36,166][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:03:36,670][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:03:37,170][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:03:37,686][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:03:38,193][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:03:38,692][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:03:39,193][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:03:39,695][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:03:40,194][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:03:40,694][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:03:41,194][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:03:41,694][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:03:42,194][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:03:42,692][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:03:43,190][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:03:43,688][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:03:44,187][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:03:44,686][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:03:45,189][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:03:45,691][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:03:46,193][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:03:46,693][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:03:47,196][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:03:47,696][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:03:48,200][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:03:48,705][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:03:49,206][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:03:49,709][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:03:50,208][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:03:50,709][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:03:51,208][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:03:51,707][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:03:52,208][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:03:52,705][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:03:53,205][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:03:53,703][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:03:54,205][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:03:54,703][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:03:55,203][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:03:55,704][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:03:56,203][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:03:56,701][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:03:57,199][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:03:57,700][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:03:58,202][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:03:58,704][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:03:59,206][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:03:59,705][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:04:00,206][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:04:00,713][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:04:01,217][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:04:01,720][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:04:02,220][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:04:02,720][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:04:03,221][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:04:03,720][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:04:04,219][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:04:04,721][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:04:05,218][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:04:05,715][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:04:06,216][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:04:06,715][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10364 tokens. +[2026-03-26 01:04:07,386][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.19%, Current % of VRAM taken: 60.62%, Block Peak % of device VRAM: 62.34%, ΔTime: 00:00:32 +[2026-03-26 01:04:08,120][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:04:08,122][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:04:08,124][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:04:08,907][__main__][INFO] - Iteration 210 took 53s (33.06% Gen, 65.47% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 21m 28s. Estimated total time: 44h 24m 48s. Time estimates for 10 more iterations: 8m 52s, 100 more iterations: 1h 28m 49s, 500 more iterations: 7h 24m 8s. +[2026-03-26 01:04:08,909][__main__][INFO] - Starting iteration 210. +[2026-03-26 01:04:09,311][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 20 and human policies 1. +[2026-03-26 01:04:09,312][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:04:13,636][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:04:27,125][__main__][INFO] - Number of regex retries in iteration 210: 1 +[2026-03-26 01:04:27,126][__main__][INFO] - agents played in iteration 210 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:04:27,894][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:04:27,917][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:04:27,940][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:04:27,963][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:04:27,964][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:04:27,964][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:04:28,597][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:04:29,053][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:04:29,559][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:04:30,058][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:04:30,558][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:04:31,054][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:04:31,555][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:04:32,054][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:04:32,556][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:04:33,052][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:04:33,553][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:04:34,051][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:04:34,547][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:04:35,044][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:04:35,544][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:04:36,044][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:04:36,541][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:04:37,044][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:04:37,544][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:04:38,047][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:04:38,550][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:04:39,054][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:04:39,554][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:04:40,057][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:04:40,557][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:04:41,056][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:04:41,557][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:04:42,057][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:04:42,557][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:04:43,056][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:04:43,553][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:04:44,051][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:04:44,548][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:04:45,048][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:04:45,546][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:04:46,044][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:04:46,541][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:04:47,039][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:04:47,537][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:04:48,038][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:04:48,536][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:04:49,033][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:04:49,530][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:04:50,028][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:04:50,526][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:04:51,024][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:04:51,520][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:04:52,017][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:04:52,514][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:04:53,014][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:04:53,514][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:04:54,015][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:04:54,515][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:04:55,017][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:04:55,532][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:04:56,031][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:04:56,527][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:04:57,027][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:04:57,531][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:04:58,030][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:04:58,527][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:04:59,026][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:04:59,526][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:05:00,023][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:05:00,523][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10237 tokens. +[2026-03-26 01:05:01,183][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.23%, Current % of VRAM taken: 60.66%, Block Peak % of device VRAM: 62.22%, ΔTime: 00:00:32 +[2026-03-26 01:05:01,955][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:05:01,957][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:05:01,959][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:05:03,237][__main__][INFO] - Iteration 211 took 53s (33.04% Gen, 64.59% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 52m 4s. Estimated total time: 44h 56m 19s. Time estimates for 10 more iterations: 8m 59s, 100 more iterations: 1h 29m 52s, 500 more iterations: 7h 29m 23s. +[2026-03-26 01:05:03,239][__main__][INFO] - Starting iteration 211. +[2026-03-26 01:05:03,707][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 21 and human policies 1. +[2026-03-26 01:05:03,708][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:05:07,566][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:05:20,891][__main__][INFO] - Number of regex retries in iteration 211: 1 +[2026-03-26 01:05:20,892][__main__][INFO] - agents played in iteration 211 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:05:21,658][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:05:21,682][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:05:21,705][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:05:21,727][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:05:21,728][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:05:21,728][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:05:22,371][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:05:22,821][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:05:23,326][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:05:23,823][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:05:24,321][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:05:24,818][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:05:25,315][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:05:25,813][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:05:26,312][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:05:26,807][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:05:27,303][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:05:27,804][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:05:28,301][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:05:28,797][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:05:29,293][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:05:29,789][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:05:30,287][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:05:30,788][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:05:31,292][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:05:31,799][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:05:32,299][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:05:32,797][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:05:33,301][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:05:33,799][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:05:34,301][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:05:34,801][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:05:35,301][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:05:35,801][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:05:36,300][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:05:36,801][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:05:37,297][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:05:37,793][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:05:38,292][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:05:38,791][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:05:39,288][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:05:39,785][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:05:40,281][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:05:40,779][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:05:41,275][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:05:41,773][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:05:42,270][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:05:42,790][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:05:43,291][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:05:43,788][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:05:44,286][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:05:44,786][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:05:45,286][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:05:45,786][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:05:46,287][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:05:46,786][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:05:47,283][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:05:47,783][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:05:48,282][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:05:48,779][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:05:49,278][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:05:49,781][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:05:50,280][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:05:50,779][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:05:51,274][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:05:51,772][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:05:52,273][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:05:52,776][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:05:53,275][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:05:53,771][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:05:54,267][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10241 tokens. +[2026-03-26 01:05:54,937][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 62.17%, ΔTime: 00:00:32 +[2026-03-26 01:05:55,702][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:05:55,704][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:05:55,706][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:05:56,480][__main__][INFO] - Iteration 212 took 52s (32.56% Gen, 65.97% Train). Generation: 17s, Training: 34s. Estimated remaining time: 40h 53m 32s. Estimated total time: 43h 58m 39s. Time estimates for 10 more iterations: 8m 47s, 100 more iterations: 1h 27m 57s, 500 more iterations: 7h 19m 46s. +[2026-03-26 01:05:56,482][__main__][INFO] - Starting iteration 212. +[2026-03-26 01:05:56,884][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 21 and human policies 1. +[2026-03-26 01:05:56,885][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:06:13,434][__main__][INFO] - Number of regex retries in iteration 212: 0 +[2026-03-26 01:06:13,435][__main__][INFO] - agents played in iteration 212 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:06:14,202][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:06:14,225][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:06:14,248][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:06:14,271][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:06:14,271][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:06:14,272][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:06:14,905][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:06:15,355][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:06:15,858][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:06:16,354][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:06:16,850][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:06:17,347][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:06:17,844][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:06:18,343][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:06:18,841][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:06:19,341][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:06:19,838][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:06:20,338][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:06:20,834][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:06:21,330][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:06:21,827][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:06:22,323][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:06:22,820][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:06:23,316][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:06:23,821][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:06:24,324][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:06:24,829][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:06:25,331][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:06:25,828][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:06:26,330][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:06:26,829][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:06:27,331][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:06:27,853][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:06:28,355][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:06:28,858][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:06:29,360][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:06:29,860][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:06:30,360][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:06:30,860][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:06:31,356][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:06:31,855][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:06:32,352][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:06:32,849][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:06:33,346][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:06:33,842][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:06:34,344][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:06:34,841][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:06:35,339][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:06:35,836][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:06:36,334][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:06:36,832][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:06:37,330][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:06:37,829][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:06:38,326][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:06:38,826][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:06:39,324][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:06:39,823][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:06:40,321][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:06:40,821][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:06:41,322][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:06:41,822][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:06:42,320][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:06:42,817][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:06:43,313][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:06:43,815][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:06:44,316][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:06:44,818][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:06:45,315][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:06:45,814][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:06:46,310][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:06:46,807][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10321 tokens. +[2026-03-26 01:06:47,479][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.16%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 01:06:48,224][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:06:48,227][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:06:48,228][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:06:48,888][__main__][INFO] - Iteration 213 took 52s (31.82% Gen, 66.90% Train). Generation: 16s, Training: 34s. Estimated remaining time: 40h 14m 17s. Estimated total time: 43h 20m 17s. Time estimates for 10 more iterations: 8m 40s, 100 more iterations: 1h 26m 40s, 500 more iterations: 7h 13m 22s. +[2026-03-26 01:06:48,891][__main__][INFO] - Starting iteration 213. +[2026-03-26 01:06:49,313][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 21 and human policies 1. +[2026-03-26 01:06:49,314][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:06:53,423][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:07:06,803][__main__][INFO] - Number of regex retries in iteration 213: 1 +[2026-03-26 01:07:06,804][__main__][INFO] - agents played in iteration 213 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:07:07,570][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:07:07,593][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:07:07,617][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:07:07,640][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:07:07,640][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:07:07,641][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:07:08,280][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:07:08,731][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:07:09,233][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:07:09,730][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:07:10,229][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:07:10,728][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:07:11,226][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:07:11,722][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:07:12,219][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:07:12,718][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:07:13,214][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:07:13,711][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:07:14,209][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:07:14,713][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:07:15,210][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:07:15,709][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:07:16,207][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:07:16,705][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:07:17,202][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:07:17,707][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:07:18,213][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:07:18,713][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:07:19,211][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:07:19,711][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:07:20,211][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:07:20,712][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:07:21,213][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:07:21,738][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:07:22,240][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:07:22,743][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:07:23,245][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:07:23,742][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:07:24,240][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:07:24,736][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:07:25,235][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:07:25,731][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:07:26,231][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:07:26,731][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:07:27,231][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:07:27,729][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:07:28,226][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:07:28,727][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:07:29,227][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:07:29,725][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:07:30,226][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:07:30,725][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:07:31,225][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:07:31,724][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:07:32,223][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:07:32,723][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:07:33,226][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:07:33,726][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:07:34,239][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:07:34,742][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:07:35,242][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:07:35,738][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:07:36,239][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:07:36,739][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:07:37,237][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:07:37,737][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:07:38,235][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:07:38,734][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:07:39,230][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:07:39,727][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:07:40,226][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10182 tokens. +[2026-03-26 01:07:40,905][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 01:07:41,659][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:07:41,662][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:07:41,664][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:07:42,385][__main__][INFO] - Iteration 214 took 53s (32.95% Gen, 65.68% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 6m 44s. Estimated total time: 44h 13m 38s. Time estimates for 10 more iterations: 8m 50s, 100 more iterations: 1h 28m 27s, 500 more iterations: 7h 22m 16s. +[2026-03-26 01:07:42,388][__main__][INFO] - Starting iteration 214. +[2026-03-26 01:07:42,788][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 21 and human policies 1. +[2026-03-26 01:07:42,789][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:08:02,032][__main__][INFO] - Number of regex retries in iteration 214: 0 +[2026-03-26 01:08:02,033][__main__][INFO] - agents played in iteration 214 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:08:02,800][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:08:02,823][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:08:02,846][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:08:02,869][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:08:02,870][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:08:02,870][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:08:03,510][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:08:03,967][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:08:04,469][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:08:04,968][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:08:05,468][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:08:05,965][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:08:06,466][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:08:06,965][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:08:07,463][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:08:07,962][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:08:08,462][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:08:08,958][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:08:09,456][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:08:09,955][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:08:10,451][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:08:10,947][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:08:11,449][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:08:11,955][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:08:12,458][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:08:12,959][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:08:13,464][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:08:13,966][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:08:14,466][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:08:14,968][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:08:15,470][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:08:15,974][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:08:16,472][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:08:16,972][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:08:17,472][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:08:17,973][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:08:18,475][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:08:18,977][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:08:19,478][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:08:19,975][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:08:20,474][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:08:20,972][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:08:21,471][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:08:21,971][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:08:22,470][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:08:22,970][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:08:23,469][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:08:23,968][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:08:24,468][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:08:24,967][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:08:25,469][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:08:25,968][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:08:26,468][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:08:26,969][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:08:27,467][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:08:27,967][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:08:28,472][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:08:28,975][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:08:29,480][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:08:29,985][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:08:30,488][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:08:30,988][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:08:31,485][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:08:31,983][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:08:32,484][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:08:32,981][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:08:33,478][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:08:33,976][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:08:34,473][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:08:34,968][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:08:35,469][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10281 tokens. +[2026-03-26 01:08:36,148][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.08%, ΔTime: 00:00:32 +[2026-03-26 01:08:36,896][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:08:36,899][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:08:36,901][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:08:37,731][__main__][INFO] - Iteration 215 took 54s (35.03% Gen, 63.46% Train). Generation: 19s, Training: 34s. Estimated remaining time: 42h 39m 20s. Estimated total time: 45h 47m 9s. Time estimates for 10 more iterations: 9m 9s, 100 more iterations: 1h 31m 34s, 500 more iterations: 7h 37m 51s. +[2026-03-26 01:08:37,733][__main__][INFO] - Starting iteration 215. +[2026-03-26 01:08:38,137][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 21 and human policies 1. +[2026-03-26 01:08:38,138][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:08:55,125][__main__][INFO] - Number of regex retries in iteration 215: 0 +[2026-03-26 01:08:55,126][__main__][INFO] - agents played in iteration 215 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:08:55,927][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:08:55,951][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:08:55,975][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:08:55,998][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:08:55,998][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:08:55,999][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:08:56,595][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:08:57,051][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:08:57,558][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:08:58,056][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:08:58,558][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:08:59,056][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:08:59,554][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:09:00,051][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:09:00,552][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:09:01,050][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:09:01,550][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:09:02,049][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:09:02,550][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:09:03,053][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:09:03,554][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:09:04,052][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:09:04,553][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:09:05,058][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:09:05,557][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:09:06,061][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:09:06,564][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:09:07,065][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:09:07,569][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:09:08,067][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:09:08,566][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:09:09,067][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:09:09,568][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:09:10,073][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:09:10,574][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:09:11,073][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:09:11,574][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:09:12,074][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:09:12,574][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:09:13,076][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:09:13,578][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:09:14,079][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:09:14,582][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:09:15,083][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:09:15,583][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:09:16,084][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:09:16,584][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:09:17,080][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:09:17,577][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:09:18,077][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:09:18,574][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:09:19,073][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:09:19,572][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:09:20,072][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:09:20,572][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:09:21,069][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:09:21,567][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:09:22,064][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:09:22,566][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:09:23,070][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:09:23,568][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:09:24,067][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:09:24,563][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:09:25,059][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:09:25,554][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:09:26,053][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:09:26,548][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:09:27,045][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:09:27,540][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:09:28,042][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:09:28,537][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10220 tokens. +[2026-03-26 01:09:29,228][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.04%, Current % of VRAM taken: 60.47%, Block Peak % of device VRAM: 62.21%, ΔTime: 00:00:32 +[2026-03-26 01:09:30,003][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:09:30,005][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:09:30,007][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:09:30,848][__main__][INFO] - Iteration 216 took 52s (32.23% Gen, 66.17% Train). Generation: 16s, Training: 34s. Estimated remaining time: 40h 46m 53s. Estimated total time: 43h 55m 35s. Time estimates for 10 more iterations: 8m 47s, 100 more iterations: 1h 27m 51s, 500 more iterations: 7h 19m 15s. +[2026-03-26 01:09:30,851][__main__][INFO] - Starting iteration 216. +[2026-03-26 01:09:31,265][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 21 and human policies 1. +[2026-03-26 01:09:31,265][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:09:48,794][__main__][INFO] - Number of regex retries in iteration 216: 0 +[2026-03-26 01:09:48,795][__main__][INFO] - agents played in iteration 216 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:09:49,584][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:09:49,607][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:09:49,630][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:09:49,653][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:09:49,654][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:09:49,654][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:09:50,314][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:09:50,769][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:09:51,272][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:09:51,767][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:09:52,266][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:09:52,767][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:09:53,268][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:09:53,767][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:09:54,266][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:09:54,765][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:09:55,264][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:09:55,767][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:09:56,272][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:09:56,775][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:09:57,277][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:09:57,777][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:09:58,280][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:09:58,782][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:09:59,286][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:09:59,785][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:10:00,284][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:10:00,784][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:10:01,286][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:10:01,786][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:10:02,284][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:10:02,785][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:10:03,284][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:10:03,784][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:10:04,282][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:10:04,779][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:10:05,276][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:10:05,776][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:10:06,278][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:10:06,776][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:10:07,274][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:10:07,771][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:10:08,269][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:10:08,769][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:10:09,267][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:10:09,767][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:10:10,265][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:10:10,763][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:10:11,261][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:10:11,758][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:10:12,260][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:10:12,764][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:10:13,266][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:10:13,766][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:10:14,263][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:10:14,762][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:10:15,262][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:10:15,760][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:10:16,257][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:10:16,752][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:10:17,249][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:10:17,748][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:10:18,247][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:10:18,749][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:10:19,247][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:10:19,746][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:10:20,244][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:10:20,744][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:10:21,241][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:10:21,740][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:10:22,240][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10280 tokens. +[2026-03-26 01:10:22,929][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.15%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.31%, ΔTime: 00:00:32 +[2026-03-26 01:10:23,699][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:10:23,701][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:10:23,703][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:10:24,546][__main__][INFO] - Iteration 217 took 53s (32.90% Gen, 65.52% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 14m 29s. Estimated total time: 44h 24m 4s. Time estimates for 10 more iterations: 8m 52s, 100 more iterations: 1h 28m 48s, 500 more iterations: 7h 24m 0s. +[2026-03-26 01:10:24,548][__main__][INFO] - Starting iteration 217. +[2026-03-26 01:10:24,955][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 21 and human policies 1. +[2026-03-26 01:10:24,956][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:10:30,000][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:10:42,926][__main__][INFO] - Number of regex retries in iteration 217: 1 +[2026-03-26 01:10:42,927][__main__][INFO] - agents played in iteration 217 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:10:43,742][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:10:43,765][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:10:43,788][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:10:43,810][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:10:43,811][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:10:43,812][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:10:44,462][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:10:44,918][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:10:45,419][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:10:45,916][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:10:46,414][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:10:46,911][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:10:47,408][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:10:47,905][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:10:48,401][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:10:48,904][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:10:49,410][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:10:49,910][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:10:50,408][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:10:50,908][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:10:51,408][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:10:51,910][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:10:52,410][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:10:52,914][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:10:53,417][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:10:53,918][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:10:54,416][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:10:54,916][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:10:55,416][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:10:55,914][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:10:56,414][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:10:56,911][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:10:57,409][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:10:57,909][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:10:58,410][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:10:58,908][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:10:59,407][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:10:59,908][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:11:00,406][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:11:00,906][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:11:01,407][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:11:01,906][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:11:02,403][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:11:02,902][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:11:03,401][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:11:03,901][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:11:04,407][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:11:04,915][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:11:05,413][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:11:05,910][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:11:06,407][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:11:06,905][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:11:07,405][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:11:07,902][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:11:08,399][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:11:08,895][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:11:09,392][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:11:09,887][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:11:10,384][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:11:10,882][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:11:11,379][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:11:11,875][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:11:12,371][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:11:12,868][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:11:13,367][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:11:13,865][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:11:14,364][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:11:14,866][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:11:15,366][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:11:15,864][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:11:16,362][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10197 tokens. +[2026-03-26 01:11:17,061][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 01:11:17,803][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:11:17,805][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:11:17,807][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:11:18,641][__main__][INFO] - Iteration 218 took 53s (33.47% Gen, 64.97% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 33m 50s. Estimated total time: 44h 44m 20s. Time estimates for 10 more iterations: 8m 56s, 100 more iterations: 1h 29m 28s, 500 more iterations: 7h 27m 23s. +[2026-03-26 01:11:18,643][__main__][INFO] - Starting iteration 218. +[2026-03-26 01:11:19,044][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 21 and human policies 1. +[2026-03-26 01:11:19,045][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:11:24,917][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:11:37,202][__main__][INFO] - Number of regex retries in iteration 218: 1 +[2026-03-26 01:11:37,203][__main__][INFO] - agents played in iteration 218 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:11:37,991][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:11:38,014][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:11:38,037][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:11:38,059][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:11:38,060][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:11:38,061][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:11:38,700][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:11:39,152][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:11:39,652][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:11:40,153][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:11:40,652][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:11:41,156][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:11:41,653][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:11:42,157][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:11:42,655][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:11:43,159][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:11:43,664][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:11:44,167][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:11:44,668][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:11:45,170][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:11:45,694][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:11:46,195][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:11:46,695][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:11:47,197][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:11:47,701][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:11:48,203][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:11:48,702][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:11:49,202][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:11:49,700][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:11:50,201][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:11:50,702][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:11:51,199][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:11:51,697][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:11:52,195][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:11:52,695][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:11:53,193][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:11:53,693][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:11:54,189][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:11:54,687][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:11:55,185][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:11:55,687][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:11:56,191][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:11:56,696][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:11:57,193][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:11:57,690][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:11:58,188][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:11:58,690][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:11:59,191][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:11:59,688][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:12:00,188][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:12:00,688][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:12:01,187][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:12:01,687][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:12:02,188][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:12:02,686][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:12:03,186][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:12:03,683][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:12:04,180][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:12:04,681][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:12:05,177][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:12:05,678][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:12:06,175][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:12:06,674][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:12:07,172][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:12:07,669][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:12:08,165][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:12:08,661][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:12:09,157][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:12:09,653][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:12:10,151][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:12:10,647][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10338 tokens. +[2026-03-26 01:12:11,323][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.18%, ΔTime: 00:00:32 +[2026-03-26 01:12:12,061][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:12:12,064][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:12:12,066][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:12:12,687][__main__][INFO] - Iteration 219 took 53s (33.85% Gen, 64.99% Train). Generation: 18s, Training: 34s. Estimated remaining time: 41h 30m 50s. Estimated total time: 44h 42m 14s. Time estimates for 10 more iterations: 8m 56s, 100 more iterations: 1h 29m 24s, 500 more iterations: 7h 27m 2s. +[2026-03-26 01:12:12,690][__main__][INFO] - Starting iteration 219. +[2026-03-26 01:12:13,093][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 21 and human policies 1. +[2026-03-26 01:12:13,094][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:12:30,007][__main__][INFO] - Number of regex retries in iteration 219: 0 +[2026-03-26 01:12:30,008][__main__][INFO] - agents played in iteration 219 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:12:30,788][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:12:30,811][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:12:30,835][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:12:30,857][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:12:30,858][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:12:30,858][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:12:31,501][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:12:31,953][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:12:32,457][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:12:32,957][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:12:33,454][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:12:33,953][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:12:34,453][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:12:34,951][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:12:35,454][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:12:35,956][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:12:36,459][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:12:36,958][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:12:37,456][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:12:37,956][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:12:38,456][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:12:38,955][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:12:39,455][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:12:39,955][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:12:40,453][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:12:40,954][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:12:41,450][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:12:41,950][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:12:42,448][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:12:42,948][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:12:43,444][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:12:43,942][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:12:44,445][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:12:44,947][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:12:45,445][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:12:45,947][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:12:46,447][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:12:46,944][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:12:47,447][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:12:47,952][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:12:48,458][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:12:48,957][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:12:49,458][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:12:49,955][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:12:50,454][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:12:50,955][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:12:51,455][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:12:51,954][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:12:52,454][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:12:52,952][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:12:53,452][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:12:53,949][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:12:54,448][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:12:54,948][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:12:55,446][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:12:55,950][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:12:56,454][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:12:56,954][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:12:57,455][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:12:57,954][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:12:58,451][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:12:58,951][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:12:59,454][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:12:59,952][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:13:00,450][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:13:00,947][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:13:01,446][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:13:01,946][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:13:02,444][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:13:02,942][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:13:03,439][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10289 tokens. +[2026-03-26 01:13:04,125][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.14%, ΔTime: 00:00:32 +[2026-03-26 01:13:04,923][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:13:04,925][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:13:04,927][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:13:05,669][__main__][INFO] - Iteration 220 took 52s (32.17% Gen, 66.42% Train). Generation: 16s, Training: 34s. Estimated remaining time: 40h 36m 33s. Estimated total time: 43h 48m 50s. Time estimates for 10 more iterations: 8m 45s, 100 more iterations: 1h 27m 37s, 500 more iterations: 7h 18m 8s. +[2026-03-26 01:13:05,672][__main__][INFO] - Starting iteration 220. +[2026-03-26 01:13:06,079][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 21 and human policies 1. +[2026-03-26 01:13:06,080][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:13:09,720][mllm.models.large_language_model_local][WARNING] - Response Proposal: 5 hats, 5 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:13:23,052][__main__][INFO] - Number of regex retries in iteration 220: 1 +[2026-03-26 01:13:23,053][__main__][INFO] - agents played in iteration 220 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:13:23,837][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:13:23,860][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:13:23,884][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:13:23,907][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:13:23,907][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:13:23,908][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:13:24,567][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:13:25,023][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:13:25,526][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:13:26,029][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:13:26,530][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:13:27,028][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:13:27,526][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:13:28,027][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:13:28,527][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:13:29,026][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:13:29,527][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:13:30,031][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:13:30,531][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:13:31,040][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:13:31,543][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:13:32,043][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:13:32,545][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:13:33,046][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:13:33,547][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:13:34,047][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:13:34,548][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:13:35,048][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:13:35,549][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:13:36,053][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:13:36,550][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:13:37,048][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:13:37,549][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:13:38,046][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:13:38,546][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:13:39,044][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:13:39,542][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:13:40,043][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:13:40,545][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:13:41,043][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:13:41,553][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:13:42,048][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:13:42,543][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:13:43,041][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:13:43,537][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:13:44,034][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:13:44,528][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:13:45,026][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:13:45,527][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:13:46,027][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:13:46,524][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:13:47,027][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:13:47,525][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:13:48,024][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:13:48,523][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:13:49,022][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:13:49,525][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:13:50,028][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:13:50,527][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:13:51,028][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:13:51,526][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:13:52,025][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:13:52,522][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:13:53,022][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:13:53,522][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:13:54,019][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:13:54,519][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:13:55,016][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:13:55,515][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:13:56,011][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:13:56,512][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10318 tokens. +[2026-03-26 01:13:57,205][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.17%, Current % of VRAM taken: 60.60%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 01:13:57,956][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:13:57,958][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:13:57,960][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:13:59,674][__main__][INFO] - Iteration 221 took 53s (31.67% Gen, 65.13% Train). Generation: 16s, Training: 34s. Estimated remaining time: 41h 26m 36s. Estimated total time: 44h 39m 47s. Time estimates for 10 more iterations: 8m 55s, 100 more iterations: 1h 29m 19s, 500 more iterations: 7h 26m 37s. +[2026-03-26 01:13:59,676][__main__][INFO] - Starting iteration 221. +[2026-03-26 01:14:00,082][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 22 and human policies 1. +[2026-03-26 01:14:00,083][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:14:17,598][__main__][INFO] - Number of regex retries in iteration 221: 0 +[2026-03-26 01:14:17,599][__main__][INFO] - agents played in iteration 221 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:14:18,383][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:14:18,406][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:14:18,430][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:14:18,452][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:14:18,453][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:14:18,454][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:14:19,100][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:14:19,604][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:14:20,107][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:14:20,605][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:14:21,102][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:14:21,598][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:14:22,093][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:14:22,588][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:14:23,086][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:14:23,590][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:14:24,095][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:14:24,600][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:14:25,098][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:14:25,600][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:14:26,100][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:14:26,602][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:14:27,102][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:14:27,607][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:14:28,107][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:14:28,629][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:14:29,135][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:14:29,636][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:14:30,137][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:14:30,639][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:14:31,141][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:14:31,643][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:14:32,142][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:14:32,648][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:14:33,152][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:14:33,658][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:14:34,160][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:14:34,662][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:14:35,159][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:14:35,660][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:14:36,162][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:14:36,660][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:14:37,161][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:14:37,659][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:14:38,161][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:14:38,662][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:14:39,162][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:14:39,663][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:14:40,162][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:14:40,665][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:14:41,168][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:14:41,667][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:14:42,170][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:14:42,672][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:14:43,173][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:14:43,675][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:14:44,174][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:14:44,673][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:14:45,170][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:14:45,667][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:14:46,164][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:14:46,663][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:14:47,163][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:14:47,664][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:14:48,169][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:14:48,670][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:14:49,169][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:14:49,669][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:14:50,170][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:14:50,671][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:14:51,172][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10326 tokens. +[2026-03-26 01:14:51,849][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.27%, Current % of VRAM taken: 60.69%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 01:14:52,597][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:14:52,599][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:14:52,601][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:14:53,368][__main__][INFO] - Iteration 222 took 53s (32.87% Gen, 65.68% Train). Generation: 17s, Training: 35s. Estimated remaining time: 41h 10m 14s. Estimated total time: 44h 24m 19s. Time estimates for 10 more iterations: 8m 52s, 100 more iterations: 1h 28m 48s, 500 more iterations: 7h 24m 3s. +[2026-03-26 01:14:53,371][__main__][INFO] - Starting iteration 222. +[2026-03-26 01:14:53,770][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 22 and human policies 1. +[2026-03-26 01:14:53,771][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:15:10,430][__main__][INFO] - Number of regex retries in iteration 222: 0 +[2026-03-26 01:15:10,431][__main__][INFO] - agents played in iteration 222 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:15:11,213][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:15:11,236][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:15:11,259][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:15:11,282][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:15:11,283][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:15:11,283][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:15:11,932][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:15:12,388][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:15:12,890][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:15:13,388][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:15:13,890][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:15:14,388][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:15:14,889][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:15:15,390][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:15:15,890][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:15:16,389][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:15:16,887][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:15:17,393][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:15:17,895][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:15:18,398][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:15:18,898][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:15:19,397][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:15:19,897][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:15:20,397][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:15:20,897][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:15:21,396][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:15:21,897][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:15:22,398][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:15:22,899][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:15:23,397][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:15:23,900][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:15:24,406][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:15:24,909][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:15:25,406][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:15:25,906][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:15:26,407][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:15:26,905][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:15:27,402][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:15:27,899][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:15:28,396][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:15:28,894][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:15:29,392][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:15:29,890][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:15:30,388][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:15:30,886][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:15:31,384][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:15:31,882][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:15:32,382][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:15:32,886][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:15:33,387][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:15:33,887][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:15:34,388][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:15:34,888][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:15:35,389][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:15:35,890][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:15:36,388][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:15:36,892][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:15:37,391][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:15:37,888][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:15:38,389][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:15:38,886][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:15:39,384][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:15:39,884][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:15:40,382][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:15:40,882][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:15:41,382][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:15:41,880][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:15:42,379][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:15:42,875][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:15:43,372][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:15:43,871][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10309 tokens. +[2026-03-26 01:15:44,550][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.16%, Current % of VRAM taken: 60.58%, Block Peak % of device VRAM: 62.27%, ΔTime: 00:00:32 +[2026-03-26 01:15:45,299][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:15:45,301][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:15:45,303][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:15:46,063][__main__][INFO] - Iteration 223 took 52s (31.86% Gen, 66.68% Train). Generation: 16s, Training: 34s. Estimated remaining time: 40h 19m 42s. Estimated total time: 43h 34m 39s. Time estimates for 10 more iterations: 8m 42s, 100 more iterations: 1h 27m 9s, 500 more iterations: 7h 15m 46s. +[2026-03-26 01:15:46,065][__main__][INFO] - Starting iteration 223. +[2026-03-26 01:15:46,466][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 22 and human policies 1. +[2026-03-26 01:15:46,466][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:16:04,201][__main__][INFO] - Number of regex retries in iteration 223: 0 +[2026-03-26 01:16:04,202][__main__][INFO] - agents played in iteration 223 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:16:04,976][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:16:05,000][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:16:05,023][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:16:05,045][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:16:05,046][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:16:05,046][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:16:05,697][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:16:06,148][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:16:06,652][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:16:07,151][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:16:07,649][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:16:08,149][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:16:08,646][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:16:09,144][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:16:09,643][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:16:10,147][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:16:10,648][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:16:11,150][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:16:11,655][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:16:12,154][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:16:12,656][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:16:13,156][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:16:13,660][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:16:14,159][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:16:14,659][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:16:15,168][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:16:15,672][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:16:16,174][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:16:16,671][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:16:17,168][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:16:17,667][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:16:18,168][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:16:18,665][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:16:19,164][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:16:19,662][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:16:20,159][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:16:20,657][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:16:21,154][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:16:21,650][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:16:22,148][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:16:22,645][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:16:23,144][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:16:23,642][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:16:24,140][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:16:24,640][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:16:25,138][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:16:25,636][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:16:26,137][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:16:26,633][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:16:27,131][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:16:27,634][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:16:28,132][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:16:28,629][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:16:29,126][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:16:29,624][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:16:30,122][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:16:30,622][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:16:31,146][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:16:31,648][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:16:32,148][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:16:32,647][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:16:33,146][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:16:33,643][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:16:34,143][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:16:34,641][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:16:35,139][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:16:35,636][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:16:36,136][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:16:36,631][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:16:37,128][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:16:37,627][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10312 tokens. +[2026-03-26 01:16:38,346][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.61%, Block Peak % of device VRAM: 62.23%, ΔTime: 00:00:32 +[2026-03-26 01:16:39,101][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:16:39,104][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:16:39,105][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:16:39,745][__main__][INFO] - Iteration 224 took 53s (33.29% Gen, 65.51% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 8m 7s. Estimated total time: 44h 23m 58s. Time estimates for 10 more iterations: 8m 52s, 100 more iterations: 1h 28m 47s, 500 more iterations: 7h 23m 59s. +[2026-03-26 01:16:39,747][__main__][INFO] - Starting iteration 224. +[2026-03-26 01:16:40,151][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 22 and human policies 1. +[2026-03-26 01:16:40,152][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:16:59,017][__main__][INFO] - Number of regex retries in iteration 224: 0 +[2026-03-26 01:16:59,018][__main__][INFO] - agents played in iteration 224 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:16:59,786][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:16:59,810][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:16:59,833][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:16:59,855][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:16:59,856][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:16:59,856][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:17:00,497][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:17:00,955][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:17:01,457][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:17:01,957][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:17:02,457][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:17:02,959][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:17:03,460][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:17:03,959][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:17:04,464][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:17:04,968][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:17:05,472][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:17:05,979][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:17:06,486][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:17:06,987][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:17:07,485][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:17:07,982][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:17:08,480][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:17:08,977][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:17:09,479][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:17:09,983][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:17:10,484][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:17:10,984][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:17:11,487][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:17:11,987][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:17:12,484][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:17:12,982][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:17:13,478][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:17:13,976][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:17:14,481][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:17:14,977][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:17:15,474][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:17:15,971][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:17:16,470][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:17:16,970][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:17:17,471][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:17:17,969][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:17:18,468][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:17:18,969][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:17:19,470][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:17:19,968][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:17:20,468][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:17:21,117][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:17:21,615][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:17:22,113][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:17:22,609][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:17:23,108][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:17:23,606][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:17:24,103][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:17:24,603][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:17:25,104][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:17:25,606][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:17:26,106][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:17:26,605][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:17:27,105][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:17:27,605][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:17:28,109][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:17:28,610][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:17:29,111][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:17:29,611][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:17:30,113][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:17:30,616][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:17:31,116][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:17:31,611][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:17:32,111][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:17:32,610][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10244 tokens. +[2026-03-26 01:17:33,292][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.21%, Current % of VRAM taken: 60.64%, Block Peak % of device VRAM: 62.32%, ΔTime: 00:00:32 +[2026-03-26 01:17:34,074][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:17:34,076][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:17:34,078][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:17:34,879][__main__][INFO] - Iteration 225 took 54s (34.47% Gen, 64.06% Train). Generation: 18s, Training: 35s. Estimated remaining time: 42h 19m 39s. Estimated total time: 45h 36m 25s. Time estimates for 10 more iterations: 9m 7s, 100 more iterations: 1h 31m 12s, 500 more iterations: 7h 36m 4s. +[2026-03-26 01:17:34,881][__main__][INFO] - Starting iteration 225. +[2026-03-26 01:17:35,304][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 22 and human policies 1. +[2026-03-26 01:17:35,305][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:18:09,115][__main__][INFO] - Number of regex retries in iteration 225: 0 +[2026-03-26 01:18:09,116][__main__][INFO] - agents played in iteration 225 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:18:09,914][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:18:09,937][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:18:09,960][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:18:09,983][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:18:09,984][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:18:09,984][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:18:10,646][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:18:11,099][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:18:11,607][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:18:12,104][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:18:12,601][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:18:13,098][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:18:13,594][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:18:14,090][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:18:14,586][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:18:15,086][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:18:15,589][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:18:16,094][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:18:16,600][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:18:17,098][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:18:17,603][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:18:18,102][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:18:18,600][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:18:19,099][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:18:19,596][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:18:20,095][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:18:20,592][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:18:21,091][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:18:21,590][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:18:22,092][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:18:22,591][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:18:23,089][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:18:23,589][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:18:24,086][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:18:24,586][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:18:25,086][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:18:25,584][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:18:26,083][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:18:26,579][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:18:27,077][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:18:27,577][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:18:28,078][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:18:28,576][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:18:29,075][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:18:29,573][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:18:30,071][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:18:30,573][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:18:31,074][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:18:31,572][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:18:32,072][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:18:32,568][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:18:33,067][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:18:33,567][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:18:34,067][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:18:34,567][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:18:35,065][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:18:35,564][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:18:36,063][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:18:36,560][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:18:37,056][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:18:37,553][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:18:38,054][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:18:38,550][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:18:39,047][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:18:39,546][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:18:40,043][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:18:40,540][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:18:41,036][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:18:41,535][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:18:42,031][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:18:42,531][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10157 tokens. +[2026-03-26 01:18:43,219][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.22%, Current % of VRAM taken: 60.64%, Block Peak % of device VRAM: 62.19%, ΔTime: 00:00:32 +[2026-03-26 01:18:43,993][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:18:43,995][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:18:43,996][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:18:44,837][__main__][INFO] - Iteration 226 took 1m 9s (48.62% Gen, 50.16% Train). Generation: 33s, Training: 34s. Estimated remaining time: 54h 38m 44s. Estimated total time: 57h 56m 40s. Time estimates for 10 more iterations: 11m 35s, 100 more iterations: 1h 55m 53s, 500 more iterations: 9h 39m 26s. +[2026-03-26 01:18:44,840][__main__][INFO] - Starting iteration 226. +[2026-03-26 01:18:45,253][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 22 and human policies 1. +[2026-03-26 01:18:45,254][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:19:02,407][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:19:03,258][__main__][INFO] - Number of regex retries in iteration 226: 1 +[2026-03-26 01:19:03,258][__main__][INFO] - agents played in iteration 226 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:19:04,047][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:19:04,070][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:19:04,093][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:19:04,116][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:19:04,117][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:19:04,118][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:19:04,761][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:19:05,221][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:19:05,725][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:19:06,228][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:19:06,729][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:19:07,233][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:19:07,734][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:19:08,238][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:19:08,740][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:19:09,239][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:19:09,739][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:19:10,241][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:19:10,739][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:19:11,238][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:19:11,738][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:19:12,239][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:19:12,737][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:19:13,235][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:19:13,736][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:19:14,238][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:19:14,745][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:19:15,264][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:19:15,765][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:19:16,267][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:19:16,766][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:19:17,266][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:19:17,767][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:19:18,268][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:19:18,770][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:19:19,270][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:19:19,773][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:19:20,273][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:19:20,774][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:19:21,276][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:19:21,781][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:19:22,282][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:19:22,781][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:19:23,280][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:19:23,781][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:19:24,278][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:19:24,778][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:19:25,277][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:19:25,774][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:19:26,271][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:19:26,768][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:19:27,268][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:19:27,768][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:19:28,267][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:19:28,766][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:19:29,262][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:19:29,757][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:19:30,251][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:19:30,746][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:19:31,242][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:19:31,738][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:19:32,236][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:19:32,731][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:19:33,231][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:19:33,728][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:19:34,228][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:19:34,728][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:19:35,227][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:19:35,727][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:19:36,225][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:19:36,726][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10259 tokens. +[2026-03-26 01:19:37,402][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.22%, Current % of VRAM taken: 60.64%, Block Peak % of device VRAM: 62.27%, ΔTime: 00:00:32 +[2026-03-26 01:19:38,170][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:19:38,172][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:19:38,174][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:19:38,878][__main__][INFO] - Iteration 227 took 53s (33.57% Gen, 65.11% Train). Generation: 18s, Training: 34s. Estimated remaining time: 41h 22m 25s. Estimated total time: 44h 41m 15s. Time estimates for 10 more iterations: 8m 56s, 100 more iterations: 1h 29m 22s, 500 more iterations: 7h 26m 52s. +[2026-03-26 01:19:38,880][__main__][INFO] - Starting iteration 227. +[2026-03-26 01:19:39,287][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 22 and human policies 1. +[2026-03-26 01:19:39,288][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:19:54,523][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:19:57,283][__main__][INFO] - Number of regex retries in iteration 227: 1 +[2026-03-26 01:19:57,284][__main__][INFO] - agents played in iteration 227 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:19:58,086][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:19:58,110][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:19:58,133][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:19:58,156][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:19:58,157][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:19:58,158][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:19:58,803][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:19:59,263][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:19:59,771][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:20:00,273][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:20:00,779][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:20:01,281][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:20:01,786][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:20:02,290][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:20:02,794][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:20:03,297][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:20:03,801][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:20:04,304][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:20:04,806][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:20:05,308][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:20:05,812][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:20:06,314][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:20:06,816][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:20:07,318][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:20:07,819][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:20:08,319][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:20:08,823][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:20:09,323][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:20:09,823][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:20:10,321][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:20:10,822][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:20:11,322][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:20:11,820][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:20:12,322][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:20:12,824][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:20:13,322][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:20:13,820][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:20:14,319][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:20:14,816][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:20:15,316][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:20:15,820][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:20:16,323][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:20:16,823][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:20:17,324][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:20:17,824][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:20:18,326][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:20:18,829][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:20:19,328][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:20:19,826][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:20:20,325][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:20:20,825][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:20:21,324][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:20:21,824][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:20:22,325][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:20:22,822][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:20:23,325][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:20:23,823][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:20:24,325][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:20:24,822][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:20:25,320][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:20:25,816][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:20:26,316][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:20:26,813][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:20:27,313][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:20:27,814][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:20:28,314][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:20:28,815][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:20:29,311][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:20:29,809][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:20:30,309][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:20:30,810][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10476 tokens. +[2026-03-26 01:20:31,490][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.19%, Current % of VRAM taken: 60.61%, Block Peak % of device VRAM: 62.25%, ΔTime: 00:00:32 +[2026-03-26 01:20:32,239][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:20:32,242][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:20:32,244][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:20:32,920][__main__][INFO] - Iteration 228 took 53s (33.55% Gen, 65.18% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 21m 58s. Estimated total time: 44h 41m 42s. Time estimates for 10 more iterations: 8m 56s, 100 more iterations: 1h 29m 23s, 500 more iterations: 7h 26m 57s. +[2026-03-26 01:20:32,922][__main__][INFO] - Starting iteration 228. +[2026-03-26 01:20:33,323][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 22 and human policies 1. +[2026-03-26 01:20:33,324][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:20:49,197][__main__][INFO] - Number of regex retries in iteration 228: 0 +[2026-03-26 01:20:49,197][__main__][INFO] - agents played in iteration 228 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:20:49,983][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:20:50,006][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:20:50,029][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:20:50,052][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:20:50,053][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:20:50,053][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:20:50,698][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:20:51,155][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:20:51,656][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:20:52,155][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:20:52,665][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:20:53,164][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:20:53,666][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:20:54,164][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:20:54,665][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:20:55,165][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:20:55,666][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:20:56,170][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:20:56,676][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:20:57,177][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:20:57,699][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:20:58,205][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:20:58,707][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:20:59,206][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:20:59,711][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:21:00,211][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:21:00,710][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:21:01,211][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:21:01,711][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:21:02,210][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:21:02,707][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:21:03,207][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:21:03,707][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:21:04,205][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:21:04,705][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:21:05,205][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:21:05,704][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:21:06,203][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:21:06,703][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:21:07,203][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:21:07,705][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:21:08,204][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:21:08,699][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:21:09,196][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:21:09,691][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:21:10,189][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:21:10,691][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:21:11,190][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:21:11,688][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:21:12,185][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:21:12,683][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:21:13,179][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:21:13,676][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:21:14,174][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:21:14,674][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:21:15,170][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:21:15,667][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:21:16,167][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:21:16,664][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:21:17,161][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:21:17,660][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:21:18,157][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:21:18,654][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:21:19,150][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:21:19,647][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:21:20,146][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:21:20,646][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:21:21,143][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:21:21,638][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:21:22,133][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:21:22,630][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10292 tokens. +[2026-03-26 01:21:23,307][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.15%, Current % of VRAM taken: 60.58%, Block Peak % of device VRAM: 62.40%, ΔTime: 00:00:32 +[2026-03-26 01:21:24,054][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:21:24,056][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:21:24,058][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:21:24,923][__main__][INFO] - Iteration 229 took 51s (30.76% Gen, 67.56% Train). Generation: 15s, Training: 34s. Estimated remaining time: 39h 39m 25s. Estimated total time: 43h 0m 2s. Time estimates for 10 more iterations: 8m 36s, 100 more iterations: 1h 26m 0s, 500 more iterations: 7h 10m 0s. +[2026-03-26 01:21:24,925][__main__][INFO] - Starting iteration 229. +[2026-03-26 01:21:25,326][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 22 and human policies 1. +[2026-03-26 01:21:25,327][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:21:42,336][__main__][INFO] - Number of regex retries in iteration 229: 0 +[2026-03-26 01:21:42,336][__main__][INFO] - agents played in iteration 229 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:21:43,171][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:21:43,195][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:21:43,218][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:21:43,240][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:21:43,241][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:21:43,242][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:21:43,887][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:21:44,346][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:21:44,852][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:21:45,352][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:21:45,850][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:21:46,352][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:21:46,853][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:21:47,358][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:21:47,856][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:21:48,359][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:21:48,862][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:21:49,364][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:21:49,865][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:21:50,365][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:21:50,864][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:21:51,361][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:21:51,863][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:21:52,364][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:21:52,864][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:21:53,364][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:21:53,864][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:21:54,365][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:21:54,865][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:21:55,364][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:21:55,863][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:21:56,359][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:21:56,856][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:21:57,351][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:21:57,847][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:21:58,346][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:21:58,846][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:21:59,343][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:21:59,843][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:22:00,342][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:22:00,842][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:22:01,345][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:22:01,847][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:22:02,344][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:22:02,842][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:22:03,340][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:22:03,837][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:22:04,335][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:22:04,836][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:22:05,336][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:22:05,836][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:22:06,332][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:22:06,832][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:22:07,332][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:22:07,828][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:22:08,329][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:22:08,826][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:22:09,322][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:22:09,819][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:22:10,317][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:22:10,816][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:22:11,312][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:22:11,810][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:22:12,309][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:22:12,807][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:22:13,306][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:22:13,805][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:22:14,305][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:22:14,802][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:22:15,302][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:22:15,800][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10283 tokens. +[2026-03-26 01:22:16,486][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.10%, ΔTime: 00:00:32 +[2026-03-26 01:22:17,256][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:22:17,258][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:22:17,260][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:22:17,910][__main__][INFO] - Iteration 230 took 52s (32.35% Gen, 66.41% Train). Generation: 17s, Training: 34s. Estimated remaining time: 40h 27m 45s. Estimated total time: 43h 49m 15s. Time estimates for 10 more iterations: 8m 45s, 100 more iterations: 1h 27m 38s, 500 more iterations: 7h 18m 12s. +[2026-03-26 01:22:17,913][__main__][INFO] - Starting iteration 230. +[2026-03-26 01:22:18,314][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 22 and human policies 1. +[2026-03-26 01:22:18,315][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:22:22,804][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:22:36,320][__main__][INFO] - Number of regex retries in iteration 230: 1 +[2026-03-26 01:22:36,321][__main__][INFO] - agents played in iteration 230 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:22:37,107][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:22:37,131][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:22:37,155][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:22:37,179][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:22:37,179][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:22:37,180][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:22:37,833][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:22:38,294][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:22:38,806][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:22:39,314][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:22:39,814][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:22:40,314][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:22:40,815][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:22:41,316][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:22:41,818][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:22:42,321][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:22:42,820][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:22:43,322][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:22:43,822][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:22:44,321][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:22:44,823][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:22:45,321][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:22:45,822][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:22:46,319][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:22:46,819][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:22:47,318][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:22:47,815][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:22:48,312][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:22:48,812][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:22:49,309][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:22:49,804][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:22:50,303][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:22:50,804][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:22:51,305][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:22:51,805][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:22:52,305][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:22:52,803][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:22:53,300][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:22:53,801][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:22:54,302][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:22:54,798][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:22:55,301][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:22:55,802][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:22:56,299][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:22:56,795][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:22:57,295][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:22:57,795][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:22:58,295][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:22:58,796][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:22:59,298][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:22:59,799][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:23:00,298][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:23:00,796][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:23:01,296][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:23:01,796][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:23:02,298][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:23:02,797][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:23:03,293][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:23:03,794][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:23:04,293][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:23:04,789][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:23:05,287][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:23:05,784][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:23:06,282][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:23:06,779][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:23:07,278][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:23:07,774][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:23:08,270][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:23:08,770][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:23:09,268][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:23:09,769][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10399 tokens. +[2026-03-26 01:23:10,449][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.61%, Block Peak % of device VRAM: 62.26%, ΔTime: 00:00:32 +[2026-03-26 01:23:11,206][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:23:11,209][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:23:11,210][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:23:12,840][__main__][INFO] - Iteration 231 took 54s (33.02% Gen, 63.99% Train). Generation: 18s, Training: 34s. Estimated remaining time: 42h 3m 56s. Estimated total time: 45h 26m 20s. Time estimates for 10 more iterations: 9m 5s, 100 more iterations: 1h 30m 52s, 500 more iterations: 7h 34m 23s. +[2026-03-26 01:23:12,843][__main__][INFO] - Starting iteration 231. +[2026-03-26 01:23:13,243][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 23 and human policies 1. +[2026-03-26 01:23:13,244][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:23:31,268][__main__][INFO] - Number of regex retries in iteration 231: 0 +[2026-03-26 01:23:31,269][__main__][INFO] - agents played in iteration 231 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:23:32,061][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:23:32,085][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:23:32,109][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:23:32,132][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:23:32,133][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:23:32,134][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:23:32,792][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:23:33,249][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:23:33,752][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:23:34,255][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:23:34,757][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:23:35,258][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:23:35,758][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:23:36,261][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:23:36,761][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:23:37,264][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:23:37,761][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:23:38,262][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:23:38,762][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:23:39,263][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:23:39,763][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:23:40,264][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:23:40,765][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:23:41,264][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:23:41,765][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:23:42,262][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:23:42,763][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:23:43,262][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:23:43,758][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:23:44,258][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:23:44,759][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:23:45,255][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:23:45,752][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:23:46,249][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:23:46,747][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:23:47,244][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:23:47,744][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:23:48,243][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:23:48,742][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:23:49,239][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:23:49,735][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:23:50,238][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:23:50,735][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:23:51,234][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:23:51,735][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:23:52,234][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:23:52,731][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:23:53,228][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:23:53,728][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:23:54,225][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:23:54,721][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:23:55,222][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:23:55,722][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:23:56,222][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:23:56,723][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:23:57,219][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:23:57,716][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:23:58,214][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:23:58,711][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:23:59,211][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:23:59,709][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:24:00,206][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:24:00,705][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:24:01,201][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:24:01,697][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:24:02,194][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:24:02,695][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:24:03,191][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:24:03,693][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:24:04,190][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:24:04,691][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10373 tokens. +[2026-03-26 01:24:05,376][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.24%, Current % of VRAM taken: 60.66%, Block Peak % of device VRAM: 62.15%, ΔTime: 00:00:32 +[2026-03-26 01:24:06,160][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:24:06,162][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:24:06,164][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:24:06,988][__main__][INFO] - Iteration 232 took 53s (33.54% Gen, 64.93% Train). Generation: 18s, Training: 34s. Estimated remaining time: 41h 23m 57s. Estimated total time: 44h 47m 16s. Time estimates for 10 more iterations: 8m 57s, 100 more iterations: 1h 29m 34s, 500 more iterations: 7h 27m 52s. +[2026-03-26 01:24:06,992][__main__][INFO] - Starting iteration 232. +[2026-03-26 01:24:07,429][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 23 and human policies 1. +[2026-03-26 01:24:07,429][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:24:11,142][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:24:23,508][__main__][INFO] - Number of regex retries in iteration 232: 1 +[2026-03-26 01:24:23,508][__main__][INFO] - agents played in iteration 232 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:24:24,286][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:24:24,310][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:24:24,333][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:24:24,355][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:24:24,356][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:24:24,357][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:24:25,001][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:24:25,460][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:24:25,966][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:24:26,470][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:24:26,974][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:24:27,476][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:24:27,978][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:24:28,477][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:24:28,980][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:24:29,484][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:24:29,986][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:24:30,489][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:24:30,995][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:24:31,497][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:24:31,998][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:24:32,500][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:24:33,002][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:24:33,503][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:24:34,003][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:24:34,500][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:24:35,000][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:24:35,501][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:24:36,001][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:24:36,498][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:24:36,994][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:24:37,498][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:24:38,016][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:24:38,515][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:24:39,017][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:24:39,517][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:24:40,012][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:24:40,516][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:24:41,017][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:24:41,515][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:24:42,015][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:24:42,511][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:24:43,012][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:24:43,511][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:24:44,007][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:24:44,504][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:24:45,000][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:24:45,501][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:24:46,002][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:24:46,500][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:24:46,996][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:24:47,495][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:24:47,994][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:24:48,493][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:24:48,990][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:24:49,491][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:24:49,993][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:24:50,494][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:24:50,993][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:24:51,491][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:24:51,989][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:24:52,490][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:24:52,991][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:24:53,489][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:24:53,989][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:24:54,490][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:24:54,991][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:24:55,492][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:24:55,992][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:24:56,497][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:24:57,003][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10508 tokens. +[2026-03-26 01:24:57,680][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.24%, Current % of VRAM taken: 60.67%, Block Peak % of device VRAM: 62.29%, ΔTime: 00:00:32 +[2026-03-26 01:24:58,436][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:24:58,438][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:24:58,440][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:24:59,272][__main__][INFO] - Iteration 233 took 51s (31.01% Gen, 67.38% Train). Generation: 16s, Training: 34s. Estimated remaining time: 39h 48m 1s. Estimated total time: 43h 12m 12s. Time estimates for 10 more iterations: 8m 38s, 100 more iterations: 1h 26m 24s, 500 more iterations: 7h 12m 2s. +[2026-03-26 01:24:59,274][__main__][INFO] - Starting iteration 233. +[2026-03-26 01:24:59,675][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 23 and human policies 1. +[2026-03-26 01:24:59,676][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:25:17,463][__main__][INFO] - Number of regex retries in iteration 233: 0 +[2026-03-26 01:25:17,464][__main__][INFO] - agents played in iteration 233 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:25:18,241][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:25:18,264][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:25:18,287][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:25:18,310][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:25:18,311][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:25:18,311][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:25:18,963][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:25:19,422][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:25:19,932][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:25:20,433][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:25:20,935][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:25:21,434][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:25:21,934][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:25:22,437][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:25:22,937][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:25:23,440][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:25:23,944][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:25:24,447][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:25:24,949][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:25:25,450][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:25:25,950][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:25:26,449][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:25:26,949][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:25:27,445][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:25:27,945][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:25:28,444][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:25:28,945][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:25:29,444][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:25:29,942][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:25:30,443][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:25:30,944][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:25:31,447][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:25:31,947][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:25:32,449][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:25:32,948][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:25:33,451][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:25:33,952][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:25:34,451][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:25:34,953][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:25:35,450][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:25:35,946][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:25:36,446][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:25:36,946][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:25:37,442][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:25:37,939][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:25:38,434][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:25:38,930][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:25:39,427][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:25:39,927][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:25:40,428][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:25:40,925][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:25:41,422][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:25:41,923][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:25:42,418][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:25:42,916][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:25:43,412][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:25:43,910][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:25:44,409][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:25:44,909][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:25:45,408][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:25:45,909][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:25:46,407][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:25:46,908][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:25:47,408][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:25:47,914][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:25:48,415][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:25:48,923][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:25:49,428][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:25:49,931][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:25:50,433][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:25:50,933][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10497 tokens. +[2026-03-26 01:25:51,617][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.25%, Current % of VRAM taken: 60.68%, Block Peak % of device VRAM: 62.31%, ΔTime: 00:00:32 +[2026-03-26 01:25:52,378][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:25:52,380][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:25:52,382][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:25:53,102][__main__][INFO] - Iteration 234 took 53s (33.29% Gen, 65.35% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 6m 17s. Estimated total time: 44h 31m 22s. Time estimates for 10 more iterations: 8m 54s, 100 more iterations: 1h 29m 2s, 500 more iterations: 7h 25m 13s. +[2026-03-26 01:25:53,104][__main__][INFO] - Starting iteration 234. +[2026-03-26 01:25:53,506][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 23 and human policies 1. +[2026-03-26 01:25:53,506][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:25:57,574][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:26:10,510][__main__][INFO] - Number of regex retries in iteration 234: 1 +[2026-03-26 01:26:10,511][__main__][INFO] - agents played in iteration 234 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:26:11,278][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:26:11,301][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:26:11,325][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:26:11,348][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:26:11,348][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:26:11,349][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:26:12,008][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:26:12,464][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:26:12,972][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:26:13,484][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:26:13,988][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:26:14,491][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:26:14,995][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:26:15,500][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:26:16,002][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:26:16,501][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:26:17,002][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:26:17,502][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:26:18,005][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:26:18,503][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:26:19,002][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:26:19,501][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:26:19,998][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:26:20,500][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:26:20,998][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:26:21,502][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:26:22,005][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:26:22,507][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:26:23,011][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:26:23,513][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:26:24,014][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:26:24,519][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:26:25,020][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:26:25,524][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:26:26,027][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:26:26,527][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:26:27,026][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:26:27,530][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:26:28,030][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:26:28,527][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:26:29,030][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:26:29,531][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:26:30,031][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:26:30,530][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:26:31,029][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:26:31,531][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:26:32,029][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:26:32,526][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:26:33,027][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:26:33,522][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:26:34,018][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:26:34,517][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:26:35,016][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:26:35,514][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:26:36,012][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:26:36,512][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:26:37,011][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:26:37,511][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:26:38,010][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:26:38,510][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:26:39,009][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:26:39,508][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:26:40,009][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:26:40,509][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:26:41,004][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:26:41,506][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:26:42,009][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:26:42,511][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:26:43,017][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:26:43,517][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:26:44,013][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10504 tokens. +[2026-03-26 01:26:44,712][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.39%, ΔTime: 00:00:32 +[2026-03-26 01:26:45,474][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:26:45,477][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:26:45,478][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:26:46,235][__main__][INFO] - Iteration 235 took 52s (32.25% Gen, 66.31% Train). Generation: 17s, Training: 34s. Estimated remaining time: 40h 30m 33s. Estimated total time: 43h 56m 30s. Time estimates for 10 more iterations: 8m 47s, 100 more iterations: 1h 27m 53s, 500 more iterations: 7h 19m 25s. +[2026-03-26 01:26:46,237][__main__][INFO] - Starting iteration 235. +[2026-03-26 01:26:46,639][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 23 and human policies 1. +[2026-03-26 01:26:46,640][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:27:04,827][__main__][INFO] - Number of regex retries in iteration 235: 0 +[2026-03-26 01:27:04,827][__main__][INFO] - agents played in iteration 235 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:27:05,617][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:27:05,641][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:27:05,664][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:27:05,687][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:27:05,687][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:27:05,688][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:27:06,344][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:27:06,803][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:27:07,312][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:27:07,814][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:27:08,316][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:27:08,820][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:27:09,323][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:27:09,825][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:27:10,326][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:27:10,830][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:27:11,330][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:27:11,834][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:27:12,363][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:27:12,866][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:27:13,368][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:27:13,867][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:27:14,371][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:27:14,874][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:27:15,373][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:27:15,871][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:27:16,370][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:27:16,870][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:27:17,369][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:27:17,871][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:27:18,370][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:27:18,869][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:27:19,369][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:27:19,869][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:27:20,370][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:27:20,871][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:27:21,372][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:27:21,874][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:27:22,375][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:27:22,876][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:27:23,376][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:27:23,878][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:27:24,380][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:27:24,880][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:27:25,380][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:27:25,879][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:27:26,379][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:27:26,875][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:27:27,375][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:27:27,873][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:27:28,369][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:27:28,870][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:27:29,368][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:27:29,866][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:27:30,366][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:27:30,866][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:27:31,364][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:27:31,864][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:27:32,370][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:27:32,874][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:27:33,380][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:27:33,878][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:27:34,379][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:27:34,875][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:27:35,373][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:27:35,873][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:27:36,374][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:27:36,871][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:27:37,369][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:27:37,869][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:27:38,371][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10411 tokens. +[2026-03-26 01:27:39,065][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.22%, Current % of VRAM taken: 60.65%, Block Peak % of device VRAM: 62.45%, ΔTime: 00:00:32 +[2026-03-26 01:27:39,817][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:27:39,819][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:27:39,821][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:27:40,541][__main__][INFO] - Iteration 236 took 53s (33.74% Gen, 64.92% Train). Generation: 18s, Training: 34s. Estimated remaining time: 41h 28m 15s. Estimated total time: 44h 55m 7s. Time estimates for 10 more iterations: 8m 59s, 100 more iterations: 1h 29m 50s, 500 more iterations: 7h 29m 11s. +[2026-03-26 01:27:40,543][__main__][INFO] - Starting iteration 236. +[2026-03-26 01:27:40,942][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 23 and human policies 1. +[2026-03-26 01:27:40,943][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:27:45,605][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:27:58,250][__main__][INFO] - Number of regex retries in iteration 236: 1 +[2026-03-26 01:27:58,251][__main__][INFO] - agents played in iteration 236 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:27:59,026][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:27:59,049][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:27:59,072][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:27:59,097][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:27:59,097][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:27:59,098][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:27:59,763][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:28:00,220][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:28:00,728][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:28:01,230][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:28:01,734][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:28:02,238][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:28:02,741][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:28:03,245][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:28:03,747][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:28:04,244][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:28:04,743][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:28:05,245][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:28:05,745][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:28:06,247][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:28:06,745][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:28:07,248][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:28:07,746][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:28:08,246][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:28:08,748][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:28:09,252][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:28:09,753][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:28:10,251][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:28:10,747][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:28:11,247][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:28:11,746][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:28:12,246][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:28:12,743][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:28:13,244][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:28:13,743][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:28:14,243][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:28:14,743][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:28:15,244][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:28:15,746][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:28:16,243][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:28:16,740][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:28:17,236][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:28:17,738][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:28:18,241][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:28:18,739][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:28:19,236][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:28:19,732][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:28:20,230][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:28:20,732][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:28:21,231][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:28:21,731][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:28:22,232][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:28:22,733][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:28:23,232][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:28:23,732][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:28:24,233][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:28:24,730][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:28:25,230][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:28:25,735][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:28:26,246][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:28:26,744][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:28:27,242][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:28:27,742][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:28:28,242][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:28:28,740][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:28:29,237][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:28:29,735][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:28:30,235][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:28:30,734][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:28:31,235][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:28:31,738][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10326 tokens. +[2026-03-26 01:28:32,428][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.22%, Current % of VRAM taken: 60.64%, Block Peak % of device VRAM: 62.28%, ΔTime: 00:00:32 +[2026-03-26 01:28:33,288][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:28:33,290][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:28:33,292][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:28:34,007][__main__][INFO] - Iteration 237 took 53s (32.62% Gen, 66.03% Train). Generation: 17s, Training: 35s. Estimated remaining time: 40h 45m 32s. Estimated total time: 44h 13m 17s. Time estimates for 10 more iterations: 8m 50s, 100 more iterations: 1h 28m 26s, 500 more iterations: 7h 22m 12s. +[2026-03-26 01:28:34,010][__main__][INFO] - Starting iteration 237. +[2026-03-26 01:28:34,410][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 23 and human policies 1. +[2026-03-26 01:28:34,410][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:28:52,205][__main__][INFO] - Number of regex retries in iteration 237: 0 +[2026-03-26 01:28:52,206][__main__][INFO] - agents played in iteration 237 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:28:52,998][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:28:53,021][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:28:53,045][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:28:53,068][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:28:53,068][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:28:53,069][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:28:53,730][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:28:54,185][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:28:54,698][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:28:55,201][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:28:55,706][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:28:56,208][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:28:56,709][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:28:57,211][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:28:57,711][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:28:58,213][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:28:58,717][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:28:59,215][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:28:59,716][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:29:00,215][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:29:00,718][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:29:01,219][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:29:01,717][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:29:02,237][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:29:02,742][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:29:03,243][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:29:03,742][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:29:04,244][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:29:04,746][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:29:05,245][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:29:05,745][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:29:06,243][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:29:06,746][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:29:07,248][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:29:07,747][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:29:08,248][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:29:08,751][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:29:09,256][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:29:09,756][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:29:10,252][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:29:10,749][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:29:11,249][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:29:11,750][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:29:12,251][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:29:12,753][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:29:13,253][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:29:13,751][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:29:14,250][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:29:14,752][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:29:15,251][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:29:15,751][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:29:16,250][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:29:16,755][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:29:17,262][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:29:17,766][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:29:18,268][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:29:18,767][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:29:19,267][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:29:19,765][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:29:20,268][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:29:20,765][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:29:21,265][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:29:21,764][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:29:22,262][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:29:22,759][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:29:23,259][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:29:23,759][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:29:24,259][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:29:24,762][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:29:25,261][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:29:25,764][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10490 tokens. +[2026-03-26 01:29:26,479][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.23%, Current % of VRAM taken: 60.65%, Block Peak % of device VRAM: 62.19%, ΔTime: 00:00:32 +[2026-03-26 01:29:27,245][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:29:27,248][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:29:27,253][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:29:28,011][__main__][INFO] - Iteration 238 took 53s (33.20% Gen, 65.38% Train). Generation: 17s, Training: 35s. Estimated remaining time: 41h 11m 27s. Estimated total time: 44h 40m 7s. Time estimates for 10 more iterations: 8m 56s, 100 more iterations: 1h 29m 20s, 500 more iterations: 7h 26m 41s. +[2026-03-26 01:29:28,014][__main__][INFO] - Starting iteration 238. +[2026-03-26 01:29:28,413][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 23 and human policies 1. +[2026-03-26 01:29:28,413][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:29:46,227][__main__][INFO] - Number of regex retries in iteration 238: 0 +[2026-03-26 01:29:46,228][__main__][INFO] - agents played in iteration 238 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:29:47,007][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:29:47,031][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:29:47,054][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:29:47,076][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:29:47,077][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:29:47,077][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:29:47,755][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:29:48,216][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:29:48,723][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:29:49,228][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:29:49,733][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:29:50,238][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:29:50,739][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:29:51,243][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:29:51,745][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:29:52,245][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:29:52,743][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:29:53,245][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:29:53,746][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:29:54,245][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:29:54,746][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:29:55,245][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:29:55,744][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:29:56,241][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:29:56,742][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:29:57,243][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:29:57,743][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:29:58,242][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:29:58,741][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:29:59,240][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:29:59,741][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:30:00,239][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:30:00,742][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:30:01,239][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:30:01,735][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:30:02,237][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:30:02,741][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:30:03,241][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:30:03,743][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:30:04,242][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:30:04,743][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:30:05,244][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:30:05,741][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:30:06,239][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:30:06,740][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:30:07,245][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:30:07,752][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:30:08,258][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:30:08,762][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:30:09,259][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:30:09,759][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:30:10,255][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:30:10,756][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:30:11,255][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:30:11,753][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:30:12,251][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:30:12,755][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:30:13,256][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:30:13,757][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:30:14,256][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:30:14,758][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:30:15,259][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:30:15,755][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:30:16,252][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:30:16,749][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:30:17,247][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:30:17,749][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:30:18,248][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:30:18,746][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:30:19,247][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:30:19,746][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10457 tokens. +[2026-03-26 01:30:20,460][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.61%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 01:30:21,214][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:30:21,216][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:30:21,218][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:30:21,981][__main__][INFO] - Iteration 239 took 53s (33.25% Gen, 65.32% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 8m 53s. Estimated total time: 44h 38m 26s. Time estimates for 10 more iterations: 8m 55s, 100 more iterations: 1h 29m 16s, 500 more iterations: 7h 26m 24s. +[2026-03-26 01:30:21,984][__main__][INFO] - Starting iteration 239. +[2026-03-26 01:30:22,386][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 23 and human policies 1. +[2026-03-26 01:30:22,386][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:30:26,406][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:30:39,929][__main__][INFO] - Number of regex retries in iteration 239: 1 +[2026-03-26 01:30:39,930][__main__][INFO] - agents played in iteration 239 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:30:40,723][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:30:40,747][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:30:40,770][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:30:40,793][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:30:40,793][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:30:40,794][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:30:41,497][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:30:41,955][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:30:42,460][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:30:42,964][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:30:43,471][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:30:43,974][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:30:44,478][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:30:44,982][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:30:45,483][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:30:45,982][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:30:46,484][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:30:46,981][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:30:47,483][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:30:47,979][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:30:48,479][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:30:48,977][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:30:49,474][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:30:49,975][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:30:50,476][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:30:50,976][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:30:51,471][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:30:51,971][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:30:52,468][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:30:52,967][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:30:53,464][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:30:53,962][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:30:54,460][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:30:54,959][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:30:55,458][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:30:55,956][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:30:56,452][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:30:56,953][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:30:57,454][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:30:57,952][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:30:58,454][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:30:58,953][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:30:59,451][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:30:59,948][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:31:00,451][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:31:00,948][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:31:01,444][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:31:01,939][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:31:02,437][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:31:02,934][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:31:03,431][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:31:03,928][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:31:04,428][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:31:04,924][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:31:05,421][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:31:05,917][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:31:06,413][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:31:06,909][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:31:07,406][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:31:07,904][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:31:08,401][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:31:08,900][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:31:09,396][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:31:09,891][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:31:10,388][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:31:10,886][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:31:11,385][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:31:11,880][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:31:12,377][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:31:12,876][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:31:13,371][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10245 tokens. +[2026-03-26 01:31:14,081][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.43%, ΔTime: 00:00:32 +[2026-03-26 01:31:14,885][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:31:14,888][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:31:14,889][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:31:15,661][__main__][INFO] - Iteration 240 took 53s (32.93% Gen, 65.62% Train). Generation: 17s, Training: 34s. Estimated remaining time: 40h 53m 21s. Estimated total time: 44h 23m 48s. Time estimates for 10 more iterations: 8m 52s, 100 more iterations: 1h 28m 47s, 500 more iterations: 7h 23m 58s. +[2026-03-26 01:31:15,663][__main__][INFO] - Starting iteration 240. +[2026-03-26 01:31:16,063][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 23 and human policies 1. +[2026-03-26 01:31:16,064][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:31:22,338][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:31:22,673][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:31:33,523][__main__][INFO] - Number of regex retries in iteration 240: 2 +[2026-03-26 01:31:33,524][__main__][INFO] - agents played in iteration 240 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:31:34,316][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:31:34,340][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:31:34,363][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:31:34,386][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:31:34,386][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:31:34,387][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:31:35,047][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:31:35,503][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:31:36,008][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:31:36,513][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:31:37,023][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:31:37,526][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:31:38,028][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:31:38,533][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:31:39,033][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:31:39,538][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:31:40,042][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:31:40,543][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:31:41,045][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:31:41,550][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:31:42,051][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:31:42,552][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:31:43,052][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:31:43,554][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:31:44,053][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:31:44,553][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:31:45,053][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:31:45,554][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:31:46,054][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:31:46,554][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:31:47,055][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:31:47,554][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:31:48,054][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:31:48,553][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:31:49,051][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:31:49,549][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:31:50,054][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:31:50,555][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:31:51,060][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:31:51,564][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:31:52,064][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:31:52,563][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:31:53,065][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:31:53,568][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:31:54,071][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:31:54,572][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:31:55,073][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:31:55,570][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:31:56,067][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:31:56,564][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:31:57,067][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:31:57,569][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:31:58,069][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:31:58,569][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:31:59,068][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:31:59,569][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:32:00,069][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:32:00,568][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:32:01,069][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:32:01,566][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:32:02,068][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:32:02,568][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:32:03,065][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:32:03,561][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:32:04,065][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:32:04,562][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:32:05,064][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:32:05,564][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:32:06,063][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:32:06,561][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:32:07,058][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10433 tokens. +[2026-03-26 01:32:07,789][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.25%, ΔTime: 00:00:32 +[2026-03-26 01:32:08,551][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:32:08,553][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:32:08,555][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:32:09,859][__main__][INFO] - Iteration 241 took 53s (32.45% Gen, 65.12% Train). Generation: 17s, Training: 35s. Estimated remaining time: 41h 18m 28s. Estimated total time: 44h 49m 50s. Time estimates for 10 more iterations: 8m 57s, 100 more iterations: 1h 29m 39s, 500 more iterations: 7h 28m 18s. +[2026-03-26 01:32:09,863][__main__][INFO] - Starting iteration 241. +[2026-03-26 01:32:10,266][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 24 and human policies 1. +[2026-03-26 01:32:10,267][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:32:28,457][__main__][INFO] - Number of regex retries in iteration 241: 0 +[2026-03-26 01:32:28,458][__main__][INFO] - agents played in iteration 241 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:32:29,237][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:32:29,260][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:32:29,283][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:32:29,306][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:32:29,306][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:32:29,307][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:32:29,954][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:32:30,411][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:32:30,919][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:32:31,426][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:32:31,929][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:32:32,428][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:32:32,929][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:32:33,435][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:32:33,934][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:32:34,436][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:32:34,938][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:32:35,436][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:32:35,938][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:32:36,438][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:32:36,934][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:32:37,434][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:32:37,933][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:32:38,432][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:32:38,928][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:32:39,429][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:32:39,929][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:32:40,427][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:32:40,924][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:32:41,423][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:32:41,922][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:32:42,421][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:32:42,919][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:32:43,426][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:32:43,932][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:32:44,436][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:32:44,937][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:32:45,436][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:32:45,936][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:32:46,432][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:32:46,930][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:32:47,432][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:32:47,933][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:32:48,436][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:32:48,931][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:32:49,426][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:32:49,927][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:32:50,426][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:32:50,924][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:32:51,420][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:32:51,917][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:32:52,413][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:32:52,908][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:32:53,405][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:32:53,904][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:32:54,400][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:32:54,897][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:32:55,396][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:32:55,896][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:32:56,396][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:32:56,895][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:32:57,394][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:32:57,893][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:32:58,393][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:32:58,892][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:32:59,393][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:32:59,893][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:33:00,393][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:33:00,893][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:33:01,392][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:33:01,892][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10428 tokens. +[2026-03-26 01:33:02,605][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.21%, Current % of VRAM taken: 60.64%, Block Peak % of device VRAM: 62.27%, ΔTime: 00:00:32 +[2026-03-26 01:33:03,374][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:33:03,377][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:33:03,378][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:33:04,097][__main__][INFO] - Iteration 242 took 53s (33.79% Gen, 64.87% Train). Generation: 18s, Training: 34s. Estimated remaining time: 41h 19m 18s. Estimated total time: 44h 51m 34s. Time estimates for 10 more iterations: 8m 58s, 100 more iterations: 1h 29m 43s, 500 more iterations: 7h 28m 35s. +[2026-03-26 01:33:04,099][__main__][INFO] - Starting iteration 242. +[2026-03-26 01:33:04,502][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 24 and human policies 1. +[2026-03-26 01:33:04,503][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:33:20,758][__main__][INFO] - Number of regex retries in iteration 242: 0 +[2026-03-26 01:33:20,759][__main__][INFO] - agents played in iteration 242 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:33:21,524][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:33:21,548][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:33:21,571][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:33:21,594][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:33:21,594][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:33:21,595][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:33:22,257][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:33:22,711][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:33:23,216][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:33:23,719][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:33:24,220][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:33:24,724][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:33:25,226][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:33:25,729][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:33:26,230][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:33:26,735][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:33:27,237][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:33:27,746][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:33:28,252][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:33:28,755][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:33:29,273][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:33:29,782][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:33:30,286][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:33:30,783][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:33:31,284][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:33:31,789][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:33:32,291][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:33:32,797][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:33:33,297][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:33:33,795][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:33:34,293][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:33:34,795][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:33:35,298][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:33:35,796][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:33:36,298][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:33:36,798][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:33:37,302][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:33:37,801][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:33:38,301][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:33:38,803][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:33:39,304][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:33:39,806][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:33:40,304][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:33:40,802][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:33:41,305][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:33:41,809][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:33:42,308][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:33:42,807][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:33:43,308][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:33:43,805][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:33:44,303][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:33:44,807][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:33:45,309][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:33:45,807][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:33:46,307][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:33:46,808][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:33:47,308][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:33:47,805][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:33:48,304][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:33:48,802][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:33:49,301][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:33:49,803][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:33:50,303][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:33:50,799][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:33:51,299][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:33:51,795][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:33:52,291][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:33:52,791][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:33:53,287][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:33:53,784][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:33:54,283][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10271 tokens. +[2026-03-26 01:33:54,979][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.60%, Block Peak % of device VRAM: 62.32%, ΔTime: 00:00:32 +[2026-03-26 01:33:55,735][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:33:55,738][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:33:55,739][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:33:56,531][__main__][INFO] - Iteration 243 took 52s (31.24% Gen, 67.23% Train). Generation: 16s, Training: 34s. Estimated remaining time: 39h 48m 19s. Estimated total time: 43h 21m 27s. Time estimates for 10 more iterations: 8m 40s, 100 more iterations: 1h 26m 42s, 500 more iterations: 7h 13m 34s. +[2026-03-26 01:33:56,533][__main__][INFO] - Starting iteration 243. +[2026-03-26 01:33:56,932][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 24 and human policies 1. +[2026-03-26 01:33:56,933][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:34:14,881][__main__][INFO] - Number of regex retries in iteration 243: 0 +[2026-03-26 01:34:14,881][__main__][INFO] - agents played in iteration 243 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:34:15,653][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:34:15,676][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:34:15,700][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:34:15,722][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:34:15,723][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:34:15,724][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:34:16,382][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:34:16,833][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:34:17,335][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:34:17,834][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:34:18,336][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:34:18,838][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:34:19,340][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:34:19,839][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:34:20,338][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:34:20,843][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:34:21,342][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:34:21,845][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:34:22,347][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:34:22,852][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:34:23,359][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:34:23,865][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:34:24,370][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:34:24,873][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:34:25,370][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:34:25,896][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:34:26,401][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:34:26,904][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:34:27,404][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:34:27,905][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:34:28,407][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:34:28,906][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:34:29,406][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:34:29,906][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:34:30,408][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:34:30,908][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:34:31,407][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:34:31,905][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:34:32,406][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:34:32,905][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:34:33,405][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:34:33,905][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:34:34,410][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:34:34,915][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:34:35,416][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:34:35,916][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:34:36,414][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:34:36,913][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:34:37,415][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:34:37,919][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:34:38,421][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:34:38,921][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:34:39,420][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:34:39,920][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:34:40,418][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:34:40,915][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:34:41,417][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:34:41,914][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:34:42,412][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:34:42,910][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:34:43,415][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:34:43,917][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:34:44,416][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:34:44,917][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:34:45,417][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:34:45,916][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:34:46,413][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:34:46,911][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:34:47,408][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:34:47,907][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:34:48,405][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10304 tokens. +[2026-03-26 01:34:49,106][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.25%, ΔTime: 00:00:32 +[2026-03-26 01:34:49,860][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:34:49,863][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:34:49,864][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:34:50,575][__main__][INFO] - Iteration 244 took 53s (33.46% Gen, 65.21% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 8m 9s. Estimated total time: 44h 42m 11s. Time estimates for 10 more iterations: 8m 56s, 100 more iterations: 1h 29m 24s, 500 more iterations: 7h 27m 1s. +[2026-03-26 01:34:50,578][__main__][INFO] - Starting iteration 244. +[2026-03-26 01:34:50,981][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 24 and human policies 1. +[2026-03-26 01:34:50,982][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:34:55,303][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:35:00,823][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:35:09,088][__main__][INFO] - Number of regex retries in iteration 244: 2 +[2026-03-26 01:35:09,089][__main__][INFO] - agents played in iteration 244 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:35:09,859][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:35:09,883][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:35:09,906][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:35:09,930][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:35:09,931][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:35:09,932][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:35:10,611][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:35:11,067][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:35:11,574][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:35:12,075][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:35:12,581][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:35:13,083][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:35:13,583][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:35:14,086][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:35:14,592][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:35:15,102][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:35:15,607][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:35:16,110][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:35:16,613][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:35:17,115][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:35:17,611][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:35:18,109][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:35:18,607][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:35:19,105][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:35:19,603][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:35:20,102][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:35:20,603][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:35:21,103][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:35:21,599][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:35:22,096][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:35:22,597][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:35:23,098][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:35:23,595][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:35:24,094][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:35:24,616][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:35:25,114][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:35:25,615][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:35:26,114][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:35:26,610][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:35:27,111][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:35:27,612][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:35:28,114][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:35:28,617][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:35:29,117][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:35:29,617][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:35:30,116][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:35:30,614][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:35:31,115][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:35:31,616][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:35:32,118][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:35:32,622][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:35:33,123][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:35:33,623][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:35:34,122][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:35:34,619][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:35:35,119][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:35:35,619][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:35:36,119][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:35:36,619][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:35:37,117][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:35:37,617][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:35:38,117][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:35:38,615][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:35:39,116][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:35:39,619][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:35:40,120][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:35:40,620][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:35:41,121][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:35:41,622][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:35:42,123][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:35:42,625][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10310 tokens. +[2026-03-26 01:35:43,339][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.17%, Current % of VRAM taken: 60.60%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 01:35:44,100][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:35:44,102][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:35:44,104][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:35:44,829][__main__][INFO] - Iteration 245 took 53s (33.63% Gen, 65.02% Train). Generation: 18s, Training: 35s. Estimated remaining time: 41h 17m 31s. Estimated total time: 44h 52m 27s. Time estimates for 10 more iterations: 8m 58s, 100 more iterations: 1h 29m 44s, 500 more iterations: 7h 28m 44s. +[2026-03-26 01:35:44,831][__main__][INFO] - Starting iteration 245. +[2026-03-26 01:35:45,231][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 24 and human policies 1. +[2026-03-26 01:35:45,232][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:36:02,615][__main__][INFO] - Number of regex retries in iteration 245: 0 +[2026-03-26 01:36:02,616][__main__][INFO] - agents played in iteration 245 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:36:03,384][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:36:03,409][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:36:03,432][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:36:03,456][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:36:03,456][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:36:03,457][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:36:04,106][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:36:04,563][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:36:05,071][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:36:05,572][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:36:06,077][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:36:06,581][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:36:07,088][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:36:07,594][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:36:08,097][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:36:08,603][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:36:09,103][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:36:09,603][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:36:10,107][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:36:10,608][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:36:11,108][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:36:11,609][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:36:12,109][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:36:12,607][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:36:13,106][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:36:13,604][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:36:14,100][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:36:14,597][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:36:15,095][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:36:15,591][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:36:16,089][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:36:16,587][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:36:17,085][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:36:17,584][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:36:18,084][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:36:18,581][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:36:19,078][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:36:19,576][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:36:20,079][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:36:20,575][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:36:21,071][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:36:21,570][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:36:22,071][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:36:22,571][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:36:23,068][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:36:23,565][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:36:24,063][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:36:24,562][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:36:25,066][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:36:25,568][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:36:26,068][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:36:26,569][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:36:27,069][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:36:27,572][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:36:28,072][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:36:28,567][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:36:29,063][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:36:29,559][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:36:30,059][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:36:30,555][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:36:31,053][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:36:31,551][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:36:32,050][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:36:32,548][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:36:33,046][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:36:33,545][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:36:34,040][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:36:34,537][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:36:35,034][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:36:35,530][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:36:36,027][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10257 tokens. +[2026-03-26 01:36:36,726][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.28%, ΔTime: 00:00:32 +[2026-03-26 01:36:37,494][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:36:37,497][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:36:37,499][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:36:38,184][__main__][INFO] - Iteration 246 took 52s (32.83% Gen, 65.87% Train). Generation: 17s, Training: 34s. Estimated remaining time: 40h 31m 52s. Estimated total time: 44h 7m 41s. Time estimates for 10 more iterations: 8m 49s, 100 more iterations: 1h 28m 15s, 500 more iterations: 7h 21m 16s. +[2026-03-26 01:36:38,187][__main__][INFO] - Starting iteration 246. +[2026-03-26 01:36:38,588][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 24 and human policies 1. +[2026-03-26 01:36:38,588][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:36:55,868][__main__][INFO] - Number of regex retries in iteration 246: 0 +[2026-03-26 01:36:55,869][__main__][INFO] - agents played in iteration 246 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:36:56,639][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:36:56,662][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:36:56,685][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:36:56,709][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:36:56,711][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:36:56,711][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:36:57,366][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:36:57,823][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:36:58,334][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:36:58,836][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:36:59,344][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:36:59,852][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:37:00,352][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:37:00,851][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:37:01,354][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:37:01,856][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:37:02,362][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:37:02,863][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:37:03,363][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:37:03,868][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:37:04,371][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:37:04,873][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:37:05,376][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:37:05,879][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:37:06,378][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:37:06,877][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:37:07,376][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:37:07,901][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:37:08,402][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:37:08,903][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:37:09,408][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:37:09,905][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:37:10,405][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:37:10,911][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:37:11,413][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:37:11,912][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:37:12,413][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:37:12,911][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:37:13,409][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:37:13,911][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:37:14,416][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:37:14,917][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:37:15,420][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:37:15,925][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:37:16,426][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:37:16,926][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:37:17,426][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:37:17,925][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:37:18,427][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:37:18,927][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:37:19,426][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:37:19,929][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:37:20,428][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:37:20,928][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:37:21,428][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:37:21,926][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:37:22,424][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:37:22,923][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:37:23,422][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:37:23,919][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:37:24,416][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:37:24,914][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:37:25,415][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:37:25,914][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:37:26,415][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:37:26,912][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:37:27,413][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:37:27,910][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:37:28,407][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:37:28,905][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:37:29,404][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10339 tokens. +[2026-03-26 01:37:30,099][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.26%, ΔTime: 00:00:32 +[2026-03-26 01:37:30,854][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:37:30,856][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:37:30,858][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:37:31,577][__main__][INFO] - Iteration 247 took 52s (32.61% Gen, 66.03% Train). Generation: 17s, Training: 34s. Estimated remaining time: 40h 32m 45s. Estimated total time: 44h 9m 28s. Time estimates for 10 more iterations: 8m 49s, 100 more iterations: 1h 28m 18s, 500 more iterations: 7h 21m 34s. +[2026-03-26 01:37:31,579][__main__][INFO] - Starting iteration 247. +[2026-03-26 01:37:31,979][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 24 and human policies 1. +[2026-03-26 01:37:31,980][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:37:43,135][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:37:45,831][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:37:49,524][__main__][INFO] - Number of regex retries in iteration 247: 2 +[2026-03-26 01:37:49,525][__main__][INFO] - agents played in iteration 247 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:37:50,307][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:37:50,330][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:37:50,354][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:37:50,377][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:37:50,378][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:37:50,378][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:37:51,040][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:37:51,492][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:37:51,997][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:37:52,496][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:37:53,000][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:37:53,502][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:37:54,006][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:37:54,515][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:37:55,021][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:37:55,526][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:37:56,031][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:37:56,535][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:37:57,036][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:37:57,540][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:37:58,046][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:37:58,550][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:37:59,054][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:37:59,555][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:38:00,059][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:38:00,560][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:38:01,059][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:38:01,557][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:38:02,056][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:38:02,555][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:38:03,052][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:38:03,554][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:38:04,054][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:38:04,578][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:38:05,079][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:38:05,580][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:38:06,085][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:38:06,584][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:38:07,087][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:38:07,589][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:38:08,090][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:38:08,590][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:38:09,090][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:38:09,593][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:38:10,091][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:38:10,592][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:38:11,096][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:38:11,599][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:38:12,099][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:38:12,601][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:38:13,101][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:38:13,603][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:38:14,107][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:38:14,610][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:38:15,111][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:38:15,614][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:38:16,112][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:38:16,613][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:38:17,115][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:38:17,617][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:38:18,117][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:38:18,620][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:38:19,128][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:38:19,627][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:38:20,128][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:38:20,631][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:38:21,129][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:38:21,627][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:38:22,128][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:38:22,627][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:38:23,124][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10301 tokens. +[2026-03-26 01:38:23,827][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.43%, ΔTime: 00:00:32 +[2026-03-26 01:38:24,580][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:38:24,583][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:38:24,585][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:38:25,301][__main__][INFO] - Iteration 248 took 53s (32.90% Gen, 65.75% Train). Generation: 17s, Training: 35s. Estimated remaining time: 40h 48m 31s. Estimated total time: 44h 26m 8s. Time estimates for 10 more iterations: 8m 53s, 100 more iterations: 1h 28m 52s, 500 more iterations: 7h 24m 21s. +[2026-03-26 01:38:25,303][__main__][INFO] - Starting iteration 248. +[2026-03-26 01:38:25,704][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 24 and human policies 1. +[2026-03-26 01:38:25,705][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:38:43,123][__main__][INFO] - Number of regex retries in iteration 248: 0 +[2026-03-26 01:38:43,124][__main__][INFO] - agents played in iteration 248 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:38:43,897][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:38:43,921][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:38:43,944][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:38:43,967][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:38:43,967][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:38:43,968][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:38:44,766][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:38:45,224][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:38:45,731][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:38:46,228][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:38:46,726][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:38:47,227][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:38:47,738][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:38:48,246][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:38:48,748][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:38:49,247][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:38:49,748][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:38:50,253][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:38:50,754][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:38:51,254][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:38:51,755][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:38:52,256][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:38:52,755][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:38:53,257][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:38:53,759][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:38:54,256][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:38:54,757][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:38:55,258][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:38:55,758][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:38:56,259][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:38:56,758][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:38:57,256][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:38:57,754][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:38:58,253][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:38:58,750][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:38:59,249][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:38:59,748][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:39:00,246][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:39:00,744][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:39:01,243][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:39:01,744][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:39:02,243][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:39:02,740][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:39:03,243][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:39:03,745][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:39:04,246][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:39:04,748][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:39:05,246][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:39:05,743][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:39:06,243][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:39:06,741][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:39:07,247][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:39:07,750][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:39:08,247][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:39:08,746][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:39:09,247][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:39:09,744][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:39:10,241][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:39:10,737][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:39:11,234][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:39:11,730][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:39:12,227][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:39:12,726][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:39:13,223][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:39:13,724][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:39:14,221][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:39:14,720][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:39:15,219][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:39:15,719][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:39:16,215][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:39:16,714][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10223 tokens. +[2026-03-26 01:39:17,404][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.61%, Block Peak % of device VRAM: 62.28%, ΔTime: 00:00:32 +[2026-03-26 01:39:18,155][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:39:18,157][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:39:18,159][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:39:18,964][__main__][INFO] - Iteration 249 took 53s (32.71% Gen, 65.78% Train). Generation: 17s, Training: 35s. Estimated remaining time: 40h 44m 30s. Estimated total time: 44h 23m 0s. Time estimates for 10 more iterations: 8m 52s, 100 more iterations: 1h 28m 46s, 500 more iterations: 7h 23m 50s. +[2026-03-26 01:39:18,966][__main__][INFO] - Starting iteration 249. +[2026-03-26 01:39:19,370][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 24 and human policies 1. +[2026-03-26 01:39:19,370][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:39:23,444][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:39:37,273][__main__][INFO] - Number of regex retries in iteration 249: 1 +[2026-03-26 01:39:37,274][__main__][INFO] - agents played in iteration 249 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:39:38,058][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:39:38,082][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:39:38,105][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:39:38,128][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:39:38,128][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:39:38,129][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:39:38,783][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:39:39,239][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:39:39,744][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:39:40,245][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:39:40,747][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:39:41,250][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:39:41,753][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:39:42,254][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:39:42,754][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:39:43,261][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:39:43,764][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:39:44,267][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:39:44,771][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:39:45,271][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:39:45,776][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:39:46,280][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:39:46,782][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:39:47,284][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:39:47,782][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:39:48,287][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:39:48,785][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:39:49,284][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:39:49,781][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:39:50,282][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:39:50,780][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:39:51,284][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:39:51,788][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:39:52,292][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:39:52,793][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:39:53,298][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:39:53,800][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:39:54,303][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:39:54,805][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:39:55,303][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:39:55,799][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:39:56,297][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:39:56,794][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:39:57,295][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:39:57,792][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:39:58,292][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:39:58,792][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:39:59,290][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:39:59,791][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:40:00,289][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:40:00,791][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:40:01,292][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:40:01,791][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:40:02,288][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:40:02,787][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:40:03,287][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:40:03,784][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:40:04,282][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:40:04,779][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:40:05,281][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:40:05,778][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:40:06,274][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:40:06,772][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:40:07,269][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:40:07,770][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:40:08,272][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:40:08,769][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:40:09,266][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:40:09,764][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:40:10,263][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:40:10,758][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10244 tokens. +[2026-03-26 01:40:11,456][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.05%, Current % of VRAM taken: 60.48%, Block Peak % of device VRAM: 62.44%, ΔTime: 00:00:32 +[2026-03-26 01:40:12,240][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:40:12,242][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:40:12,244][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:40:13,091][__main__][INFO] - Iteration 250 took 53s (33.33% Gen, 65.10% Train). Generation: 17s, Training: 34s. Estimated remaining time: 41h 6m 40s. Estimated total time: 44h 46m 5s. Time estimates for 10 more iterations: 8m 57s, 100 more iterations: 1h 29m 32s, 500 more iterations: 7h 27m 40s. +[2026-03-26 01:40:13,093][__main__][INFO] - Starting iteration 250. +[2026-03-26 01:40:13,496][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 24 and human policies 1. +[2026-03-26 01:40:13,497][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:40:30,807][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:40:31,644][__main__][INFO] - Number of regex retries in iteration 250: 1 +[2026-03-26 01:40:31,645][__main__][INFO] - agents played in iteration 250 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:40:32,432][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:40:32,455][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:40:32,478][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:40:32,501][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:40:32,502][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:40:32,502][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:40:33,156][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:40:33,611][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:40:34,113][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:40:34,613][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:40:35,120][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:40:35,623][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:40:36,126][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:40:36,634][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:40:37,135][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:40:37,637][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:40:38,143][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:40:38,644][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:40:39,148][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:40:39,647][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:40:40,145][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:40:40,644][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:40:41,152][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:40:41,653][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:40:42,156][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:40:42,654][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:40:43,159][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:40:43,661][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:40:44,159][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:40:44,663][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:40:45,168][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:40:45,668][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:40:46,168][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:40:46,666][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:40:47,166][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:40:47,666][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:40:48,165][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:40:48,665][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:40:49,167][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:40:49,668][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:40:50,168][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:40:50,668][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:40:51,170][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:40:51,667][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:40:52,167][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:40:52,669][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:40:53,169][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:40:53,669][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:40:54,168][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:40:54,667][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:40:55,169][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:40:55,666][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:40:56,165][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:40:56,665][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:40:57,162][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:40:57,662][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:40:58,164][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:40:58,665][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:40:59,164][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:40:59,662][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:41:00,159][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:41:00,660][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:41:01,157][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:41:01,653][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:41:02,149][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:41:02,646][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:41:03,144][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:41:03,640][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:41:04,140][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:41:04,640][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:41:05,137][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10204 tokens. +[2026-03-26 01:41:05,834][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.04%, Current % of VRAM taken: 60.47%, Block Peak % of device VRAM: 62.17%, ΔTime: 00:00:32 +[2026-03-26 01:41:06,595][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:41:06,597][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:41:06,599][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:41:08,147][__main__][INFO] - Iteration 251 took 54s (33.21% Gen, 63.96% Train). Generation: 18s, Training: 34s. Estimated remaining time: 41h 52m 16s. Estimated total time: 45h 32m 35s. Time estimates for 10 more iterations: 9m 6s, 100 more iterations: 1h 31m 5s, 500 more iterations: 7h 35m 25s. +[2026-03-26 01:41:08,150][__main__][INFO] - Starting iteration 251. +[2026-03-26 01:41:08,551][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 25 and human policies 1. +[2026-03-26 01:41:08,552][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:41:26,031][__main__][INFO] - Number of regex retries in iteration 251: 0 +[2026-03-26 01:41:26,032][__main__][INFO] - agents played in iteration 251 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:41:26,814][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:41:26,838][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:41:26,861][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:41:26,884][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:41:26,884][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:41:26,885][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:41:27,563][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:41:28,015][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:41:28,523][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:41:29,024][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:41:29,527][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:41:30,031][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:41:30,539][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:41:31,041][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:41:31,545][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:41:32,047][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:41:32,556][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:41:33,059][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:41:33,564][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:41:34,067][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:41:34,572][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:41:35,077][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:41:35,598][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:41:36,097][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:41:36,596][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:41:37,096][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:41:37,594][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:41:38,093][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:41:38,592][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:41:39,090][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:41:39,595][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:41:40,095][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:41:40,596][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:41:41,096][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:41:41,597][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:41:42,098][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:41:42,598][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:41:43,096][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:41:43,597][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:41:44,098][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:41:44,599][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:41:45,099][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:41:45,602][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:41:46,103][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:41:46,608][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:41:47,112][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:41:47,613][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:41:48,117][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:41:48,619][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:41:49,123][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:41:49,628][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:41:50,128][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:41:50,627][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:41:51,127][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:41:51,627][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:41:52,123][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:41:52,623][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:41:53,121][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:41:53,617][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:41:54,114][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:41:54,614][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:41:55,114][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:41:55,611][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:41:56,113][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:41:56,609][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:41:57,110][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:41:57,612][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:41:58,109][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:41:58,606][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:41:59,103][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:41:59,600][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10262 tokens. +[2026-03-26 01:42:00,307][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.22%, Current % of VRAM taken: 60.64%, Block Peak % of device VRAM: 62.44%, ΔTime: 00:00:32 +[2026-03-26 01:42:01,072][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:42:01,075][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:42:01,077][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:42:01,805][__main__][INFO] - Iteration 252 took 53s (32.82% Gen, 65.81% Train). Generation: 17s, Training: 35s. Estimated remaining time: 40h 41m 31s. Estimated total time: 44h 22m 44s. Time estimates for 10 more iterations: 8m 52s, 100 more iterations: 1h 28m 45s, 500 more iterations: 7h 23m 47s. +[2026-03-26 01:42:01,807][__main__][INFO] - Starting iteration 252. +[2026-03-26 01:42:02,207][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 25 and human policies 1. +[2026-03-26 01:42:02,207][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:42:10,571][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:42:16,891][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 1, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:42:19,501][__main__][INFO] - Number of regex retries in iteration 252: 2 +[2026-03-26 01:42:19,502][__main__][INFO] - agents played in iteration 252 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:42:20,285][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.59%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:42:20,308][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.59%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:42:20,331][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.59%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:42:20,354][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.59%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:42:20,355][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:42:20,356][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:42:21,014][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:42:21,469][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:42:21,970][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:42:22,469][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:42:22,974][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:42:23,478][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:42:23,981][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:42:24,486][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:42:24,990][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:42:25,494][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:42:25,997][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:42:26,499][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:42:27,002][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:42:27,501][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:42:28,000][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:42:28,503][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:42:29,006][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:42:29,503][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:42:30,001][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:42:30,497][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:42:30,993][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:42:31,492][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:42:31,990][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:42:32,487][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:42:32,994][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:42:33,491][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:42:33,989][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:42:34,489][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:42:34,986][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:42:35,485][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:42:35,984][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:42:36,482][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:42:36,982][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:42:37,479][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:42:37,977][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:42:38,475][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:42:38,973][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:42:39,470][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:42:39,968][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:42:40,469][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:42:40,968][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:42:41,470][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:42:41,969][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:42:42,467][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:42:42,966][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:42:43,468][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:42:43,966][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:42:44,462][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:42:44,959][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:42:45,462][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:42:45,964][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:42:46,460][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:42:46,960][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:42:47,456][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:42:47,956][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:42:48,455][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:42:48,956][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:42:49,455][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:42:49,953][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:42:50,454][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:42:50,954][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:42:51,451][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:42:51,948][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:42:52,446][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:42:52,946][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10290 tokens. +[2026-03-26 01:42:53,648][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.16%, Current % of VRAM taken: 60.58%, Block Peak % of device VRAM: 62.23%, ΔTime: 00:00:32 +[2026-03-26 01:42:54,602][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:42:54,606][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:42:54,608][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:42:55,315][__main__][INFO] - Iteration 253 took 53s (32.56% Gen, 66.10% Train). Generation: 17s, Training: 35s. Estimated remaining time: 40h 33m 23s. Estimated total time: 44h 15m 29s. Time estimates for 10 more iterations: 8m 51s, 100 more iterations: 1h 28m 30s, 500 more iterations: 7h 22m 34s. +[2026-03-26 01:42:55,318][__main__][INFO] - Starting iteration 253. +[2026-03-26 01:42:55,738][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 25 and human policies 1. +[2026-03-26 01:42:55,739][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:43:12,578][__main__][INFO] - Number of regex retries in iteration 253: 0 +[2026-03-26 01:43:12,579][__main__][INFO] - agents played in iteration 253 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:43:13,357][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:43:13,380][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:43:13,404][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:43:13,427][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:43:13,428][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:43:13,428][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:43:14,095][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:43:14,553][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:43:15,063][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:43:15,565][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:43:16,065][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:43:16,565][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:43:17,065][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:43:17,572][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:43:18,077][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:43:18,579][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:43:19,077][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:43:19,577][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:43:20,077][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:43:20,577][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:43:21,077][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:43:21,578][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:43:22,081][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:43:22,582][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:43:23,082][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:43:23,582][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:43:24,078][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:43:24,577][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:43:25,078][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:43:25,583][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:43:26,082][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:43:26,582][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:43:27,081][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:43:27,584][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:43:28,087][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:43:28,588][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:43:29,088][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:43:29,587][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:43:30,087][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:43:30,587][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:43:31,089][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:43:31,586][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:43:32,089][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:43:32,590][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:43:33,090][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:43:33,591][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:43:34,095][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:43:34,599][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:43:35,102][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:43:35,603][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:43:36,103][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:43:36,601][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:43:37,106][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:43:37,606][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:43:38,104][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:43:38,605][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:43:39,105][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:43:39,604][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:43:40,107][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:43:40,605][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:43:41,105][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:43:41,604][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:43:42,104][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:43:42,604][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:43:43,101][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:43:43,601][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:43:44,102][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:43:44,600][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:43:45,097][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:43:45,596][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:43:46,094][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10338 tokens. +[2026-03-26 01:43:46,808][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.12%, ΔTime: 00:00:32 +[2026-03-26 01:43:47,572][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:43:47,574][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:43:47,576][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:43:48,344][__main__][INFO] - Iteration 254 took 52s (32.01% Gen, 66.53% Train). Generation: 16s, Training: 34s. Estimated remaining time: 40h 7m 21s. Estimated total time: 43h 50m 21s. Time estimates for 10 more iterations: 8m 46s, 100 more iterations: 1h 27m 40s, 500 more iterations: 7h 18m 23s. +[2026-03-26 01:43:48,346][__main__][INFO] - Starting iteration 254. +[2026-03-26 01:43:49,149][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 25 and human policies 1. +[2026-03-26 01:43:49,150][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:44:10,567][__main__][INFO] - Number of regex retries in iteration 254: 0 +[2026-03-26 01:44:10,568][__main__][INFO] - agents played in iteration 254 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:44:11,372][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:44:11,396][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:44:11,419][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:44:11,442][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:44:11,443][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:44:11,443][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:44:12,107][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:44:12,565][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:44:13,069][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:44:13,571][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:44:14,076][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:44:14,573][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:44:15,073][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:44:15,574][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:44:16,074][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:44:16,575][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:44:17,077][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:44:17,575][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:44:18,073][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:44:18,574][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:44:19,076][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:44:19,576][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:44:20,076][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:44:20,578][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:44:21,078][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:44:21,577][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:44:22,079][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:44:22,579][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:44:23,076][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:44:23,576][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:44:24,074][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:44:24,572][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:44:25,072][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:44:25,573][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:44:26,074][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:44:26,576][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:44:27,073][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:44:27,570][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:44:28,068][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:44:28,568][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:44:29,066][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:44:29,563][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:44:30,060][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:44:30,562][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:44:31,058][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:44:31,557][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:44:32,057][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:44:32,557][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:44:33,056][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:44:33,555][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:44:34,054][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:44:34,554][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:44:35,057][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:44:35,560][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:44:36,060][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:44:36,557][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:44:37,053][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:44:37,549][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:44:38,044][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:44:38,539][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:44:39,037][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:44:39,537][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:44:40,036][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:44:40,533][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:44:41,032][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:44:41,530][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:44:42,030][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:44:42,536][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:44:43,041][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:44:43,540][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:44:44,040][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10197 tokens. +[2026-03-26 01:44:44,739][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.24%, Current % of VRAM taken: 60.67%, Block Peak % of device VRAM: 62.26%, ΔTime: 00:00:32 +[2026-03-26 01:44:45,498][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:44:45,501][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:44:45,502][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:44:46,363][__main__][INFO] - Iteration 255 took 57s (37.17% Gen, 60.63% Train). Generation: 21s, Training: 34s. Estimated remaining time: 44h 16m 49s. Estimated total time: 48h 0m 47s. Time estimates for 10 more iterations: 9m 36s, 100 more iterations: 1h 36m 1s, 500 more iterations: 8h 0m 7s. +[2026-03-26 01:44:46,365][__main__][INFO] - Starting iteration 255. +[2026-03-26 01:44:46,765][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 25 and human policies 1. +[2026-03-26 01:44:46,766][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:45:03,788][__main__][INFO] - Number of regex retries in iteration 255: 0 +[2026-03-26 01:45:03,789][__main__][INFO] - agents played in iteration 255 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:45:04,572][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:45:04,595][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:45:04,619][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:45:04,642][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:45:04,642][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:45:04,643][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:45:05,306][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:45:05,764][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:45:06,270][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:45:06,769][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:45:07,274][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:45:07,775][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:45:08,275][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:45:08,779][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:45:09,286][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:45:09,792][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:45:10,301][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:45:10,801][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:45:11,305][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:45:11,808][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:45:12,309][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:45:12,816][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:45:13,316][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:45:13,816][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:45:14,316][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:45:14,815][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:45:15,312][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:45:15,809][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:45:16,306][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:45:16,802][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:45:17,301][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:45:17,799][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:45:18,303][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:45:18,804][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:45:19,306][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:45:19,806][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:45:20,308][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:45:20,810][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:45:21,310][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:45:21,809][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:45:22,310][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:45:22,809][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:45:23,307][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:45:23,805][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:45:24,303][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:45:24,806][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:45:25,304][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:45:25,804][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:45:26,307][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:45:26,806][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:45:27,304][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:45:27,801][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:45:28,298][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:45:28,798][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:45:29,298][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:45:29,800][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:45:30,297][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:45:30,799][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:45:31,297][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:45:31,795][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:45:32,295][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:45:32,795][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:45:33,296][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:45:33,795][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:45:34,293][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:45:34,790][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:45:35,290][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:45:35,789][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:45:36,287][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:45:36,787][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:45:37,293][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10235 tokens. +[2026-03-26 01:45:37,996][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.15%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.29%, ΔTime: 00:00:32 +[2026-03-26 01:45:38,765][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:45:38,767][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:45:38,769][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:45:39,599][__main__][INFO] - Iteration 256 took 52s (32.22% Gen, 66.21% Train). Generation: 17s, Training: 34s. Estimated remaining time: 40h 16m 51s. Estimated total time: 44h 1m 42s. Time estimates for 10 more iterations: 8m 48s, 100 more iterations: 1h 28m 3s, 500 more iterations: 7h 20m 17s. +[2026-03-26 01:45:39,601][__main__][INFO] - Starting iteration 256. +[2026-03-26 01:45:40,001][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 25 and human policies 1. +[2026-03-26 01:45:40,002][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:46:00,725][__main__][INFO] - Number of regex retries in iteration 256: 0 +[2026-03-26 01:46:00,726][__main__][INFO] - agents played in iteration 256 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:46:01,519][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:46:01,543][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:46:01,566][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:46:01,589][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:46:01,589][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:46:01,590][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:46:02,251][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:46:02,708][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:46:03,218][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:46:03,718][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:46:04,215][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:46:04,716][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:46:05,216][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:46:05,721][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:46:06,221][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:46:06,727][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:46:07,228][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:46:07,729][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:46:08,230][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:46:08,732][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:46:09,230][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:46:09,730][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:46:10,229][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:46:10,732][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:46:11,236][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:46:11,735][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:46:12,233][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:46:12,734][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:46:13,231][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:46:13,729][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:46:14,227][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:46:14,726][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:46:15,226][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:46:15,724][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:46:16,222][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:46:16,720][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:46:17,218][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:46:17,716][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:46:18,217][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:46:18,739][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:46:19,240][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:46:19,742][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:46:20,246][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:46:20,749][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:46:21,250][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:46:21,752][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:46:22,252][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:46:22,751][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:46:23,252][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:46:23,753][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:46:24,254][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:46:24,756][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:46:25,253][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:46:25,753][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:46:26,253][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:46:26,751][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:46:27,249][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:46:27,747][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:46:28,253][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:46:28,756][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:46:29,256][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:46:29,758][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:46:30,256][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:46:30,752][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:46:31,251][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:46:31,751][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:46:32,248][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:46:32,745][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:46:33,246][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:46:33,743][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:46:34,241][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10185 tokens. +[2026-03-26 01:46:34,937][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.29%, ΔTime: 00:00:32 +[2026-03-26 01:46:35,680][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:46:35,682][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:46:35,685][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:46:36,401][__main__][INFO] - Iteration 257 took 56s (36.74% Gen, 61.98% Train). Generation: 20s, Training: 34s. Estimated remaining time: 43h 14m 16s. Estimated total time: 47h 0m 3s. Time estimates for 10 more iterations: 9m 24s, 100 more iterations: 1h 34m 0s, 500 more iterations: 7h 50m 0s. +[2026-03-26 01:46:36,404][__main__][INFO] - Starting iteration 257. +[2026-03-26 01:46:36,805][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 25 and human policies 1. +[2026-03-26 01:46:36,806][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:46:41,856][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:46:55,478][__main__][INFO] - Number of regex retries in iteration 257: 1 +[2026-03-26 01:46:55,479][__main__][INFO] - agents played in iteration 257 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:46:56,268][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:46:56,292][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:46:56,315][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:46:56,338][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:46:56,338][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:46:56,339][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:46:57,000][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:46:57,458][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:46:57,965][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:46:58,470][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:46:58,971][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:46:59,473][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:46:59,976][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:47:00,476][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:47:00,976][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:47:01,475][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:47:01,978][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:47:02,476][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:47:02,976][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:47:03,476][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:47:03,972][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:47:04,469][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:47:04,968][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:47:05,472][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:47:05,990][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:47:06,492][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:47:06,993][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:47:07,493][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:47:07,997][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:47:08,498][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:47:08,998][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:47:09,496][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:47:09,994][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:47:10,493][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:47:10,990][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:47:11,492][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:47:11,990][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:47:12,489][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:47:12,988][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:47:13,489][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:47:13,993][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:47:14,493][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:47:14,993][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:47:15,492][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:47:15,993][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:47:16,495][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:47:16,996][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:47:17,498][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:47:17,999][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:47:18,499][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:47:19,000][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:47:19,502][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:47:20,009][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:47:20,516][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:47:21,026][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:47:21,525][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:47:22,026][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:47:22,526][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:47:23,024][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:47:23,525][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:47:24,026][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:47:24,523][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:47:25,024][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:47:25,521][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:47:26,019][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:47:26,518][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:47:27,015][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:47:27,513][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:47:28,013][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:47:28,512][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:47:29,014][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10472 tokens. +[2026-03-26 01:47:29,706][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.25%, Current % of VRAM taken: 60.68%, Block Peak % of device VRAM: 62.39%, ΔTime: 00:00:32 +[2026-03-26 01:47:30,446][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:47:30,449][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:47:30,451][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:47:31,142][__main__][INFO] - Iteration 258 took 54s (34.36% Gen, 64.36% Train). Generation: 18s, Training: 34s. Estimated remaining time: 41h 30m 11s. Estimated total time: 45h 16m 53s. Time estimates for 10 more iterations: 9m 3s, 100 more iterations: 1h 30m 33s, 500 more iterations: 7h 32m 48s. +[2026-03-26 01:47:31,144][__main__][INFO] - Starting iteration 258. +[2026-03-26 01:47:31,544][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 25 and human policies 1. +[2026-03-26 01:47:31,544][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:47:36,714][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:47:48,299][__main__][INFO] - Number of regex retries in iteration 258: 1 +[2026-03-26 01:47:48,299][__main__][INFO] - agents played in iteration 258 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:47:49,091][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:47:49,114][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:47:49,138][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:47:49,161][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:47:49,162][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:47:49,162][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:47:49,835][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:47:50,301][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:47:50,808][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:47:51,314][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:47:51,819][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:47:52,323][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:47:52,831][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:47:53,337][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:47:53,841][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:47:54,343][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:47:54,843][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:47:55,342][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:47:55,839][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:47:56,337][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:47:56,836][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:47:57,334][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:47:57,831][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:47:58,331][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:47:58,833][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:47:59,335][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:47:59,836][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:48:00,337][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:48:00,839][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:48:01,340][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:48:01,837][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:48:02,336][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:48:02,841][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:48:03,341][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:48:03,840][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:48:04,337][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:48:04,833][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:48:05,333][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:48:05,832][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:48:06,331][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:48:06,832][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:48:07,329][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:48:07,832][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:48:08,334][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:48:08,832][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:48:09,335][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:48:09,835][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:48:10,336][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:48:10,836][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:48:11,335][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:48:11,835][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:48:12,335][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:48:12,837][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:48:13,342][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:48:13,848][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:48:14,345][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:48:14,846][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:48:15,346][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:48:15,842][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:48:16,338][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:48:16,835][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:48:17,332][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:48:17,829][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:48:18,327][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:48:18,829][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:48:19,331][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:48:19,829][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:48:20,330][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:48:20,832][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:48:21,328][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:48:21,826][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10383 tokens. +[2026-03-26 01:48:22,544][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.61%, Block Peak % of device VRAM: 62.29%, ΔTime: 00:00:32 +[2026-03-26 01:48:23,294][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:48:23,296][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:48:23,298][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:48:24,024][__main__][INFO] - Iteration 259 took 52s (31.93% Gen, 66.69% Train). Generation: 16s, Training: 34s. Estimated remaining time: 39h 56m 25s. Estimated total time: 43h 44m 0s. Time estimates for 10 more iterations: 8m 44s, 100 more iterations: 1h 27m 28s, 500 more iterations: 7h 17m 20s. +[2026-03-26 01:48:24,026][__main__][INFO] - Starting iteration 259. +[2026-03-26 01:48:24,423][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 25 and human policies 1. +[2026-03-26 01:48:24,424][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:48:40,578][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:48:41,369][__main__][INFO] - Number of regex retries in iteration 259: 1 +[2026-03-26 01:48:41,370][__main__][INFO] - agents played in iteration 259 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:48:42,150][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:48:42,173][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:48:42,197][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:48:42,219][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:48:42,220][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:48:42,221][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:48:42,879][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:48:43,344][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:48:43,852][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:48:44,356][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:48:44,854][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:48:45,353][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:48:45,854][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:48:46,357][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:48:46,859][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:48:47,357][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:48:47,857][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:48:48,359][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:48:48,859][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:48:49,355][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:48:49,853][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:48:50,357][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:48:50,856][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:48:51,355][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:48:51,852][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:48:52,353][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:48:52,859][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:48:53,357][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:48:53,861][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:48:54,359][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:48:54,857][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:48:55,356][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:48:55,853][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:48:56,353][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:48:56,850][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:48:57,351][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:48:57,849][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:48:58,348][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:48:58,846][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:48:59,344][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:48:59,846][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:49:00,345][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:49:00,846][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:49:01,345][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:49:01,842][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:49:02,342][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:49:02,843][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:49:03,344][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:49:03,845][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:49:04,349][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:49:04,853][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:49:05,358][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:49:05,860][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:49:06,360][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:49:06,859][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:49:07,359][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:49:07,858][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:49:08,354][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:49:08,854][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:49:09,351][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:49:09,850][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:49:10,352][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:49:10,848][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:49:11,349][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:49:11,850][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:49:12,352][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:49:12,852][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:49:13,350][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:49:13,850][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:49:14,351][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:49:14,852][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10471 tokens. +[2026-03-26 01:49:15,571][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.20%, Current % of VRAM taken: 60.63%, Block Peak % of device VRAM: 62.22%, ΔTime: 00:00:32 +[2026-03-26 01:49:16,318][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:49:16,321][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:49:16,323][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:49:17,040][__main__][INFO] - Iteration 260 took 52s (32.21% Gen, 66.43% Train). Generation: 16s, Training: 34s. Estimated remaining time: 40h 2m 25s. Estimated total time: 43h 50m 53s. Time estimates for 10 more iterations: 8m 46s, 100 more iterations: 1h 27m 41s, 500 more iterations: 7h 18m 28s. +[2026-03-26 01:49:17,043][__main__][INFO] - Starting iteration 260. +[2026-03-26 01:49:17,444][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 25 and human policies 1. +[2026-03-26 01:49:17,445][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:49:35,442][__main__][INFO] - Number of regex retries in iteration 260: 0 +[2026-03-26 01:49:35,443][__main__][INFO] - agents played in iteration 260 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:49:36,233][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:49:36,256][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:49:36,280][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:49:36,303][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:49:36,304][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:49:36,304][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:49:36,977][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:49:37,440][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:49:37,953][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:49:38,459][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:49:38,960][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:49:39,463][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:49:39,972][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:49:40,474][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:49:40,979][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:49:41,479][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:49:41,988][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:49:42,490][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:49:42,995][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:49:43,495][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:49:43,997][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:49:44,497][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:49:44,997][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:49:45,504][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:49:46,002][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:49:46,502][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:49:46,999][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:49:47,499][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:49:47,996][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:49:48,496][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:49:49,001][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:49:49,499][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:49:49,998][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:49:50,496][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:49:50,994][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:49:51,492][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:49:51,990][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:49:52,488][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:49:52,995][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:49:53,493][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:49:53,995][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:49:54,496][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:49:54,995][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:49:55,498][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:49:55,995][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:49:56,498][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:49:57,000][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:49:57,503][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:49:58,010][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:49:58,509][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:49:59,011][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:49:59,514][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:50:00,011][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:50:00,508][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:50:01,007][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:50:01,506][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:50:02,005][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:50:02,507][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:50:03,005][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:50:03,509][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:50:04,009][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:50:04,512][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:50:05,009][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:50:05,510][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:50:06,009][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:50:06,507][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:50:07,008][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:50:07,511][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:50:08,009][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:50:08,510][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:50:09,013][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10364 tokens. +[2026-03-26 01:50:09,728][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.23%, Current % of VRAM taken: 60.66%, Block Peak % of device VRAM: 62.21%, ΔTime: 00:00:32 +[2026-03-26 01:50:10,478][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:50:10,481][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:50:10,482][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:50:11,837][__main__][INFO] - Iteration 261 took 54s (33.09% Gen, 64.42% Train). Generation: 17s, Training: 35s. Estimated remaining time: 41h 30m 15s. Estimated total time: 45h 19m 38s. Time estimates for 10 more iterations: 9m 3s, 100 more iterations: 1h 30m 39s, 500 more iterations: 7h 33m 16s. +[2026-03-26 01:50:11,839][__main__][INFO] - Starting iteration 261. +[2026-03-26 01:50:12,240][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 26 and human policies 1. +[2026-03-26 01:50:12,240][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:50:29,140][__main__][INFO] - Number of regex retries in iteration 261: 0 +[2026-03-26 01:50:29,141][__main__][INFO] - agents played in iteration 261 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:50:29,909][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:50:29,933][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:50:29,956][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:50:29,979][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:50:29,979][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:50:29,980][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:50:30,666][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:50:31,118][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:50:31,627][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:50:32,133][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:50:32,635][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:50:33,141][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:50:33,645][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:50:34,145][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:50:34,647][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:50:35,147][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:50:35,649][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:50:36,153][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:50:36,656][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:50:37,158][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:50:37,658][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:50:38,159][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:50:38,656][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:50:39,161][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:50:39,658][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:50:40,161][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:50:40,657][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:50:41,160][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:50:41,661][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:50:42,159][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:50:42,658][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:50:43,160][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:50:43,692][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:50:44,194][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:50:44,699][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:50:45,201][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:50:45,705][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:50:46,208][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:50:46,711][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:50:47,221][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:50:47,731][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:50:48,240][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:50:48,743][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:50:49,244][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:50:49,747][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:50:50,250][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:50:50,752][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:50:51,253][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:50:51,756][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:50:52,256][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:50:52,756][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:50:53,254][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:50:53,754][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:50:54,256][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:50:54,756][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:50:55,253][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:50:55,752][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:50:56,250][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:50:56,753][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:50:57,255][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:50:57,754][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:50:58,256][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:50:58,756][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:50:59,252][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:50:59,752][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:51:00,251][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:51:00,748][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:51:01,245][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:51:01,741][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:51:02,239][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:51:02,739][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10474 tokens. +[2026-03-26 01:51:03,458][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.20%, Current % of VRAM taken: 60.63%, Block Peak % of device VRAM: 62.53%, ΔTime: 00:00:32 +[2026-03-26 01:51:04,251][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:51:04,253][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:51:04,255][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:51:05,161][__main__][INFO] - Iteration 262 took 52s (31.93% Gen, 66.35% Train). Generation: 16s, Training: 35s. Estimated remaining time: 40h 15m 49s. Estimated total time: 44h 6m 6s. Time estimates for 10 more iterations: 8m 49s, 100 more iterations: 1h 28m 12s, 500 more iterations: 7h 21m 1s. +[2026-03-26 01:51:05,163][__main__][INFO] - Starting iteration 262. +[2026-03-26 01:51:05,562][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 26 and human policies 1. +[2026-03-26 01:51:05,563][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:51:10,568][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:51:23,960][__main__][INFO] - Number of regex retries in iteration 262: 1 +[2026-03-26 01:51:23,961][__main__][INFO] - agents played in iteration 262 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:51:24,741][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:51:24,764][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:51:24,788][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:51:24,811][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:51:24,811][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:51:24,812][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:51:25,471][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:51:25,928][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:51:26,437][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:51:26,934][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:51:27,436][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:51:27,939][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:51:28,441][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:51:28,942][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:51:29,446][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:51:29,946][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:51:30,447][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:51:30,948][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:51:31,453][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:51:31,951][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:51:32,451][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:51:32,954][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:51:33,454][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:51:33,952][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:51:34,448][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:51:34,945][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:51:35,441][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:51:35,938][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:51:36,435][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:51:36,931][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:51:37,428][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:51:37,933][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:51:38,437][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:51:38,942][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:51:39,451][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:51:39,950][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:51:40,452][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:51:40,951][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:51:41,449][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:51:41,950][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:51:42,471][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:51:42,971][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:51:43,469][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:51:43,967][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:51:44,470][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:51:44,970][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:51:45,472][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:51:45,973][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:51:46,472][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:51:46,972][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:51:47,476][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:51:47,976][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:51:48,474][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:51:48,981][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:51:49,484][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:51:49,981][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:51:50,479][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:51:50,977][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:51:51,478][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:51:51,975][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:51:52,475][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:51:52,976][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:51:53,473][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:51:53,971][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:51:54,470][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:51:54,970][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:51:55,469][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:51:55,970][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:51:56,469][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:51:56,967][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:51:57,468][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10340 tokens. +[2026-03-26 01:51:58,183][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.20%, Current % of VRAM taken: 60.63%, Block Peak % of device VRAM: 62.14%, ΔTime: 00:00:32 +[2026-03-26 01:51:58,929][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:51:58,931][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:51:58,933][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:51:59,779][__main__][INFO] - Iteration 263 took 54s (33.93% Gen, 64.50% Train). Generation: 18s, Training: 34s. Estimated remaining time: 41h 19m 42s. Estimated total time: 45h 10m 53s. Time estimates for 10 more iterations: 9m 2s, 100 more iterations: 1h 30m 21s, 500 more iterations: 7h 31m 48s. +[2026-03-26 01:51:59,781][__main__][INFO] - Starting iteration 263. +[2026-03-26 01:52:00,180][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 26 and human policies 1. +[2026-03-26 01:52:00,181][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:52:13,501][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:52:18,888][__main__][INFO] - Number of regex retries in iteration 263: 1 +[2026-03-26 01:52:18,889][__main__][INFO] - agents played in iteration 263 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:52:19,673][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:52:19,696][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:52:19,719][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:52:19,742][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:52:19,743][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:52:19,743][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:52:20,415][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:52:20,874][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:52:21,382][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:52:21,887][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:52:22,391][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:52:22,891][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:52:23,396][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:52:23,894][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:52:24,394][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:52:24,899][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:52:25,398][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:52:25,900][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:52:26,418][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:52:26,921][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:52:27,426][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:52:27,926][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:52:28,427][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:52:28,925][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:52:29,428][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:52:29,929][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:52:30,437][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:52:30,944][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:52:31,446][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:52:31,948][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:52:32,446][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:52:32,948][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:52:33,446][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:52:33,944][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:52:34,444][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:52:34,946][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:52:35,442][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:52:35,941][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:52:36,439][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:52:36,939][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:52:37,443][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:52:37,945][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:52:38,448][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:52:38,949][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:52:39,449][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:52:39,958][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:52:40,460][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:52:40,957][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:52:41,455][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:52:41,954][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:52:42,452][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:52:42,950][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:52:43,447][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:52:43,945][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:52:44,444][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:52:44,941][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:52:45,444][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:52:45,951][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:52:46,449][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:52:46,951][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:52:47,453][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:52:47,950][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:52:48,452][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:52:48,952][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:52:49,449][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:52:49,950][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:52:50,453][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:52:50,952][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:52:51,450][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:52:51,950][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:52:52,451][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10362 tokens. +[2026-03-26 01:52:53,167][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.23%, Current % of VRAM taken: 60.65%, Block Peak % of device VRAM: 62.26%, ΔTime: 00:00:32 +[2026-03-26 01:52:53,807][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:52:53,810][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:52:53,811][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:52:54,540][__main__][INFO] - Iteration 264 took 54s (34.42% Gen, 64.24% Train). Generation: 18s, Training: 34s. Estimated remaining time: 41h 25m 55s. Estimated total time: 45h 18m 1s. Time estimates for 10 more iterations: 9m 3s, 100 more iterations: 1h 30m 36s, 500 more iterations: 7h 33m 0s. +[2026-03-26 01:52:54,542][__main__][INFO] - Starting iteration 264. +[2026-03-26 01:52:54,943][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 26 and human policies 1. +[2026-03-26 01:52:54,944][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:53:16,968][__main__][INFO] - Number of regex retries in iteration 264: 0 +[2026-03-26 01:53:16,969][__main__][INFO] - agents played in iteration 264 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:53:17,752][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:53:17,776][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:53:17,799][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:53:17,823][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:53:17,824][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:53:17,824][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:53:18,521][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:53:18,984][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:53:19,494][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:53:19,994][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:53:20,496][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:53:20,997][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:53:21,504][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:53:22,008][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:53:22,513][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:53:23,019][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:53:23,523][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:53:24,024][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:53:24,526][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:53:25,028][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:53:25,531][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:53:26,033][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:53:26,536][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:53:27,033][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:53:27,531][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:53:28,032][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:53:28,529][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:53:29,026][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:53:29,536][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:53:30,039][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:53:30,549][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:53:31,054][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:53:31,557][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:53:32,055][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:53:32,557][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:53:33,057][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:53:33,558][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:53:34,061][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:53:34,568][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:53:35,065][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:53:35,564][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:53:36,061][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:53:36,563][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:53:37,060][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:53:37,561][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:53:38,057][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:53:38,554][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:53:39,055][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:53:39,556][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:53:40,056][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:53:40,559][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:53:41,057][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:53:41,558][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:53:42,055][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:53:42,555][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:53:43,052][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:53:43,552][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:53:44,053][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:53:44,553][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:53:45,053][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:53:45,553][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:53:46,052][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:53:46,572][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:53:47,068][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:53:47,567][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:53:48,070][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:53:48,568][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:53:49,065][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:53:49,563][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:53:50,061][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:53:50,558][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10357 tokens. +[2026-03-26 01:53:51,276][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.19%, ΔTime: 00:00:32 +[2026-03-26 01:53:52,021][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:53:52,023][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:53:52,025][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:53:52,720][__main__][INFO] - Iteration 265 took 57s (38.12% Gen, 60.67% Train). Generation: 22s, Training: 35s. Estimated remaining time: 44h 15m 48s. Estimated total time: 48h 8m 52s. Time estimates for 10 more iterations: 9m 37s, 100 more iterations: 1h 36m 17s, 500 more iterations: 8h 1m 28s. +[2026-03-26 01:53:52,722][__main__][INFO] - Starting iteration 265. +[2026-03-26 01:53:53,125][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 26 and human policies 1. +[2026-03-26 01:53:53,126][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:54:01,560][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:54:11,737][__main__][INFO] - Number of regex retries in iteration 265: 1 +[2026-03-26 01:54:11,738][__main__][INFO] - agents played in iteration 265 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:54:12,523][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:54:12,547][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:54:12,571][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:54:12,594][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:54:12,594][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:54:12,595][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:54:13,274][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:54:13,732][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:54:14,239][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:54:14,740][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:54:15,240][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:54:15,743][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:54:16,249][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:54:16,754][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:54:17,256][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:54:17,757][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:54:18,258][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:54:18,758][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:54:19,255][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:54:19,755][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:54:20,262][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:54:20,766][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:54:21,274][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:54:21,777][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:54:22,274][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:54:22,773][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:54:23,272][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:54:23,771][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:54:24,270][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:54:24,771][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:54:25,271][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:54:25,768][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:54:26,268][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:54:26,768][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:54:27,266][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:54:27,768][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:54:28,267][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:54:28,769][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:54:29,267][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:54:29,774][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:54:30,280][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:54:30,783][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:54:31,302][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:54:31,805][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:54:32,306][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:54:32,805][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:54:33,302][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:54:33,803][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:54:34,304][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:54:34,804][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:54:35,305][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:54:35,805][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:54:36,304][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:54:36,804][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:54:37,300][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:54:37,800][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:54:38,300][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:54:38,801][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:54:39,296][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:54:39,794][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:54:40,294][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:54:40,796][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:54:41,296][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:54:41,792][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:54:42,290][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:54:42,786][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:54:43,286][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:54:43,783][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:54:44,279][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:54:44,778][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:54:45,274][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10439 tokens. +[2026-03-26 01:54:45,999][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 01:54:46,745][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:54:46,747][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:54:46,749][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:54:47,481][__main__][INFO] - Iteration 266 took 54s (34.24% Gen, 64.41% Train). Generation: 18s, Training: 35s. Estimated remaining time: 41h 23m 52s. Estimated total time: 45h 17m 50s. Time estimates for 10 more iterations: 9m 3s, 100 more iterations: 1h 30m 35s, 500 more iterations: 7h 32m 58s. +[2026-03-26 01:54:47,484][__main__][INFO] - Starting iteration 266. +[2026-03-26 01:54:47,882][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 26 and human policies 1. +[2026-03-26 01:54:47,882][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:55:06,344][__main__][INFO] - Number of regex retries in iteration 266: 0 +[2026-03-26 01:55:06,344][__main__][INFO] - agents played in iteration 266 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:55:07,131][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:55:07,154][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:55:07,177][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:55:07,200][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:55:07,201][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:55:07,202][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:55:07,897][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:55:08,355][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:55:08,859][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:55:09,360][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:55:09,865][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:55:10,365][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:55:10,864][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:55:11,363][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:55:11,862][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:55:12,360][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:55:12,860][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:55:13,357][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:55:13,854][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:55:14,356][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:55:14,853][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:55:15,355][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:55:15,853][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:55:16,352][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:55:16,850][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:55:17,358][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:55:17,863][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:55:18,376][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:55:18,875][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:55:19,373][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:55:19,872][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:55:20,373][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:55:20,875][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:55:21,378][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:55:21,880][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:55:22,378][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:55:22,881][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:55:23,379][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:55:23,877][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:55:24,374][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:55:24,876][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:55:25,378][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:55:25,879][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:55:26,378][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:55:26,888][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:55:27,394][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:55:27,895][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:55:28,392][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:55:28,890][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:55:29,389][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:55:29,887][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:55:30,384][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:55:30,883][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:55:31,378][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:55:31,874][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:55:32,375][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:55:32,871][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:55:33,369][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:55:33,867][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:55:34,367][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:55:34,864][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:55:35,360][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:55:35,855][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:55:36,355][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:55:36,853][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:55:37,352][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:55:37,851][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:55:38,351][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:55:38,851][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:55:39,351][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:55:39,852][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10305 tokens. +[2026-03-26 01:55:40,583][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.27%, Current % of VRAM taken: 60.69%, Block Peak % of device VRAM: 62.32%, ΔTime: 00:00:32 +[2026-03-26 01:55:41,335][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:55:41,338][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:55:41,339][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:55:42,195][__main__][INFO] - Iteration 267 took 54s (33.99% Gen, 64.43% Train). Generation: 18s, Training: 34s. Estimated remaining time: 41h 20m 47s. Estimated total time: 45h 15m 40s. Time estimates for 10 more iterations: 9m 3s, 100 more iterations: 1h 30m 31s, 500 more iterations: 7h 32m 36s. +[2026-03-26 01:55:42,197][__main__][INFO] - Starting iteration 267. +[2026-03-26 01:55:42,597][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 26 and human policies 1. +[2026-03-26 01:55:42,598][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:55:47,552][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:55:47,554][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:56:01,269][__main__][INFO] - Number of regex retries in iteration 267: 2 +[2026-03-26 01:56:01,269][__main__][INFO] - agents played in iteration 267 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:56:02,052][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:56:02,076][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:56:02,099][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:56:02,122][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:56:02,123][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:56:02,124][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:56:02,809][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:56:03,266][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:56:03,770][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:56:04,270][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:56:04,768][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:56:05,271][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:56:05,773][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:56:06,273][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:56:06,771][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:56:07,273][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:56:07,770][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:56:08,271][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:56:08,769][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:56:09,267][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:56:09,768][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:56:10,269][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:56:10,766][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:56:11,269][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:56:11,771][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:56:12,269][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:56:12,770][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:56:13,267][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:56:13,768][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:56:14,273][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:56:14,780][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:56:15,281][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:56:15,786][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:56:16,284][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:56:16,783][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:56:17,287][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:56:17,789][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:56:18,289][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:56:18,787][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:56:19,289][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:56:19,789][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:56:20,289][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:56:20,790][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:56:21,290][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:56:21,788][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:56:22,290][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:56:22,790][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:56:23,290][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:56:23,790][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:56:24,289][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:56:24,790][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:56:25,289][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:56:25,790][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:56:26,290][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:56:26,790][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:56:27,288][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:56:27,787][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:56:28,285][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:56:28,786][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:56:29,284][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:56:29,785][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:56:30,283][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:56:30,781][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:56:31,279][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:56:31,779][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:56:32,280][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:56:32,782][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:56:33,282][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:56:33,783][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:56:34,283][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:56:34,787][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10387 tokens. +[2026-03-26 01:56:35,512][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.22%, Current % of VRAM taken: 60.64%, Block Peak % of device VRAM: 62.28%, ΔTime: 00:00:32 +[2026-03-26 01:56:36,255][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:56:36,258][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:56:36,259][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:56:36,924][__main__][INFO] - Iteration 268 took 54s (34.37% Gen, 64.41% Train). Generation: 18s, Training: 34s. Estimated remaining time: 41h 20m 34s. Estimated total time: 45h 16m 22s. Time estimates for 10 more iterations: 9m 3s, 100 more iterations: 1h 30m 32s, 500 more iterations: 7h 32m 43s. +[2026-03-26 01:56:36,926][__main__][INFO] - Starting iteration 268. +[2026-03-26 01:56:37,327][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 26 and human policies 1. +[2026-03-26 01:56:37,328][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:56:54,151][__main__][INFO] - Number of regex retries in iteration 268: 0 +[2026-03-26 01:56:54,152][__main__][INFO] - agents played in iteration 268 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:56:54,938][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:56:54,962][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:56:54,985][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:56:55,008][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:56:55,009][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:56:55,009][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:56:55,685][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:56:56,145][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:56:56,652][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:56:57,153][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:56:57,653][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:56:58,153][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:56:58,655][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:56:59,155][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:56:59,656][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:57:00,158][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:57:00,655][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:57:01,155][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:57:01,656][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:57:02,152][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:57:02,654][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:57:03,154][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:57:03,653][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:57:04,153][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:57:04,650][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:57:05,148][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:57:05,652][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:57:06,151][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:57:06,653][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:57:07,150][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:57:07,646][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:57:08,149][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:57:08,650][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:57:09,150][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:57:09,648][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:57:10,149][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:57:10,647][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:57:11,146][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:57:11,648][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:57:12,148][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:57:12,646][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:57:13,144][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:57:13,642][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:57:14,139][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:57:14,642][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:57:15,139][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:57:15,638][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:57:16,134][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:57:16,630][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:57:17,128][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:57:17,626][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:57:18,122][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:57:18,619][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:57:19,115][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:57:19,612][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:57:20,111][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:57:20,609][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:57:21,106][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:57:21,604][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:57:22,103][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:57:22,601][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:57:23,099][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:57:23,602][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:57:24,101][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:57:24,598][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:57:25,095][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:57:25,593][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:57:26,091][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:57:26,589][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:57:27,087][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:57:27,584][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10202 tokens. +[2026-03-26 01:57:28,300][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 01:57:29,040][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:57:29,043][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:57:29,044][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:57:29,749][__main__][INFO] - Iteration 269 took 52s (32.09% Gen, 66.56% Train). Generation: 16s, Training: 34s. Estimated remaining time: 39h 44m 26s. Estimated total time: 43h 41m 7s. Time estimates for 10 more iterations: 8m 44s, 100 more iterations: 1h 27m 22s, 500 more iterations: 7h 16m 51s. +[2026-03-26 01:57:29,751][__main__][INFO] - Starting iteration 269. +[2026-03-26 01:57:30,150][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 26 and human policies 1. +[2026-03-26 01:57:30,151][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:57:47,748][__main__][INFO] - Number of regex retries in iteration 269: 0 +[2026-03-26 01:57:47,749][__main__][INFO] - agents played in iteration 269 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:57:48,534][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:57:48,558][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:57:48,581][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:57:48,604][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:57:48,605][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:57:48,605][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:57:49,297][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:57:49,754][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:57:50,266][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:57:50,767][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:57:51,272][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:57:51,774][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:57:52,275][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:57:52,773][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:57:53,275][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:57:53,778][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:57:54,282][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:57:54,785][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:57:55,285][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:57:55,783][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:57:56,282][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:57:56,788][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:57:57,289][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:57:57,790][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:57:58,290][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:57:58,789][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:57:59,287][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:57:59,788][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:58:00,289][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:58:00,791][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:58:01,293][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:58:01,793][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:58:02,293][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:58:02,795][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:58:03,295][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:58:03,806][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:58:04,308][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:58:04,811][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:58:05,311][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:58:05,811][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:58:06,309][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:58:06,807][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:58:07,320][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:58:07,818][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:58:08,319][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:58:08,817][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:58:09,315][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:58:09,816][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:58:10,318][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:58:10,846][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:58:11,346][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:58:11,845][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:58:12,349][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:58:12,850][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:58:13,348][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:58:13,849][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:58:14,348][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:58:14,850][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:58:15,350][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:58:15,849][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:58:16,348][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:58:16,848][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:58:17,352][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:58:17,853][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:58:18,352][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:58:18,853][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:58:19,349][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:58:19,849][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:58:20,352][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:58:20,848][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:58:21,346][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10381 tokens. +[2026-03-26 01:58:22,053][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.26%, ΔTime: 00:00:32 +[2026-03-26 01:58:22,795][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:58:22,797][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:58:22,799][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:58:23,537][__main__][INFO] - Iteration 270 took 53s (32.96% Gen, 65.65% Train). Generation: 17s, Training: 35s. Estimated remaining time: 40h 31m 46s. Estimated total time: 44h 29m 21s. Time estimates for 10 more iterations: 8m 53s, 100 more iterations: 1h 28m 58s, 500 more iterations: 7h 24m 53s. +[2026-03-26 01:58:23,539][__main__][INFO] - Starting iteration 270. +[2026-03-26 01:58:23,938][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 26 and human policies 1. +[2026-03-26 01:58:23,939][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:58:28,538][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 01:58:41,618][__main__][INFO] - Number of regex retries in iteration 270: 1 +[2026-03-26 01:58:41,619][__main__][INFO] - agents played in iteration 270 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:58:42,401][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:58:42,425][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:58:42,448][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:58:42,471][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:58:42,472][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:58:42,472][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:58:43,155][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:58:43,612][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:58:44,118][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:58:44,626][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:58:45,126][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:58:45,626][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:58:46,128][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:58:46,629][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:58:47,131][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:58:47,635][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:58:48,140][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:58:48,639][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:58:49,137][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:58:49,635][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:58:50,138][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:58:50,636][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:58:51,137][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:58:51,634][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:58:52,132][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:58:52,631][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:58:53,130][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:58:53,628][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:58:54,126][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:58:54,624][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:58:55,122][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:58:55,621][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:58:56,118][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:58:56,618][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:58:57,115][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:58:57,618][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:58:58,116][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:58:58,617][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:58:59,115][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:58:59,615][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:59:00,112][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:59:00,610][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:59:01,110][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:59:01,607][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:59:02,107][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:59:02,603][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:59:03,100][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:59:03,596][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:59:04,098][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:59:04,594][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:59:05,091][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:59:05,589][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:59:06,084][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 01:59:06,580][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 01:59:07,075][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 01:59:07,570][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 01:59:08,067][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 01:59:08,566][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 01:59:09,062][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 01:59:09,559][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 01:59:10,058][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 01:59:10,555][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 01:59:11,052][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 01:59:11,552][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 01:59:12,050][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 01:59:12,547][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 01:59:13,046][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 01:59:13,548][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 01:59:14,053][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 01:59:14,551][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 01:59:15,050][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10205 tokens. +[2026-03-26 01:59:15,766][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.18%, ΔTime: 00:00:32 +[2026-03-26 01:59:16,515][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 01:59:16,518][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 01:59:16,519][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 01:59:17,805][__main__][INFO] - Iteration 271 took 53s (32.82% Gen, 64.79% Train). Generation: 17s, Training: 34s. Estimated remaining time: 40h 54m 52s. Estimated total time: 44h 53m 22s. Time estimates for 10 more iterations: 8m 58s, 100 more iterations: 1h 29m 46s, 500 more iterations: 7h 28m 53s. +[2026-03-26 01:59:17,807][__main__][INFO] - Starting iteration 271. +[2026-03-26 01:59:18,206][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 27 and human policies 1. +[2026-03-26 01:59:18,206][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 01:59:35,199][__main__][INFO] - Number of regex retries in iteration 271: 0 +[2026-03-26 01:59:35,200][__main__][INFO] - agents played in iteration 271 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 01:59:35,982][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:59:36,006][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:59:36,030][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:59:36,053][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 01:59:36,053][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 01:59:36,054][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 01:59:36,750][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 01:59:37,204][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 01:59:37,715][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 01:59:38,212][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 01:59:38,718][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 01:59:39,224][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 01:59:39,735][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 01:59:40,235][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 01:59:40,733][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 01:59:41,230][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 01:59:41,728][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 01:59:42,226][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 01:59:42,723][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 01:59:43,220][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 01:59:43,720][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 01:59:44,218][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 01:59:44,718][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 01:59:45,215][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 01:59:45,712][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 01:59:46,213][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 01:59:46,711][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 01:59:47,211][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 01:59:47,717][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 01:59:48,217][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 01:59:48,718][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 01:59:49,216][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 01:59:49,716][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 01:59:50,215][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 01:59:50,717][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 01:59:51,218][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 01:59:51,723][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 01:59:52,224][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 01:59:52,721][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 01:59:53,220][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 01:59:53,717][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 01:59:54,216][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 01:59:54,714][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 01:59:55,211][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 01:59:55,712][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 01:59:56,209][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 01:59:56,707][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 01:59:57,203][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 01:59:57,703][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 01:59:58,202][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 01:59:58,702][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 01:59:59,198][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 01:59:59,695][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:00:00,195][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:00:00,694][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:00:01,193][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:00:01,692][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:00:02,193][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:00:02,691][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:00:03,188][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:00:03,690][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:00:04,190][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:00:04,690][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:00:05,189][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:00:05,690][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:00:06,188][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:00:06,685][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:00:07,187][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:00:07,688][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:00:08,188][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:00:08,685][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10377 tokens. +[2026-03-26 02:00:09,405][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.15%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.20%, ΔTime: 00:00:32 +[2026-03-26 02:00:10,144][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:00:10,147][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:00:10,158][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:00:10,872][__main__][INFO] - Iteration 272 took 52s (32.27% Gen, 66.38% Train). Generation: 16s, Training: 34s. Estimated remaining time: 39h 53m 58s. Estimated total time: 43h 53m 20s. Time estimates for 10 more iterations: 8m 46s, 100 more iterations: 1h 27m 46s, 500 more iterations: 7h 18m 53s. +[2026-03-26 02:00:10,875][__main__][INFO] - Starting iteration 272. +[2026-03-26 02:00:11,277][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 27 and human policies 1. +[2026-03-26 02:00:11,278][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:00:30,006][__main__][INFO] - Number of regex retries in iteration 272: 0 +[2026-03-26 02:00:30,007][__main__][INFO] - agents played in iteration 272 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:00:30,798][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:00:30,821][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:00:30,844][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:00:30,867][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:00:30,868][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:00:30,868][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:00:31,551][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:00:32,006][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:00:32,517][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:00:33,018][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:00:33,522][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:00:34,019][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:00:34,518][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:00:35,022][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:00:35,522][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:00:36,022][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:00:36,523][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:00:37,022][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:00:37,519][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:00:38,022][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:00:38,526][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:00:39,025][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:00:39,527][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:00:40,029][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:00:40,549][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:00:41,054][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:00:41,556][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:00:42,056][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:00:42,565][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:00:43,068][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:00:43,569][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:00:44,067][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:00:44,566][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:00:45,063][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:00:45,558][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:00:46,055][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:00:46,555][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:00:47,051][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:00:47,546][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:00:48,044][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:00:48,542][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:00:49,039][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:00:49,537][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:00:50,035][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:00:50,534][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:00:51,037][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:00:51,534][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:00:52,032][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:00:52,533][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:00:53,030][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:00:53,529][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:00:54,030][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:00:54,527][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:00:55,027][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:00:55,527][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:00:56,027][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:00:56,531][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:00:57,033][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:00:57,529][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:00:58,027][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:00:58,524][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:00:59,021][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:00:59,519][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:01:00,016][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:01:00,512][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:01:01,009][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:01:01,505][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:01:02,005][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:01:02,502][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:01:03,003][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:01:03,503][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10230 tokens. +[2026-03-26 02:01:04,221][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.33%, ΔTime: 00:00:32 +[2026-03-26 02:01:05,014][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:01:05,016][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:01:05,018][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:01:05,840][__main__][INFO] - Iteration 273 took 54s (34.33% Gen, 64.16% Train). Generation: 18s, Training: 35s. Estimated remaining time: 41h 27m 53s. Estimated total time: 45h 28m 11s. Time estimates for 10 more iterations: 9m 5s, 100 more iterations: 1h 30m 56s, 500 more iterations: 7h 34m 41s. +[2026-03-26 02:01:05,843][__main__][INFO] - Starting iteration 273. +[2026-03-26 02:01:06,241][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 27 and human policies 1. +[2026-03-26 02:01:06,242][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:01:11,139][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:01:24,518][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:01:25,509][__main__][INFO] - Number of regex retries in iteration 273: 2 +[2026-03-26 02:01:25,510][__main__][INFO] - agents played in iteration 273 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:01:26,296][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:01:26,320][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:01:26,343][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:01:26,366][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:01:26,367][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:01:26,367][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:01:27,050][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:01:27,509][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:01:28,020][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:01:28,523][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:01:29,027][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:01:29,529][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:01:30,029][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:01:30,536][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:01:31,037][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:01:31,537][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:01:32,037][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:01:32,538][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:01:33,038][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:01:33,537][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:01:34,036][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:01:34,533][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:01:35,033][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:01:35,532][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:01:36,030][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:01:36,526][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:01:37,026][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:01:37,532][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:01:38,032][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:01:38,534][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:01:39,036][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:01:39,535][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:01:40,031][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:01:40,529][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:01:41,034][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:01:41,532][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:01:42,032][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:01:42,530][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:01:43,028][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:01:43,527][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:01:44,028][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:01:44,528][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:01:45,029][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:01:45,532][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:01:46,031][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:01:46,530][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:01:47,031][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:01:47,535][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:01:48,035][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:01:48,532][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:01:49,030][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:01:49,531][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:01:50,029][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:01:50,530][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:01:51,028][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:01:51,529][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:01:52,029][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:01:52,533][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:01:53,034][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:01:53,531][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:01:54,033][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:01:54,530][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:01:55,028][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:01:55,528][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:01:56,028][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:01:56,528][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:01:57,028][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:01:57,524][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:01:58,024][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:01:58,531][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:01:59,028][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10414 tokens. +[2026-03-26 02:01:59,755][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.60%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 02:02:00,499][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:02:00,501][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:02:00,503][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:02:01,220][__main__][INFO] - Iteration 274 took 54s (35.05% Gen, 63.65% Train). Generation: 19s, Training: 34s. Estimated remaining time: 41h 47m 43s. Estimated total time: 45h 48m 56s. Time estimates for 10 more iterations: 9m 9s, 100 more iterations: 1h 31m 37s, 500 more iterations: 7h 38m 9s. +[2026-03-26 02:02:01,222][__main__][INFO] - Starting iteration 274. +[2026-03-26 02:02:01,624][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 27 and human policies 1. +[2026-03-26 02:02:01,624][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:02:07,827][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:02:18,592][__main__][INFO] - Number of regex retries in iteration 274: 1 +[2026-03-26 02:02:18,593][__main__][INFO] - agents played in iteration 274 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:02:19,362][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:02:19,385][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:02:19,408][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:02:19,431][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:02:19,432][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:02:19,432][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:02:20,117][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:02:20,572][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:02:21,077][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:02:21,580][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:02:22,084][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:02:22,582][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:02:23,081][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:02:23,586][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:02:24,084][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:02:24,586][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:02:25,086][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:02:25,590][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:02:26,090][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:02:26,592][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:02:27,092][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:02:27,593][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:02:28,094][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:02:28,592][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:02:29,089][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:02:29,589][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:02:30,093][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:02:30,592][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:02:31,095][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:02:31,592][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:02:32,093][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:02:32,593][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:02:33,094][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:02:33,611][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:02:34,115][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:02:34,615][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:02:35,116][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:02:35,616][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:02:36,118][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:02:36,619][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:02:37,120][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:02:37,617][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:02:38,115][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:02:38,615][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:02:39,110][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:02:39,608][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:02:40,106][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:02:40,603][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:02:41,099][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:02:41,597][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:02:42,095][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:02:42,592][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:02:43,090][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:02:43,588][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:02:44,084][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:02:44,584][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:02:45,081][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:02:45,580][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:02:46,080][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:02:46,578][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:02:47,082][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:02:47,580][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:02:48,083][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:02:48,581][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:02:49,078][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:02:49,575][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:02:50,075][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:02:50,574][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:02:51,074][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:02:51,571][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:02:52,067][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10333 tokens. +[2026-03-26 02:02:52,782][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.61%, Block Peak % of device VRAM: 62.23%, ΔTime: 00:00:32 +[2026-03-26 02:02:53,538][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:02:53,541][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:02:53,542][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:02:54,265][__main__][INFO] - Iteration 275 took 52s (32.23% Gen, 66.39% Train). Generation: 16s, Training: 34s. Estimated remaining time: 39h 49m 59s. Estimated total time: 43h 52m 4s. Time estimates for 10 more iterations: 8m 46s, 100 more iterations: 1h 27m 44s, 500 more iterations: 7h 18m 40s. +[2026-03-26 02:02:54,267][__main__][INFO] - Starting iteration 275. +[2026-03-26 02:02:54,667][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 27 and human policies 1. +[2026-03-26 02:02:54,668][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:03:06,879][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 20 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:03:08,402][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:03:12,205][__main__][INFO] - Number of regex retries in iteration 275: 2 +[2026-03-26 02:03:12,206][__main__][INFO] - agents played in iteration 275 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:03:13,005][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:03:13,028][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:03:13,052][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:03:13,075][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:03:13,076][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:03:13,077][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:03:13,772][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:03:14,233][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:03:14,744][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:03:15,249][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:03:15,748][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:03:16,246][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:03:16,751][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:03:17,255][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:03:17,757][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:03:18,259][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:03:18,757][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:03:19,258][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:03:19,762][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:03:20,272][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:03:20,775][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:03:21,276][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:03:21,778][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:03:22,279][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:03:22,776][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:03:23,276][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:03:23,773][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:03:24,279][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:03:24,778][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:03:25,280][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:03:25,781][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:03:26,283][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:03:26,780][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:03:27,278][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:03:27,776][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:03:28,279][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:03:28,786][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:03:29,286][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:03:29,788][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:03:30,286][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:03:30,785][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:03:31,283][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:03:31,781][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:03:32,279][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:03:32,776][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:03:33,276][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:03:33,774][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:03:34,278][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:03:34,776][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:03:35,274][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:03:35,772][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:03:36,269][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:03:36,765][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:03:37,261][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:03:37,756][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:03:38,257][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:03:38,756][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:03:39,253][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:03:39,752][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:03:40,252][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:03:40,749][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:03:41,250][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:03:41,747][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:03:42,247][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:03:42,745][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:03:43,241][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:03:43,739][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:03:44,239][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:03:44,738][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:03:45,240][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:03:45,742][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10296 tokens. +[2026-03-26 02:03:46,475][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.23%, Current % of VRAM taken: 60.66%, Block Peak % of device VRAM: 62.44%, ΔTime: 00:00:32 +[2026-03-26 02:03:47,217][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:03:47,219][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:03:47,221][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:03:48,047][__main__][INFO] - Iteration 276 took 53s (32.86% Gen, 65.59% Train). Generation: 17s, Training: 35s. Estimated remaining time: 40h 26m 2s. Estimated total time: 44h 29m 2s. Time estimates for 10 more iterations: 8m 53s, 100 more iterations: 1h 28m 58s, 500 more iterations: 7h 24m 50s. +[2026-03-26 02:03:48,050][__main__][INFO] - Starting iteration 276. +[2026-03-26 02:03:48,451][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 27 and human policies 1. +[2026-03-26 02:03:48,452][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:04:04,459][__main__][INFO] - Number of regex retries in iteration 276: 0 +[2026-03-26 02:04:04,460][__main__][INFO] - agents played in iteration 276 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:04:05,245][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:04:05,269][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:04:05,293][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:04:05,316][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:04:05,317][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:04:05,317][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:04:06,006][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:04:06,463][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:04:06,969][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:04:07,471][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:04:07,970][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:04:08,469][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:04:08,969][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:04:09,473][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:04:09,978][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:04:10,479][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:04:10,983][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:04:11,481][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:04:11,981][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:04:12,482][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:04:12,980][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:04:13,478][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:04:13,978][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:04:14,475][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:04:14,974][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:04:15,471][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:04:15,970][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:04:16,466][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:04:16,962][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:04:17,459][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:04:17,959][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:04:18,460][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:04:18,959][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:04:19,457][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:04:19,956][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:04:20,455][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:04:20,954][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:04:21,454][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:04:21,959][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:04:22,456][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:04:22,957][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:04:23,455][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:04:23,957][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:04:24,458][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:04:24,962][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:04:25,460][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:04:25,959][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:04:26,457][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:04:26,958][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:04:27,459][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:04:27,957][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:04:28,454][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:04:28,954][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:04:29,452][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:04:29,951][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:04:30,449][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:04:30,948][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:04:31,448][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:04:31,945][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:04:32,443][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:04:32,940][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:04:33,438][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:04:33,935][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:04:34,437][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:04:34,938][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:04:35,438][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:04:35,938][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:04:36,438][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:04:36,937][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:04:37,435][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:04:37,936][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10209 tokens. +[2026-03-26 02:04:38,660][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.19%, Current % of VRAM taken: 60.62%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 02:04:39,403][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:04:39,407][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:04:39,408][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:04:40,164][__main__][INFO] - Iteration 277 took 51s (30.95% Gen, 67.58% Train). Generation: 16s, Training: 34s. Estimated remaining time: 39h 1m 47s. Estimated total time: 43h 5m 38s. Time estimates for 10 more iterations: 8m 37s, 100 more iterations: 1h 26m 11s, 500 more iterations: 7h 10m 56s. +[2026-03-26 02:04:40,166][__main__][INFO] - Starting iteration 277. +[2026-03-26 02:04:40,568][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 27 and human policies 1. +[2026-03-26 02:04:40,568][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:04:45,181][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:04:45,343][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:04:58,987][__main__][INFO] - Number of regex retries in iteration 277: 2 +[2026-03-26 02:04:58,988][__main__][INFO] - agents played in iteration 277 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:04:59,773][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:04:59,797][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:04:59,820][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:04:59,844][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:04:59,844][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:04:59,845][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:05:00,538][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:05:00,998][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:05:01,508][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:05:02,013][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:05:02,515][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:05:03,015][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:05:03,519][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:05:04,018][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:05:04,520][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:05:05,022][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:05:05,519][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:05:06,021][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:05:06,522][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:05:07,020][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:05:07,522][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:05:08,023][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:05:08,525][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:05:09,020][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:05:09,517][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:05:10,013][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:05:10,511][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:05:11,006][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:05:11,505][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:05:12,001][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:05:12,497][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:05:12,995][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:05:13,492][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:05:13,989][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:05:14,490][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:05:14,991][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:05:15,491][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:05:15,988][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:05:16,492][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:05:16,990][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:05:17,489][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:05:17,987][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:05:18,485][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:05:18,987][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:05:19,485][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:05:19,983][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:05:20,482][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:05:20,979][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:05:21,480][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:05:21,979][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:05:22,479][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:05:22,979][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:05:23,480][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:05:23,980][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:05:24,482][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:05:24,979][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:05:25,479][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:05:25,975][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:05:26,477][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:05:26,975][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:05:27,472][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:05:27,969][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:05:28,467][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:05:28,966][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:05:29,466][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:05:29,963][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:05:30,465][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:05:30,962][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:05:31,461][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:05:31,960][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:05:32,459][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10359 tokens. +[2026-03-26 02:05:33,174][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.15%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.21%, ΔTime: 00:00:32 +[2026-03-26 02:05:33,951][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:05:33,953][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:05:33,955][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:05:34,634][__main__][INFO] - Iteration 278 took 54s (34.07% Gen, 64.67% Train). Generation: 18s, Training: 34s. Estimated remaining time: 40h 58m 33s. Estimated total time: 45h 3m 19s. Time estimates for 10 more iterations: 9m 0s, 100 more iterations: 1h 30m 6s, 500 more iterations: 7h 30m 33s. +[2026-03-26 02:05:34,636][__main__][INFO] - Starting iteration 278. +[2026-03-26 02:05:35,036][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 27 and human policies 1. +[2026-03-26 02:05:35,037][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:05:52,434][__main__][INFO] - Number of regex retries in iteration 278: 0 +[2026-03-26 02:05:52,434][__main__][INFO] - agents played in iteration 278 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:05:53,229][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:05:53,253][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:05:53,276][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:05:53,300][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:05:53,301][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:05:53,301][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:05:54,005][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:05:54,463][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:05:54,970][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:05:55,471][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:05:55,979][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:05:56,480][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:05:56,981][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:05:57,480][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:05:57,981][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:05:58,479][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:05:58,981][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:05:59,485][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:05:59,987][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:06:00,486][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:06:00,983][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:06:01,483][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:06:01,986][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:06:02,483][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:06:02,985][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:06:03,490][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:06:03,992][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:06:04,492][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:06:04,991][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:06:05,490][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:06:05,988][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:06:06,487][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:06:06,992][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:06:07,495][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:06:07,995][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:06:08,495][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:06:08,995][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:06:09,496][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:06:09,995][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:06:10,495][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:06:10,996][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:06:11,493][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:06:11,991][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:06:12,491][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:06:12,993][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:06:13,495][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:06:13,992][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:06:14,492][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:06:15,000][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:06:15,500][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:06:15,998][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:06:16,495][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:06:16,995][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:06:17,500][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:06:17,997][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:06:18,494][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:06:18,994][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:06:19,491][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:06:19,998][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:06:20,497][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:06:20,996][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:06:21,496][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:06:21,994][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:06:22,493][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:06:22,995][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:06:23,493][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:06:23,994][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:06:24,491][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:06:24,991][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:06:25,488][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:06:25,988][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10412 tokens. +[2026-03-26 02:06:26,713][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.16%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.22%, ΔTime: 00:00:32 +[2026-03-26 02:06:27,459][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:06:27,462][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:06:27,463][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:06:28,190][__main__][INFO] - Iteration 279 took 53s (32.73% Gen, 65.90% Train). Generation: 17s, Training: 35s. Estimated remaining time: 40h 12m 4s. Estimated total time: 44h 17m 44s. Time estimates for 10 more iterations: 8m 51s, 100 more iterations: 1h 28m 35s, 500 more iterations: 7h 22m 57s. +[2026-03-26 02:06:28,192][__main__][INFO] - Starting iteration 279. +[2026-03-26 02:06:28,593][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 27 and human policies 1. +[2026-03-26 02:06:28,593][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:06:45,364][__main__][INFO] - Number of regex retries in iteration 279: 0 +[2026-03-26 02:06:45,365][__main__][INFO] - agents played in iteration 279 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:06:46,149][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:06:46,172][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:06:46,196][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:06:46,219][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:06:46,219][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:06:46,220][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:06:46,919][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:06:47,373][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:06:47,882][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:06:48,392][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:06:48,896][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:06:49,395][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:06:49,896][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:06:50,396][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:06:50,897][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:06:51,395][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:06:51,895][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:06:52,393][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:06:52,892][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:06:53,390][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:06:53,889][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:06:54,387][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:06:54,889][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:06:55,389][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:06:55,893][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:06:56,400][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:06:56,902][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:06:57,406][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:06:57,904][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:06:58,407][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:06:58,907][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:06:59,405][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:06:59,902][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:07:00,400][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:07:00,898][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:07:01,399][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:07:01,895][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:07:02,393][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:07:02,894][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:07:03,392][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:07:03,894][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:07:04,392][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:07:04,891][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:07:05,391][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:07:05,893][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:07:06,393][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:07:06,897][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:07:07,395][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:07:07,893][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:07:08,391][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:07:08,888][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:07:09,387][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:07:09,885][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:07:10,383][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:07:10,880][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:07:11,377][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:07:11,875][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:07:12,379][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:07:12,878][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:07:13,382][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:07:13,883][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:07:14,386][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:07:14,887][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:07:15,385][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:07:15,887][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:07:16,383][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:07:16,886][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:07:17,383][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:07:17,883][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:07:18,387][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:07:18,888][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10198 tokens. +[2026-03-26 02:07:19,617][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.61%, Block Peak % of device VRAM: 62.21%, ΔTime: 00:00:32 +[2026-03-26 02:07:20,389][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:07:20,391][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:07:20,393][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:07:21,227][__main__][INFO] - Iteration 280 took 52s (31.86% Gen, 66.55% Train). Generation: 16s, Training: 35s. Estimated remaining time: 39h 45m 12s. Estimated total time: 43h 51m 44s. Time estimates for 10 more iterations: 8m 46s, 100 more iterations: 1h 27m 43s, 500 more iterations: 7h 18m 37s. +[2026-03-26 02:07:21,231][__main__][INFO] - Starting iteration 280. +[2026-03-26 02:07:21,632][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 27 and human policies 1. +[2026-03-26 02:07:21,633][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:07:30,936][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 20 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:07:39,700][__main__][INFO] - Number of regex retries in iteration 280: 1 +[2026-03-26 02:07:39,701][__main__][INFO] - agents played in iteration 280 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:07:40,488][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:07:40,511][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:07:40,534][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:07:40,557][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:07:40,558][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:07:40,558][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:07:41,232][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:07:41,688][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:07:42,194][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:07:42,694][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:07:43,198][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:07:43,697][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:07:44,198][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:07:44,704][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:07:45,205][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:07:45,705][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:07:46,207][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:07:46,708][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:07:47,207][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:07:47,714][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:07:48,216][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:07:48,714][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:07:49,220][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:07:49,719][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:07:50,218][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:07:50,717][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:07:51,216][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:07:51,715][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:07:52,212][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:07:52,710][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:07:53,209][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:07:53,711][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:07:54,208][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:07:54,707][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:07:55,204][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:07:55,701][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:07:56,205][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:07:56,703][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:07:57,200][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:07:57,700][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:07:58,198][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:07:58,699][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:07:59,196][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:07:59,696][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:08:00,197][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:08:00,704][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:08:01,207][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:08:01,703][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:08:02,201][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:08:02,699][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:08:03,195][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:08:03,693][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:08:04,192][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:08:04,692][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:08:05,192][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:08:05,690][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:08:06,190][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:08:06,689][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:08:07,193][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:08:07,691][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:08:08,190][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:08:08,693][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:08:09,190][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:08:09,689][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:08:10,185][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:08:10,683][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:08:11,180][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:08:11,676][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:08:12,174][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:08:12,673][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:08:13,174][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10230 tokens. +[2026-03-26 02:08:13,897][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.17%, ΔTime: 00:00:32 +[2026-03-26 02:08:14,728][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:08:15,027][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:08:15,029][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:08:16,673][__main__][INFO] - Iteration 281 took 55s (32.83% Gen, 64.18% Train). Generation: 18s, Training: 35s. Estimated remaining time: 41h 44m 35s. Estimated total time: 45h 52m 3s. Time estimates for 10 more iterations: 9m 10s, 100 more iterations: 1h 31m 44s, 500 more iterations: 7h 38m 40s. +[2026-03-26 02:08:16,676][__main__][INFO] - Starting iteration 281. +[2026-03-26 02:08:17,077][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 28 and human policies 1. +[2026-03-26 02:08:17,078][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:08:33,904][__main__][INFO] - Number of regex retries in iteration 281: 0 +[2026-03-26 02:08:33,905][__main__][INFO] - agents played in iteration 281 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:08:34,675][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:08:34,699][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:08:34,722][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:08:34,745][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:08:34,746][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:08:34,746][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:08:35,419][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:08:35,872][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:08:36,376][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:08:36,874][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:08:37,372][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:08:37,873][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:08:38,375][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:08:38,881][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:08:39,384][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:08:39,887][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:08:40,388][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:08:40,887][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:08:41,389][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:08:41,889][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:08:42,387][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:08:42,886][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:08:43,384][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:08:43,882][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:08:44,380][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:08:44,877][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:08:45,376][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:08:45,876][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:08:46,374][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:08:46,871][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:08:47,367][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:08:47,888][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:08:48,395][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:08:48,899][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:08:49,402][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:08:49,903][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:08:50,405][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:08:50,908][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:08:51,409][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:08:51,911][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:08:52,419][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:08:52,921][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:08:53,422][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:08:53,924][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:08:54,426][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:08:54,929][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:08:55,430][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:08:55,930][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:08:56,428][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:08:56,927][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:08:57,426][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:08:57,924][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:08:58,421][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:08:58,918][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:08:59,415][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:08:59,917][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:09:00,416][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:09:00,912][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:09:01,415][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:09:01,916][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:09:02,418][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:09:02,918][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:09:03,419][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:09:03,919][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:09:04,416][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:09:04,917][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:09:05,418][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:09:05,915][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:09:06,417][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:09:06,916][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:09:07,416][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10478 tokens. +[2026-03-26 02:09:08,120][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.25%, Current % of VRAM taken: 60.67%, Block Peak % of device VRAM: 62.40%, ΔTime: 00:00:32 +[2026-03-26 02:09:08,862][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:09:08,865][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:09:08,866][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:09:09,686][__main__][INFO] - Iteration 282 took 52s (31.98% Gen, 66.45% Train). Generation: 16s, Training: 34s. Estimated remaining time: 39h 42m 7s. Estimated total time: 43h 50m 28s. Time estimates for 10 more iterations: 8m 46s, 100 more iterations: 1h 27m 40s, 500 more iterations: 7h 18m 24s. +[2026-03-26 02:09:09,688][__main__][INFO] - Starting iteration 282. +[2026-03-26 02:09:10,087][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 28 and human policies 1. +[2026-03-26 02:09:10,088][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:09:27,324][__main__][INFO] - Number of regex retries in iteration 282: 0 +[2026-03-26 02:09:27,324][__main__][INFO] - agents played in iteration 282 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:09:28,104][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:09:28,127][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:09:28,150][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:09:28,173][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:09:28,174][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:09:28,174][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:09:28,844][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:09:29,299][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:09:29,806][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:09:30,302][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:09:30,801][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:09:31,300][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:09:31,803][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:09:32,304][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:09:32,801][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:09:33,305][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:09:33,803][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:09:34,302][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:09:34,800][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:09:35,301][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:09:35,802][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:09:36,300][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:09:36,798][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:09:37,296][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:09:37,796][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:09:38,294][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:09:38,795][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:09:39,293][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:09:39,793][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:09:40,325][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:09:40,825][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:09:41,323][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:09:41,820][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:09:42,318][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:09:42,820][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:09:43,318][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:09:43,818][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:09:44,315][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:09:44,814][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:09:45,313][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:09:45,813][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:09:46,312][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:09:46,812][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:09:47,310][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:09:47,810][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:09:48,309][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:09:48,805][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:09:49,301][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:09:49,800][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:09:50,304][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:09:50,800][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:09:51,297][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:09:51,798][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:09:52,295][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:09:52,795][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:09:53,295][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:09:53,799][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:09:54,298][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:09:54,798][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:09:55,297][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:09:55,797][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:09:56,297][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:09:56,796][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:09:57,293][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:09:57,793][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:09:58,290][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:09:58,792][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:09:59,294][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:09:59,794][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:10:00,295][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:10:00,796][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10442 tokens. +[2026-03-26 02:10:01,498][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.25%, Current % of VRAM taken: 60.68%, Block Peak % of device VRAM: 62.26%, ΔTime: 00:00:32 +[2026-03-26 02:10:02,247][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:10:02,250][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:10:02,251][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:10:02,974][__main__][INFO] - Iteration 283 took 52s (32.59% Gen, 66.04% Train). Generation: 17s, Training: 34s. Estimated remaining time: 39h 55m 6s. Estimated total time: 44h 4m 20s. Time estimates for 10 more iterations: 8m 48s, 100 more iterations: 1h 28m 8s, 500 more iterations: 7h 20m 43s. +[2026-03-26 02:10:02,976][__main__][INFO] - Starting iteration 283. +[2026-03-26 02:10:03,378][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 28 and human policies 1. +[2026-03-26 02:10:03,378][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:10:20,101][__main__][INFO] - Number of regex retries in iteration 283: 0 +[2026-03-26 02:10:20,102][__main__][INFO] - agents played in iteration 283 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:10:20,877][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:10:20,900][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:10:20,924][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:10:20,947][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:10:20,947][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:10:20,948][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:10:21,638][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:10:22,093][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:10:22,598][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:10:23,099][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:10:23,599][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:10:24,098][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:10:24,599][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:10:25,103][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:10:25,604][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:10:26,106][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:10:26,607][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:10:27,107][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:10:27,607][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:10:28,107][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:10:28,604][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:10:29,104][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:10:29,605][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:10:30,101][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:10:30,597][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:10:31,095][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:10:31,593][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:10:32,090][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:10:32,591][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:10:33,090][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:10:33,587][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:10:34,089][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:10:34,593][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:10:35,094][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:10:35,594][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:10:36,094][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:10:36,593][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:10:37,094][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:10:37,594][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:10:38,094][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:10:38,595][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:10:39,095][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:10:39,594][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:10:40,095][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:10:40,595][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:10:41,094][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:10:41,594][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:10:42,091][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:10:42,591][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:10:43,089][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:10:43,589][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:10:44,089][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:10:44,589][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:10:45,087][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:10:45,584][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:10:46,080][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:10:46,578][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:10:47,075][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:10:47,572][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:10:48,070][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:10:48,568][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:10:49,065][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:10:49,566][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:10:50,063][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:10:50,564][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:10:51,061][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:10:51,561][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:10:52,057][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:10:52,553][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:10:53,051][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:10:53,549][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10319 tokens. +[2026-03-26 02:10:54,270][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.32%, ΔTime: 00:00:32 +[2026-03-26 02:10:55,015][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:10:55,017][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:10:55,019][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:10:55,854][__main__][INFO] - Iteration 284 took 52s (31.87% Gen, 66.54% Train). Generation: 16s, Training: 34s. Estimated remaining time: 39h 33m 47s. Estimated total time: 43h 43m 54s. Time estimates for 10 more iterations: 8m 44s, 100 more iterations: 1h 27m 27s, 500 more iterations: 7h 17m 19s. +[2026-03-26 02:10:55,856][__main__][INFO] - Starting iteration 284. +[2026-03-26 02:10:56,258][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 28 and human policies 1. +[2026-03-26 02:10:56,258][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:11:13,624][__main__][INFO] - Number of regex retries in iteration 284: 0 +[2026-03-26 02:11:13,625][__main__][INFO] - agents played in iteration 284 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:11:14,403][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:11:14,426][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:11:14,449][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:11:14,472][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:11:14,473][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:11:14,473][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:11:15,149][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:11:15,602][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:11:16,107][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:11:16,607][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:11:17,107][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:11:17,606][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:11:18,104][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:11:18,604][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:11:19,101][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:11:19,599][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:11:20,103][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:11:20,603][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:11:21,103][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:11:21,607][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:11:22,108][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:11:22,608][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:11:23,109][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:11:23,607][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:11:24,107][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:11:24,606][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:11:25,109][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:11:25,611][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:11:26,109][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:11:26,611][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:11:27,112][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:11:27,613][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:11:28,114][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:11:28,613][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:11:29,114][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:11:29,613][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:11:30,113][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:11:30,614][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:11:31,111][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:11:31,608][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:11:32,105][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:11:32,606][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:11:33,102][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:11:33,599][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:11:34,097][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:11:34,595][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:11:35,091][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:11:35,589][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:11:36,090][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:11:36,588][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:11:37,088][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:11:37,587][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:11:38,084][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:11:38,583][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:11:39,080][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:11:39,577][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:11:40,074][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:11:40,571][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:11:41,068][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:11:41,566][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:11:42,063][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:11:42,560][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:11:43,059][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:11:43,556][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:11:44,053][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:11:44,550][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:11:45,048][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:11:45,545][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:11:46,043][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:11:46,539][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:11:47,036][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10184 tokens. +[2026-03-26 02:11:47,771][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.04%, Current % of VRAM taken: 60.47%, Block Peak % of device VRAM: 62.34%, ΔTime: 00:00:32 +[2026-03-26 02:11:48,528][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:11:48,530][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:11:48,532][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:11:49,349][__main__][INFO] - Iteration 285 took 53s (32.71% Gen, 65.75% Train). Generation: 17s, Training: 34s. Estimated remaining time: 40h 3m 34s. Estimated total time: 44h 14m 35s. Time estimates for 10 more iterations: 8m 50s, 100 more iterations: 1h 28m 29s, 500 more iterations: 7h 22m 25s. +[2026-03-26 02:11:49,351][__main__][INFO] - Starting iteration 285. +[2026-03-26 02:11:49,752][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 28 and human policies 1. +[2026-03-26 02:11:49,753][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:12:08,021][__main__][INFO] - Number of regex retries in iteration 285: 0 +[2026-03-26 02:12:08,022][__main__][INFO] - agents played in iteration 285 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:12:08,796][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:12:08,820][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:12:08,843][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:12:08,866][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:12:08,866][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:12:08,867][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:12:09,550][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:12:10,007][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:12:10,521][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:12:11,023][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:12:11,521][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:12:12,019][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:12:12,518][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:12:13,016][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:12:13,518][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:12:14,018][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:12:14,515][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:12:15,014][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:12:15,515][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:12:16,014][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:12:16,515][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:12:17,017][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:12:17,523][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:12:18,030][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:12:18,533][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:12:19,035][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:12:19,533][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:12:20,036][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:12:20,539][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:12:21,044][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:12:21,546][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:12:22,043][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:12:22,544][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:12:23,041][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:12:23,539][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:12:24,035][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:12:24,535][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:12:25,035][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:12:25,534][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:12:26,031][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:12:26,529][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:12:27,029][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:12:27,530][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:12:28,027][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:12:28,525][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:12:29,026][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:12:29,524][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:12:30,022][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:12:30,520][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:12:31,018][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:12:31,517][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:12:32,017][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:12:32,515][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:12:33,012][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:12:33,511][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:12:34,011][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:12:34,508][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:12:35,010][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:12:35,506][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:12:36,006][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:12:36,512][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:12:37,012][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:12:37,512][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:12:38,010][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:12:38,508][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:12:39,009][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:12:39,512][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:12:40,017][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:12:40,519][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:12:41,028][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:12:41,527][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10341 tokens. +[2026-03-26 02:12:42,253][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.20%, ΔTime: 00:00:32 +[2026-03-26 02:12:42,997][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:12:42,999][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:12:43,001][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:12:43,694][__main__][INFO] - Iteration 286 took 53s (33.87% Gen, 64.84% Train). Generation: 18s, Training: 34s. Estimated remaining time: 40h 45m 11s. Estimated total time: 44h 57m 6s. Time estimates for 10 more iterations: 8m 59s, 100 more iterations: 1h 29m 54s, 500 more iterations: 7h 29m 31s. +[2026-03-26 02:12:43,696][__main__][INFO] - Starting iteration 286. +[2026-03-26 02:12:44,101][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 28 and human policies 1. +[2026-03-26 02:12:44,102][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:12:48,955][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:13:01,908][__main__][INFO] - Number of regex retries in iteration 286: 1 +[2026-03-26 02:13:01,910][__main__][INFO] - agents played in iteration 286 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:13:02,686][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:13:02,711][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:13:02,736][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:13:02,760][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:13:02,761][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:13:02,761][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:13:03,453][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:13:03,908][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:13:04,412][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:13:04,915][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:13:05,412][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:13:05,915][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:13:06,413][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:13:06,915][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:13:07,415][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:13:07,921][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:13:08,448][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:13:08,949][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:13:09,451][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:13:09,954][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:13:10,454][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:13:10,955][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:13:11,456][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:13:11,960][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:13:12,461][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:13:12,962][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:13:13,461][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:13:13,964][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:13:14,465][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:13:14,968][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:13:15,469][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:13:15,974][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:13:16,472][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:13:16,972][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:13:17,472][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:13:17,986][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:13:18,490][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:13:18,994][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:13:19,495][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:13:19,995][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:13:20,495][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:13:20,997][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:13:21,496][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:13:21,995][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:13:22,493][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:13:22,995][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:13:23,494][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:13:23,996][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:13:24,497][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:13:24,998][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:13:25,505][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:13:26,004][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:13:26,506][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:13:27,008][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:13:27,542][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:13:28,040][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:13:28,542][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:13:29,038][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:13:29,539][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:13:30,036][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:13:30,537][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:13:31,034][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:13:31,536][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:13:32,040][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:13:32,544][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:13:33,046][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:13:33,546][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:13:34,043][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:13:34,546][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:13:35,044][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:13:35,544][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10453 tokens. +[2026-03-26 02:13:36,262][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.21%, Current % of VRAM taken: 60.63%, Block Peak % of device VRAM: 62.33%, ΔTime: 00:00:32 +[2026-03-26 02:13:37,008][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:13:37,010][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:13:37,012][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:13:37,721][__main__][INFO] - Iteration 287 took 53s (33.21% Gen, 65.46% Train). Generation: 17s, Training: 35s. Estimated remaining time: 40h 28m 11s. Estimated total time: 44h 41m 0s. Time estimates for 10 more iterations: 8m 56s, 100 more iterations: 1h 29m 22s, 500 more iterations: 7h 26m 50s. +[2026-03-26 02:13:37,723][__main__][INFO] - Starting iteration 287. +[2026-03-26 02:13:38,121][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 28 and human policies 1. +[2026-03-26 02:13:38,121][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:13:42,370][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:13:55,389][__main__][INFO] - Number of regex retries in iteration 287: 1 +[2026-03-26 02:13:55,390][__main__][INFO] - agents played in iteration 287 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:13:56,167][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:13:56,190][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:13:56,214][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:13:56,237][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:13:56,238][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:13:56,238][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:13:56,952][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:13:57,407][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:13:57,913][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:13:58,417][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:13:58,918][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:13:59,423][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:13:59,924][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:14:00,423][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:14:00,926][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:14:01,424][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:14:01,927][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:14:02,428][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:14:02,926][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:14:03,424][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:14:03,922][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:14:04,420][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:14:04,923][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:14:05,426][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:14:05,931][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:14:06,431][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:14:06,932][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:14:07,433][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:14:07,935][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:14:08,442][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:14:08,945][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:14:09,443][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:14:09,940][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:14:10,438][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:14:10,936][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:14:11,432][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:14:11,931][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:14:12,435][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:14:12,933][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:14:13,430][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:14:13,929][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:14:14,428][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:14:14,930][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:14:15,428][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:14:15,927][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:14:16,427][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:14:16,928][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:14:17,426][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:14:17,923][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:14:18,423][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:14:18,919][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:14:19,416][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:14:19,914][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:14:20,412][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:14:20,910][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:14:21,421][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:14:21,918][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:14:22,417][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:14:22,921][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:14:23,424][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:14:23,927][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:14:24,427][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:14:24,928][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:14:25,425][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:14:25,928][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:14:26,429][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:14:26,929][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:14:27,429][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:14:27,934][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:14:28,433][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:14:28,935][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10258 tokens. +[2026-03-26 02:14:29,682][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.20%, Current % of VRAM taken: 60.63%, Block Peak % of device VRAM: 62.37%, ΔTime: 00:00:32 +[2026-03-26 02:14:30,437][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:14:30,439][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:14:30,441][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:14:31,173][__main__][INFO] - Iteration 288 took 53s (32.55% Gen, 66.07% Train). Generation: 17s, Training: 35s. Estimated remaining time: 39h 58m 57s. Estimated total time: 44h 12m 40s. Time estimates for 10 more iterations: 8m 50s, 100 more iterations: 1h 28m 25s, 500 more iterations: 7h 22m 6s. +[2026-03-26 02:14:31,176][__main__][INFO] - Starting iteration 288. +[2026-03-26 02:14:31,576][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 28 and human policies 1. +[2026-03-26 02:14:31,577][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:14:36,274][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:14:48,931][__main__][INFO] - Number of regex retries in iteration 288: 1 +[2026-03-26 02:14:48,932][__main__][INFO] - agents played in iteration 288 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:14:49,720][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:14:49,743][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:14:49,766][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:14:49,789][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:14:49,790][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:14:49,791][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:14:50,503][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:14:50,958][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:14:51,465][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:14:51,967][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:14:52,468][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:14:52,970][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:14:53,472][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:14:53,977][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:14:54,484][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:14:54,987][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:14:55,488][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:14:55,988][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:14:56,491][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:14:56,999][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:14:57,500][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:14:58,008][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:14:58,512][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:14:59,017][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:14:59,520][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:15:00,028][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:15:00,532][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:15:01,046][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:15:01,550][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:15:02,054][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:15:02,555][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:15:03,057][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:15:03,557][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:15:04,056][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:15:04,554][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:15:05,056][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:15:05,561][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:15:06,062][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:15:06,561][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:15:07,059][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:15:07,556][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:15:08,053][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:15:08,554][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:15:09,051][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:15:09,551][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:15:10,052][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:15:10,553][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:15:11,052][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:15:11,556][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:15:12,055][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:15:12,555][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:15:13,054][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:15:13,555][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:15:14,063][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:15:14,569][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:15:15,069][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:15:15,567][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:15:16,064][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:15:16,561][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:15:17,057][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:15:17,555][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:15:18,059][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:15:18,557][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:15:19,058][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:15:19,558][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:15:20,057][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:15:20,558][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:15:21,061][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:15:21,562][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:15:22,066][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:15:22,564][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10223 tokens. +[2026-03-26 02:15:23,297][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.38%, ΔTime: 00:00:32 +[2026-03-26 02:15:24,047][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:15:24,049][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:15:24,051][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:15:24,890][__main__][INFO] - Iteration 289 took 53s (32.55% Gen, 65.87% Train). Generation: 17s, Training: 35s. Estimated remaining time: 40h 11m 10s. Estimated total time: 44h 25m 46s. Time estimates for 10 more iterations: 8m 53s, 100 more iterations: 1h 28m 51s, 500 more iterations: 7h 24m 17s. +[2026-03-26 02:15:24,893][__main__][INFO] - Starting iteration 289. +[2026-03-26 02:15:25,296][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 28 and human policies 1. +[2026-03-26 02:15:25,297][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:15:43,445][__main__][INFO] - Number of regex retries in iteration 289: 0 +[2026-03-26 02:15:43,446][__main__][INFO] - agents played in iteration 289 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:15:44,222][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:15:44,245][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:15:44,269][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:15:44,291][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:15:44,292][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:15:44,293][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:15:44,990][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:15:45,443][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:15:45,951][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:15:46,451][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:15:46,954][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:15:47,455][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:15:47,953][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:15:48,453][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:15:48,952][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:15:49,451][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:15:49,953][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:15:50,451][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:15:50,952][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:15:51,459][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:15:51,959][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:15:52,458][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:15:52,958][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:15:53,483][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:15:53,988][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:15:54,501][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:15:55,006][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:15:55,510][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:15:56,013][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:15:56,514][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:15:57,015][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:15:57,517][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:15:58,019][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:15:58,519][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:15:59,018][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:15:59,520][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:16:00,021][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:16:00,523][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:16:01,024][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:16:01,521][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:16:02,017][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:16:02,514][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:16:03,019][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:16:03,519][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:16:04,025][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:16:04,520][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:16:05,021][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:16:05,520][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:16:06,018][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:16:06,517][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:16:07,015][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:16:07,518][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:16:08,015][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:16:08,514][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:16:09,012][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:16:09,508][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:16:10,008][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:16:10,507][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:16:11,006][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:16:11,503][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:16:12,005][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:16:12,506][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:16:13,004][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:16:13,503][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:16:14,002][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:16:14,497][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:16:14,995][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:16:15,497][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:16:15,996][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:16:16,496][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:16:16,992][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10345 tokens. +[2026-03-26 02:16:17,730][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.38%, ΔTime: 00:00:32 +[2026-03-26 02:16:18,483][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:16:18,485][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:16:18,487][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:16:19,209][__main__][INFO] - Iteration 290 took 53s (33.66% Gen, 65.00% Train). Generation: 18s, Training: 35s. Estimated remaining time: 40h 40m 7s. Estimated total time: 44h 55m 38s. Time estimates for 10 more iterations: 8m 59s, 100 more iterations: 1h 29m 51s, 500 more iterations: 7h 29m 16s. +[2026-03-26 02:16:19,211][__main__][INFO] - Starting iteration 290. +[2026-03-26 02:16:19,613][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 28 and human policies 1. +[2026-03-26 02:16:19,614][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:16:37,587][__main__][INFO] - Number of regex retries in iteration 290: 0 +[2026-03-26 02:16:37,588][__main__][INFO] - agents played in iteration 290 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:16:38,373][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:16:38,396][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:16:38,419][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:16:38,442][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:16:38,443][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:16:38,443][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:16:39,134][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:16:39,592][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:16:40,093][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:16:40,595][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:16:41,092][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:16:41,591][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:16:42,094][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:16:42,596][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:16:43,095][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:16:43,594][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:16:44,090][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:16:44,588][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:16:45,089][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:16:45,590][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:16:46,091][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:16:46,588][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:16:47,086][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:16:47,584][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:16:48,087][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:16:48,589][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:16:49,090][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:16:49,587][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:16:50,086][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:16:50,587][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:16:51,083][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:16:51,584][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:16:52,081][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:16:52,583][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:16:53,081][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:16:53,584][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:16:54,084][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:16:54,604][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:16:55,109][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:16:55,609][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:16:56,112][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:16:56,614][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:16:57,112][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:16:57,614][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:16:58,114][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:16:58,615][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:16:59,116][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:16:59,618][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:17:00,114][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:17:00,611][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:17:01,109][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:17:01,608][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:17:02,109][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:17:02,606][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:17:03,103][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:17:03,608][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:17:04,108][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:17:04,609][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:17:05,110][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:17:05,610][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:17:06,110][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:17:06,612][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:17:07,115][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:17:07,613][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:17:08,110][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:17:08,611][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:17:09,109][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:17:09,606][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:17:10,104][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:17:10,601][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:17:11,100][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10326 tokens. +[2026-03-26 02:17:11,845][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.33%, ΔTime: 00:00:32 +[2026-03-26 02:17:12,596][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:17:12,599][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:17:12,601][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:17:13,940][__main__][INFO] - Iteration 291 took 54s (33.08% Gen, 64.45% Train). Generation: 17s, Training: 35s. Estimated remaining time: 40h 59m 57s. Estimated total time: 45h 16m 22s. Time estimates for 10 more iterations: 9m 3s, 100 more iterations: 1h 30m 32s, 500 more iterations: 7h 32m 43s. +[2026-03-26 02:17:13,942][__main__][INFO] - Starting iteration 291. +[2026-03-26 02:17:14,349][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 29 and human policies 1. +[2026-03-26 02:17:14,350][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:17:21,365][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:17:31,409][__main__][INFO] - Number of regex retries in iteration 291: 1 +[2026-03-26 02:17:31,410][__main__][INFO] - agents played in iteration 291 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:17:32,184][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:17:32,207][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:17:32,230][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:17:32,253][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:17:32,254][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:17:32,255][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:17:32,952][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:17:33,405][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:17:33,910][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:17:34,414][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:17:34,916][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:17:35,418][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:17:35,916][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:17:36,415][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:17:36,914][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:17:37,415][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:17:37,916][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:17:38,418][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:17:38,923][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:17:39,423][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:17:39,926][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:17:40,429][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:17:40,928][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:17:41,434][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:17:41,933][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:17:42,432][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:17:42,932][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:17:43,434][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:17:43,935][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:17:44,446][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:17:44,952][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:17:45,462][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:17:45,965][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:17:46,467][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:17:46,967][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:17:47,469][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:17:47,980][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:17:48,482][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:17:48,980][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:17:49,478][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:17:49,975][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:17:50,473][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:17:50,974][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:17:51,472][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:17:51,970][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:17:52,469][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:17:52,968][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:17:53,471][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:17:53,969][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:17:54,471][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:17:54,971][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:17:55,471][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:17:55,975][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:17:56,476][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:17:56,976][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:17:57,476][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:17:57,976][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:17:58,473][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:17:58,978][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:17:59,476][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:17:59,977][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:18:00,478][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:18:00,975][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:18:01,478][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:18:01,979][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:18:02,481][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:18:02,985][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:18:03,489][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:18:03,994][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:18:04,497][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:18:04,997][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10305 tokens. +[2026-03-26 02:18:05,725][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.60%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 02:18:06,499][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:18:06,501][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:18:06,503][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:18:07,236][__main__][INFO] - Iteration 292 took 52s (32.26% Gen, 66.35% Train). Generation: 17s, Training: 35s. Estimated remaining time: 39h 47m 5s. Estimated total time: 44h 4m 23s. Time estimates for 10 more iterations: 8m 48s, 100 more iterations: 1h 28m 8s, 500 more iterations: 7h 20m 43s. +[2026-03-26 02:18:07,238][__main__][INFO] - Starting iteration 292. +[2026-03-26 02:18:07,640][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 29 and human policies 1. +[2026-03-26 02:18:07,641][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:18:24,599][__main__][INFO] - Number of regex retries in iteration 292: 0 +[2026-03-26 02:18:24,600][__main__][INFO] - agents played in iteration 292 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:18:25,374][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:18:25,397][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:18:25,420][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:18:25,443][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:18:25,444][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:18:25,444][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:18:26,124][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:18:26,583][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:18:27,089][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:18:27,591][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:18:28,090][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:18:28,592][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:18:29,093][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:18:29,592][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:18:30,094][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:18:30,597][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:18:31,103][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:18:31,605][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:18:32,105][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:18:32,604][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:18:33,106][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:18:33,611][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:18:34,114][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:18:34,610][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:18:35,108][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:18:35,611][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:18:36,118][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:18:36,624][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:18:37,129][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:18:37,629][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:18:38,126][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:18:38,627][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:18:39,124][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:18:39,625][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:18:40,126][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:18:40,623][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:18:41,120][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:18:41,617][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:18:42,114][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:18:42,612][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:18:43,111][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:18:43,609][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:18:44,111][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:18:44,616][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:18:45,114][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:18:45,613][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:18:46,115][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:18:46,610][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:18:47,107][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:18:47,603][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:18:48,100][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:18:48,596][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:18:49,091][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:18:49,590][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:18:50,089][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:18:50,590][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:18:51,089][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:18:51,589][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:18:52,089][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:18:52,585][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:18:53,085][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:18:53,585][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:18:54,081][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:18:54,577][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:18:55,074][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:18:55,571][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:18:56,071][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:18:56,567][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:18:57,067][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:18:57,564][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:18:58,060][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10333 tokens. +[2026-03-26 02:18:58,773][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.23%, ΔTime: 00:00:32 +[2026-03-26 02:18:59,514][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:18:59,517][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:18:59,519][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:19:00,274][__main__][INFO] - Iteration 293 took 52s (32.22% Gen, 66.34% Train). Generation: 16s, Training: 34s. Estimated remaining time: 39h 33m 34s. Estimated total time: 43h 51m 45s. Time estimates for 10 more iterations: 8m 46s, 100 more iterations: 1h 27m 43s, 500 more iterations: 7h 18m 37s. +[2026-03-26 02:19:00,277][__main__][INFO] - Starting iteration 293. +[2026-03-26 02:19:00,675][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 29 and human policies 1. +[2026-03-26 02:19:00,675][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:19:18,443][__main__][INFO] - Number of regex retries in iteration 293: 0 +[2026-03-26 02:19:18,444][__main__][INFO] - agents played in iteration 293 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:19:19,216][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:19:19,240][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:19:19,263][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:19:19,286][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:19:19,286][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:19:19,287][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:19:19,965][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:19:20,421][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:19:20,924][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:19:21,422][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:19:21,920][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:19:22,422][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:19:22,919][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:19:23,419][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:19:23,918][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:19:24,418][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:19:24,918][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:19:25,421][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:19:25,924][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:19:26,422][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:19:26,920][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:19:27,418][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:19:27,917][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:19:28,420][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:19:28,925][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:19:29,428][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:19:29,930][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:19:30,431][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:19:30,933][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:19:31,434][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:19:31,939][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:19:32,446][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:19:32,948][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:19:33,447][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:19:33,946][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:19:34,445][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:19:34,946][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:19:35,445][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:19:35,945][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:19:36,443][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:19:36,940][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:19:37,438][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:19:37,936][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:19:38,435][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:19:38,933][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:19:39,431][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:19:39,929][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:19:40,427][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:19:40,926][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:19:41,423][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:19:41,920][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:19:42,417][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:19:42,914][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:19:43,411][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:19:43,910][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:19:44,412][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:19:44,922][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:19:45,423][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:19:45,924][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:19:46,422][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:19:46,922][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:19:47,419][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:19:47,919][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:19:48,419][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:19:48,920][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:19:49,418][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:19:49,916][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:19:50,418][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:19:50,917][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:19:51,417][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:19:51,914][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10243 tokens. +[2026-03-26 02:19:52,633][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 02:19:53,389][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:19:53,392][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:19:53,393][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:19:54,090][__main__][INFO] - Iteration 294 took 53s (33.26% Gen, 65.43% Train). Generation: 17s, Training: 34s. Estimated remaining time: 40h 11m 42s. Estimated total time: 44h 30m 47s. Time estimates for 10 more iterations: 8m 54s, 100 more iterations: 1h 29m 1s, 500 more iterations: 7h 25m 7s. +[2026-03-26 02:19:54,092][__main__][INFO] - Starting iteration 294. +[2026-03-26 02:19:54,494][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 29 and human policies 1. +[2026-03-26 02:19:54,495][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:20:12,670][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:20:12,911][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:20:26,338][__main__][INFO] - Number of regex retries in iteration 294: 2 +[2026-03-26 02:20:26,339][__main__][INFO] - agents played in iteration 294 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:20:27,119][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:20:27,142][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:20:27,165][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:20:27,189][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:20:27,189][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:20:27,190][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:20:27,878][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:20:28,332][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:20:28,851][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:20:29,354][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:20:29,859][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:20:30,358][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:20:30,857][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:20:31,356][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:20:31,863][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:20:32,366][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:20:32,867][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:20:33,366][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:20:33,866][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:20:34,365][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:20:34,870][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:20:35,376][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:20:35,876][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:20:36,380][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:20:36,882][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:20:37,383][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:20:37,903][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:20:38,407][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:20:38,913][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:20:39,415][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:20:39,915][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:20:40,413][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:20:40,911][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:20:41,409][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:20:41,913][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:20:42,419][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:20:42,919][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:20:43,419][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:20:43,919][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:20:44,418][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:20:44,918][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:20:45,418][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:20:45,916][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:20:46,413][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:20:46,914][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:20:47,422][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:20:47,931][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:20:48,437][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:20:48,939][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:20:49,442][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:20:49,945][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:20:50,448][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:20:50,948][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:20:51,450][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:20:51,951][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:20:52,449][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:20:52,947][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:20:53,456][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:20:53,956][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:20:54,455][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:20:54,953][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:20:55,455][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:20:55,953][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:20:56,455][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:20:56,956][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:20:57,457][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:20:57,958][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:20:58,456][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:20:58,954][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:20:59,452][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:20:59,951][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10240 tokens. +[2026-03-26 02:21:00,682][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.23%, ΔTime: 00:00:32 +[2026-03-26 02:21:01,441][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:21:01,443][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:21:01,445][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:21:02,178][__main__][INFO] - Iteration 295 took 1m 7s (47.05% Gen, 51.87% Train). Generation: 31s, Training: 35s. Estimated remaining time: 52h 4m 1s. Estimated total time: 56h 24m 15s. Time estimates for 10 more iterations: 11m 16s, 100 more iterations: 1h 52m 48s, 500 more iterations: 9h 24m 2s. +[2026-03-26 02:21:02,181][__main__][INFO] - Starting iteration 295. +[2026-03-26 02:21:02,579][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 29 and human policies 1. +[2026-03-26 02:21:02,580][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:21:22,260][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:21:27,701][__main__][INFO] - Number of regex retries in iteration 295: 1 +[2026-03-26 02:21:27,701][__main__][INFO] - agents played in iteration 295 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:21:28,475][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:21:28,499][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:21:28,522][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:21:28,545][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:21:28,545][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:21:28,546][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:21:29,241][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:21:29,695][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:21:30,199][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:21:30,698][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:21:31,197][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:21:31,698][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:21:32,203][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:21:32,701][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:21:33,200][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:21:33,701][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:21:34,201][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:21:34,701][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:21:35,198][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:21:35,698][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:21:36,199][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:21:36,708][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:21:37,210][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:21:37,710][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:21:38,213][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:21:38,710][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:21:39,209][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:21:39,709][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:21:40,208][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:21:40,707][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:21:41,207][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:21:41,703][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:21:42,201][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:21:42,698][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:21:43,193][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:21:43,691][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:21:44,187][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:21:44,684][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:21:45,185][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:21:45,707][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:21:46,211][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:21:46,713][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:21:47,212][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:21:47,714][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:21:48,216][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:21:48,717][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:21:49,217][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:21:49,717][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:21:50,217][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:21:50,717][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:21:51,218][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:21:51,718][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:21:52,215][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:21:52,717][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:21:53,216][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:21:53,713][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:21:54,210][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:21:54,709][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:21:55,207][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:21:55,705][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:21:56,201][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:21:56,706][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:21:57,213][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:21:57,715][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:21:58,214][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:21:58,716][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:21:59,220][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:21:59,722][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:22:00,226][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:22:00,725][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:22:01,223][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10260 tokens. +[2026-03-26 02:22:01,951][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.47%, ΔTime: 00:00:32 +[2026-03-26 02:22:02,700][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:22:02,702][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:22:02,704][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:22:03,432][__main__][INFO] - Iteration 296 took 1m 0s (41.28% Gen, 57.52% Train). Generation: 25s, Training: 35s. Estimated remaining time: 46h 21m 25s. Estimated total time: 50h 42m 40s. Time estimates for 10 more iterations: 10m 8s, 100 more iterations: 1h 41m 25s, 500 more iterations: 8h 27m 6s. +[2026-03-26 02:22:03,434][__main__][INFO] - Starting iteration 296. +[2026-03-26 02:22:03,834][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 29 and human policies 1. +[2026-03-26 02:22:03,835][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:22:13,752][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:22:20,278][__main__][INFO] - Number of regex retries in iteration 296: 1 +[2026-03-26 02:22:20,279][__main__][INFO] - agents played in iteration 296 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:22:21,052][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:22:21,076][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:22:21,099][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:22:21,122][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:22:21,123][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:22:21,124][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:22:21,803][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:22:22,259][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:22:22,763][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:22:23,260][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:22:23,758][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:22:24,260][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:22:24,760][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:22:25,259][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:22:25,756][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:22:26,259][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:22:26,756][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:22:27,256][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:22:27,756][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:22:28,255][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:22:28,754][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:22:29,253][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:22:29,755][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:22:30,259][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:22:30,920][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:22:31,420][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:22:31,930][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:22:32,431][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:22:32,934][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:22:33,434][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:22:33,937][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:22:34,435][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:22:34,934][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:22:35,433][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:22:35,930][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:22:36,426][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:22:36,922][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:22:37,418][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:22:37,916][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:22:38,414][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:22:38,913][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:22:39,415][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:22:39,914][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:22:40,416][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:22:40,914][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:22:41,413][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:22:41,912][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:22:42,413][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:22:42,912][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:22:43,413][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:22:43,913][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:22:44,414][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:22:44,914][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:22:45,424][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:22:45,925][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:22:46,428][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:22:46,928][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:22:47,429][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:22:47,927][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:22:48,424][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:22:48,926][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:22:49,423][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:22:49,922][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:22:50,424][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:22:50,925][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:22:51,424][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:22:51,923][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:22:52,422][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:22:52,918][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:22:53,417][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:22:53,918][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10279 tokens. +[2026-03-26 02:22:54,628][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.27%, Current % of VRAM taken: 60.70%, Block Peak % of device VRAM: 62.35%, ΔTime: 00:00:32 +[2026-03-26 02:22:55,370][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:22:55,372][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:22:55,374][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:22:56,094][__main__][INFO] - Iteration 297 took 52s (31.47% Gen, 67.15% Train). Generation: 16s, Training: 35s. Estimated remaining time: 39h 10m 53s. Estimated total time: 43h 33m 1s. Time estimates for 10 more iterations: 8m 42s, 100 more iterations: 1h 27m 6s, 500 more iterations: 7h 15m 30s. +[2026-03-26 02:22:56,096][__main__][INFO] - Starting iteration 297. +[2026-03-26 02:22:56,495][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 29 and human policies 1. +[2026-03-26 02:22:56,495][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:23:21,487][__main__][INFO] - Number of regex retries in iteration 297: 0 +[2026-03-26 02:23:21,488][__main__][INFO] - agents played in iteration 297 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:23:22,260][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:23:22,284][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:23:22,308][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:23:22,331][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:23:22,331][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:23:22,332][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:23:23,006][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:23:23,463][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:23:23,971][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:23:24,472][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:23:24,973][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:23:25,474][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:23:25,975][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:23:26,475][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:23:26,976][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:23:27,473][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:23:27,974][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:23:28,471][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:23:28,969][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:23:29,470][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:23:29,970][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:23:30,471][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:23:30,973][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:23:31,471][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:23:31,971][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:23:32,470][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:23:32,973][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:23:33,474][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:23:33,976][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:23:34,478][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:23:34,981][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:23:35,477][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:23:35,975][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:23:36,474][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:23:36,972][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:23:37,471][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:23:37,969][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:23:38,467][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:23:38,964][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:23:39,468][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:23:39,969][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:23:40,467][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:23:40,969][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:23:41,471][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:23:41,972][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:23:42,469][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:23:42,969][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:23:43,469][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:23:43,968][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:23:44,466][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:23:44,966][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:23:45,463][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:23:45,962][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:23:46,462][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:23:46,962][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:23:47,459][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:23:47,961][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:23:48,461][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:23:48,960][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:23:49,458][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:23:49,957][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:23:50,462][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:23:50,967][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:23:51,473][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:23:51,976][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:23:52,475][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:23:52,972][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:23:53,473][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:23:53,974][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:23:54,474][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:23:54,973][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10387 tokens. +[2026-03-26 02:23:55,677][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.23%, ΔTime: 00:00:32 +[2026-03-26 02:23:56,415][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:23:56,418][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:23:56,419][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:23:57,098][__main__][INFO] - Iteration 298 took 1m 0s (41.24% Gen, 57.64% Train). Generation: 24s, Training: 34s. Estimated remaining time: 46h 7m 2s. Estimated total time: 50h 30m 11s. Time estimates for 10 more iterations: 10m 6s, 100 more iterations: 1h 41m 0s, 500 more iterations: 8h 25m 1s. +[2026-03-26 02:23:57,100][__main__][INFO] - Starting iteration 298. +[2026-03-26 02:23:57,508][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 29 and human policies 1. +[2026-03-26 02:23:57,508][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:24:04,837][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:24:17,163][__main__][INFO] - Number of regex retries in iteration 298: 1 +[2026-03-26 02:24:17,164][__main__][INFO] - agents played in iteration 298 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:24:17,947][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:24:17,971][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:24:17,994][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:24:18,017][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:24:18,017][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:24:18,018][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:24:18,715][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:24:19,170][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:24:19,676][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:24:20,176][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:24:20,683][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:24:21,186][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:24:21,687][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:24:22,190][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:24:22,690][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:24:23,190][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:24:23,689][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:24:24,189][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:24:24,688][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:24:25,187][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:24:25,689][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:24:26,189][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:24:26,686][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:24:27,185][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:24:27,682][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:24:28,179][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:24:28,675][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:24:29,170][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:24:29,668][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:24:30,166][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:24:30,667][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:24:31,164][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:24:31,662][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:24:32,161][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:24:32,658][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:24:33,157][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:24:33,653][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:24:34,150][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:24:34,648][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:24:35,150][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:24:35,649][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:24:36,146][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:24:36,648][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:24:37,149][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:24:37,650][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:24:38,150][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:24:38,653][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:24:39,151][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:24:39,648][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:24:40,145][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:24:40,642][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:24:41,138][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:24:41,634][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:24:42,132][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:24:42,629][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:24:43,129][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:24:43,627][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:24:44,127][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:24:44,624][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:24:45,126][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:24:45,622][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:24:46,122][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:24:46,621][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:24:47,115][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:24:47,612][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:24:48,110][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:24:48,608][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:24:49,104][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:24:49,601][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:24:50,100][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:24:50,597][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10175 tokens. +[2026-03-26 02:24:51,308][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 62.26%, ΔTime: 00:00:32 +[2026-03-26 02:24:52,064][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:24:52,067][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:24:52,068][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:24:52,932][__main__][INFO] - Iteration 299 took 55s (35.46% Gen, 62.98% Train). Generation: 19s, Training: 34s. Estimated remaining time: 41h 47m 11s. Estimated total time: 46h 11m 15s. Time estimates for 10 more iterations: 9m 14s, 100 more iterations: 1h 32m 22s, 500 more iterations: 7h 41m 52s. +[2026-03-26 02:24:52,934][__main__][INFO] - Starting iteration 299. +[2026-03-26 02:24:53,384][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 29 and human policies 1. +[2026-03-26 02:24:53,385][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:24:58,146][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:25:00,692][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:25:11,005][__main__][INFO] - Number of regex retries in iteration 299: 2 +[2026-03-26 02:25:11,006][__main__][INFO] - agents played in iteration 299 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:25:11,780][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:25:11,804][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:25:11,827][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:25:11,850][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:25:11,850][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:25:11,851][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:25:12,543][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:25:12,997][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:25:13,502][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:25:14,004][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:25:14,506][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:25:15,009][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:25:15,508][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:25:16,012][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:25:16,516][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:25:17,016][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:25:17,516][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:25:18,017][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:25:18,516][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:25:19,014][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:25:19,512][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:25:20,011][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:25:20,510][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:25:21,012][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:25:21,512][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:25:22,013][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:25:22,514][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:25:23,015][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:25:23,516][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:25:24,017][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:25:24,517][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:25:25,014][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:25:25,511][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:25:26,007][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:25:26,507][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:25:27,004][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:25:27,500][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:25:27,997][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:25:28,495][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:25:28,994][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:25:29,491][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:25:29,993][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:25:30,490][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:25:30,987][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:25:31,482][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:25:31,978][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:25:32,474][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:25:32,972][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:25:33,470][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:25:33,966][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:25:34,463][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:25:34,963][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:25:35,458][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:25:35,957][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:25:36,453][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:25:36,954][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:25:37,456][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:25:37,953][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:25:38,452][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:25:38,952][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:25:39,449][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:25:39,948][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:25:40,448][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:25:40,953][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:25:41,451][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:25:41,949][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:25:42,447][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:25:42,948][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:25:43,444][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:25:43,943][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:25:44,441][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10255 tokens. +[2026-03-26 02:25:45,180][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.17%, ΔTime: 00:00:32 +[2026-03-26 02:25:45,918][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:25:45,920][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:25:45,921][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:25:46,786][__main__][INFO] - Iteration 300 took 53s (33.00% Gen, 65.38% Train). Generation: 17s, Training: 34s. Estimated remaining time: 40h 5m 10s. Estimated total time: 44h 30m 8s. Time estimates for 10 more iterations: 8m 54s, 100 more iterations: 1h 29m 0s, 500 more iterations: 7h 25m 1s. +[2026-03-26 02:25:46,789][__main__][INFO] - Starting iteration 300. +[2026-03-26 02:25:47,191][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 29 and human policies 1. +[2026-03-26 02:25:47,192][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:26:01,836][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 11 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:26:03,318][__main__][INFO] - Number of regex retries in iteration 300: 1 +[2026-03-26 02:26:03,319][__main__][INFO] - agents played in iteration 300 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:26:04,091][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:26:04,114][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:26:04,137][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:26:04,160][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:26:04,161][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:26:04,161][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:26:04,879][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:26:05,336][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:26:05,842][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:26:06,342][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:26:06,848][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:26:07,351][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:26:07,851][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:26:08,352][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:26:08,856][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:26:09,355][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:26:09,857][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:26:10,359][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:26:10,858][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:26:11,355][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:26:11,854][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:26:12,353][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:26:12,851][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:26:13,354][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:26:13,851][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:26:14,349][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:26:14,848][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:26:15,345][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:26:15,842][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:26:16,339][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:26:16,837][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:26:17,338][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:26:17,837][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:26:18,338][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:26:18,835][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:26:19,333][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:26:19,831][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:26:20,331][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:26:20,830][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:26:21,330][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:26:21,832][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:26:22,330][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:26:22,828][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:26:23,330][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:26:23,827][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:26:24,324][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:26:24,822][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:26:25,318][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:26:25,814][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:26:26,309][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:26:26,805][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:26:27,301][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:26:27,795][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:26:28,290][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:26:28,784][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:26:29,282][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:26:29,783][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:26:30,284][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:26:30,786][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:26:31,287][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:26:31,786][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:26:32,287][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:26:32,799][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:26:33,298][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:26:33,797][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:26:34,299][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:26:34,798][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:26:35,298][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:26:35,796][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:26:36,294][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:26:36,793][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10069 tokens. +[2026-03-26 02:26:37,575][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 02:26:38,350][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:26:38,352][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:26:38,354][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:26:39,687][__main__][INFO] - Iteration 301 took 52s (30.72% Gen, 66.74% Train). Generation: 16s, Training: 35s. Estimated remaining time: 39h 18m 57s. Estimated total time: 43h 44m 48s. Time estimates for 10 more iterations: 8m 44s, 100 more iterations: 1h 27m 29s, 500 more iterations: 7h 17m 28s. +[2026-03-26 02:26:39,689][__main__][INFO] - Starting iteration 301. +[2026-03-26 02:26:40,090][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 30 and human policies 1. +[2026-03-26 02:26:40,091][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:26:57,093][__main__][INFO] - Number of regex retries in iteration 301: 0 +[2026-03-26 02:26:57,094][__main__][INFO] - agents played in iteration 301 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:26:57,865][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:26:57,889][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:26:57,912][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:26:57,935][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:26:57,936][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:26:57,936][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:26:58,620][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:26:59,077][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:26:59,583][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:27:00,090][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:27:00,591][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:27:01,096][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:27:01,598][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:27:02,099][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:27:02,608][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:27:03,109][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:27:03,610][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:27:04,110][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:27:04,610][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:27:05,107][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:27:05,607][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:27:06,105][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:27:06,606][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:27:07,117][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:27:07,618][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:27:08,118][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:27:08,616][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:27:09,119][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:27:09,616][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:27:10,118][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:27:10,626][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:27:11,126][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:27:11,629][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:27:12,131][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:27:12,633][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:27:13,134][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:27:13,634][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:27:14,131][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:27:14,633][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:27:15,137][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:27:15,637][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:27:16,138][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:27:16,639][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:27:17,145][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:27:17,645][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:27:18,144][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:27:18,644][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:27:19,141][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:27:19,639][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:27:20,143][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:27:20,642][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:27:21,143][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:27:21,644][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:27:22,143][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:27:22,648][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:27:23,145][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:27:23,641][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:27:24,138][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:27:24,634][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:27:25,133][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:27:25,630][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:27:26,135][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:27:26,634][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:27:27,133][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:27:27,635][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:27:28,135][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:27:28,634][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:27:29,132][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:27:29,632][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:27:30,132][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:27:30,632][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10304 tokens. +[2026-03-26 02:27:31,369][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.16%, Current % of VRAM taken: 60.58%, Block Peak % of device VRAM: 62.28%, ΔTime: 00:00:32 +[2026-03-26 02:27:32,115][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:27:32,117][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:27:32,119][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:27:32,847][__main__][INFO] - Iteration 302 took 52s (32.23% Gen, 66.39% Train). Generation: 17s, Training: 35s. Estimated remaining time: 39h 31m 8s. Estimated total time: 43h 57m 52s. Time estimates for 10 more iterations: 8m 47s, 100 more iterations: 1h 27m 55s, 500 more iterations: 7h 19m 38s. +[2026-03-26 02:27:32,849][__main__][INFO] - Starting iteration 302. +[2026-03-26 02:27:33,250][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 30 and human policies 1. +[2026-03-26 02:27:33,251][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:27:38,039][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:27:38,217][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:27:39,227][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:27:51,059][__main__][INFO] - Number of regex retries in iteration 302: 3 +[2026-03-26 02:27:51,060][__main__][INFO] - agents played in iteration 302 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:27:51,833][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:27:51,857][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:27:51,880][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:27:51,903][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:27:51,903][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:27:51,904][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:27:52,621][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:27:53,073][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:27:53,575][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:27:54,074][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:27:54,574][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:27:55,073][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:27:55,574][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:27:56,072][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:27:56,570][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:27:57,074][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:27:57,597][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:27:58,094][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:27:58,595][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:27:59,096][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:27:59,594][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:28:00,094][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:28:00,592][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:28:01,090][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:28:01,588][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:28:02,088][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:28:02,584][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:28:03,080][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:28:03,575][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:28:04,073][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:28:04,570][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:28:05,068][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:28:05,569][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:28:06,069][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:28:06,567][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:28:07,073][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:28:07,573][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:28:08,075][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:28:08,577][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:28:09,078][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:28:09,578][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:28:10,078][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:28:10,574][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:28:11,074][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:28:11,575][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:28:12,075][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:28:12,575][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:28:13,073][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:28:13,570][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:28:14,071][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:28:14,569][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:28:15,067][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:28:15,564][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:28:16,061][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:28:16,557][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:28:17,054][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:28:17,549][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:28:18,050][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:28:18,548][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:28:19,045][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:28:19,542][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:28:20,039][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:28:20,537][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:28:21,034][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:28:21,532][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:28:22,030][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:28:22,532][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:28:23,037][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:28:23,537][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:28:24,035][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:28:24,534][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10129 tokens. +[2026-03-26 02:28:25,312][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.19%, ΔTime: 00:00:32 +[2026-03-26 02:28:26,075][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:28:26,077][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:28:26,079][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:28:26,924][__main__][INFO] - Iteration 303 took 53s (33.18% Gen, 65.24% Train). Generation: 17s, Training: 35s. Estimated remaining time: 40h 16m 3s. Estimated total time: 44h 43m 42s. Time estimates for 10 more iterations: 8m 56s, 100 more iterations: 1h 29m 27s, 500 more iterations: 7h 27m 17s. +[2026-03-26 02:28:26,928][__main__][INFO] - Starting iteration 303. +[2026-03-26 02:28:27,332][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 30 and human policies 1. +[2026-03-26 02:28:27,333][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:28:31,919][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:28:32,152][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:28:44,565][__main__][INFO] - Number of regex retries in iteration 303: 2 +[2026-03-26 02:28:44,566][__main__][INFO] - agents played in iteration 303 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:28:45,343][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:28:45,367][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:28:45,390][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:28:45,413][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:28:45,414][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:28:45,415][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:28:46,117][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:28:46,574][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:28:47,081][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:28:47,581][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:28:48,083][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:28:48,586][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:28:49,084][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:28:49,582][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:28:50,087][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:28:50,587][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:28:51,094][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:28:51,599][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:28:52,100][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:28:52,602][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:28:53,103][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:28:53,603][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:28:54,102][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:28:54,609][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:28:55,110][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:28:55,613][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:28:56,115][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:28:56,613][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:28:57,112][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:28:57,610][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:28:58,108][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:28:58,616][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:28:59,115][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:28:59,614][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:29:00,112][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:29:00,610][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:29:01,108][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:29:01,610][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:29:02,107][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:29:02,608][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:29:03,106][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:29:03,606][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:29:04,107][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:29:04,606][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:29:05,104][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:29:05,602][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:29:06,100][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:29:06,597][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:29:07,099][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:29:07,598][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:29:08,097][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:29:08,600][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:29:09,099][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:29:09,599][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:29:10,097][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:29:10,596][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:29:11,094][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:29:11,595][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:29:12,094][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:29:12,592][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:29:13,092][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:29:13,591][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:29:14,099][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:29:14,600][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:29:15,099][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:29:15,596][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:29:16,094][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:29:16,596][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:29:17,097][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:29:17,595][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:29:18,100][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10280 tokens. +[2026-03-26 02:29:18,832][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.22%, Current % of VRAM taken: 60.64%, Block Peak % of device VRAM: 62.27%, ΔTime: 00:00:32 +[2026-03-26 02:29:19,594][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:29:19,596][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:29:19,598][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:29:20,413][__main__][INFO] - Iteration 304 took 53s (32.47% Gen, 66.00% Train). Generation: 17s, Training: 35s. Estimated remaining time: 39h 45m 33s. Estimated total time: 44h 14m 5s. Time estimates for 10 more iterations: 8m 50s, 100 more iterations: 1h 28m 28s, 500 more iterations: 7h 22m 20s. +[2026-03-26 02:29:20,415][__main__][INFO] - Starting iteration 304. +[2026-03-26 02:29:20,855][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 30 and human policies 1. +[2026-03-26 02:29:20,856][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:29:38,004][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:29:41,551][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:29:49,149][__main__][INFO] - Number of regex retries in iteration 304: 2 +[2026-03-26 02:29:49,150][__main__][INFO] - agents played in iteration 304 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:29:49,933][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:29:49,956][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:29:49,980][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:29:50,003][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:29:50,004][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:29:50,004][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:29:50,707][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:29:51,160][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:29:51,663][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:29:52,161][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:29:52,659][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:29:53,165][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:29:53,665][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:29:54,164][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:29:54,662][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:29:55,161][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:29:55,662][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:29:56,162][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:29:56,660][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:29:57,159][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:29:57,657][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:29:58,154][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:29:58,655][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:29:59,156][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:29:59,660][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:30:00,177][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:30:00,678][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:30:01,176][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:30:01,678][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:30:02,176][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:30:02,677][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:30:03,173][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:30:03,670][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:30:04,168][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:30:04,667][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:30:05,164][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:30:05,660][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:30:06,155][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:30:06,651][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:30:07,151][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:30:07,649][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:30:08,147][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:30:08,644][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:30:09,146][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:30:09,650][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:30:10,156][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:30:10,654][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:30:11,154][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:30:11,654][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:30:12,152][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:30:12,652][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:30:13,158][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:30:13,659][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:30:14,166][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:30:14,668][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:30:15,174][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:30:15,678][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:30:16,183][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:30:16,688][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:30:17,190][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:30:17,691][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:30:18,192][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:30:18,692][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:30:19,191][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:30:19,692][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:30:20,190][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:30:20,687][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:30:21,184][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:30:21,689][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:30:22,189][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:30:22,690][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10159 tokens. +[2026-03-26 02:30:23,454][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.37%, ΔTime: 00:00:32 +[2026-03-26 02:30:24,204][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:30:24,207][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:30:24,209][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:30:24,923][__main__][INFO] - Iteration 305 took 1m 4s (44.16% Gen, 54.72% Train). Generation: 28s, Training: 35s. Estimated remaining time: 48h 53m 47s. Estimated total time: 53h 23m 23s. Time estimates for 10 more iterations: 10m 40s, 100 more iterations: 1h 46m 46s, 500 more iterations: 8h 53m 53s. +[2026-03-26 02:30:24,925][__main__][INFO] - Starting iteration 305. +[2026-03-26 02:30:25,328][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 30 and human policies 1. +[2026-03-26 02:30:25,329][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:30:40,744][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 20 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:30:41,155][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 20 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 2/3 +[2026-03-26 02:30:42,011][__main__][INFO] - Number of regex retries in iteration 305: 2 +[2026-03-26 02:30:42,012][__main__][INFO] - agents played in iteration 305 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:30:42,797][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:30:42,821][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:30:42,844][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:30:42,867][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:30:42,867][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:30:42,868][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:30:43,576][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:30:44,029][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:30:44,549][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:30:45,048][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:30:45,546][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:30:46,047][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:30:46,545][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:30:47,044][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:30:47,552][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:30:48,049][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:30:48,546][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:30:49,043][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:30:49,539][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:30:50,035][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:30:50,534][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:30:51,031][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:30:51,530][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:30:52,039][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:30:52,542][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:30:53,040][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:30:53,540][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:30:54,037][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:30:54,536][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:30:55,036][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:30:55,545][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:30:56,049][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:30:56,547][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:30:57,046][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:30:57,547][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:30:58,048][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:30:58,546][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:30:59,047][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:30:59,558][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:31:00,054][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:31:00,555][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:31:01,063][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:31:01,568][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:31:02,070][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:31:02,570][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:31:03,070][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:31:03,570][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:31:04,072][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:31:04,579][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:31:05,078][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:31:05,581][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:31:06,081][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:31:06,588][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:31:07,090][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:31:07,590][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:31:08,092][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:31:08,594][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:31:09,093][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:31:09,597][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:31:10,103][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:31:10,607][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:31:11,116][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:31:11,620][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:31:12,130][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:31:12,631][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:31:13,134][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:31:13,637][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:31:14,141][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:31:14,645][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:31:15,148][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:31:15,651][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10310 tokens. +[2026-03-26 02:31:16,411][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.16%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.17%, ΔTime: 00:00:32 +[2026-03-26 02:31:17,168][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:31:17,171][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:31:17,172][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:31:18,043][__main__][INFO] - Iteration 306 took 52s (31.65% Gen, 66.70% Train). Generation: 16s, Training: 35s. Estimated remaining time: 39h 25m 19s. Estimated total time: 43h 55m 48s. Time estimates for 10 more iterations: 8m 47s, 100 more iterations: 1h 27m 51s, 500 more iterations: 7h 19m 18s. +[2026-03-26 02:31:18,046][__main__][INFO] - Starting iteration 306. +[2026-03-26 02:31:18,451][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 30 and human policies 1. +[2026-03-26 02:31:18,451][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:31:22,692][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:31:35,525][__main__][INFO] - Number of regex retries in iteration 306: 1 +[2026-03-26 02:31:35,526][__main__][INFO] - agents played in iteration 306 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:31:36,306][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:31:36,330][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:31:36,354][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:31:36,377][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:31:36,378][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:31:36,378][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:31:37,086][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:31:37,543][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:31:38,049][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:31:38,557][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:31:39,061][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:31:39,563][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:31:40,063][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:31:40,565][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:31:41,069][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:31:41,568][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:31:42,073][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:31:42,579][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:31:43,083][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:31:43,583][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:31:44,083][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:31:44,580][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:31:45,078][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:31:45,577][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:31:46,075][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:31:46,574][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:31:47,073][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:31:47,572][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:31:48,070][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:31:48,569][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:31:49,067][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:31:49,570][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:31:50,074][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:31:50,592][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:31:51,093][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:31:51,596][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:31:52,095][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:31:52,594][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:31:53,094][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:31:53,593][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:31:54,091][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:31:54,590][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:31:55,089][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:31:55,592][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:31:56,094][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:31:56,596][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:31:57,094][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:31:57,593][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:31:58,091][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:31:58,590][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:31:59,089][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:31:59,589][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:32:00,089][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:32:00,586][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:32:01,088][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:32:01,585][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:32:02,082][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:32:02,580][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:32:03,079][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:32:03,588][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:32:04,091][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:32:04,603][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:32:05,104][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:32:05,611][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:32:06,118][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:32:06,619][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:32:07,123][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:32:07,623][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:32:08,124][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:32:08,627][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:32:09,128][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10195 tokens. +[2026-03-26 02:32:09,876][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 62.15%, ΔTime: 00:00:32 +[2026-03-26 02:32:10,628][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:32:10,630][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:32:10,632][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:32:11,451][__main__][INFO] - Iteration 307 took 53s (32.22% Gen, 66.24% Train). Generation: 17s, Training: 35s. Estimated remaining time: 39h 38m 41s. Estimated total time: 44h 10m 3s. Time estimates for 10 more iterations: 8m 50s, 100 more iterations: 1h 28m 20s, 500 more iterations: 7h 21m 40s. +[2026-03-26 02:32:11,453][__main__][INFO] - Starting iteration 307. +[2026-03-26 02:32:11,854][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 30 and human policies 1. +[2026-03-26 02:32:11,855][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:32:17,025][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:32:29,574][__main__][INFO] - Number of regex retries in iteration 307: 1 +[2026-03-26 02:32:29,575][__main__][INFO] - agents played in iteration 307 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:32:30,355][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:32:30,379][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:32:30,402][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:32:30,425][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:32:30,426][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:32:30,426][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:32:31,133][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:32:31,587][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:32:32,094][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:32:32,593][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:32:33,091][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:32:33,589][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:32:34,088][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:32:34,587][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:32:35,089][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:32:35,585][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:32:36,081][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:32:36,578][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:32:37,074][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:32:37,575][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:32:38,072][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:32:38,573][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:32:39,069][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:32:39,570][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:32:40,073][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:32:40,575][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:32:41,074][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:32:41,576][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:32:42,075][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:32:42,573][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:32:43,073][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:32:43,570][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:32:44,070][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:32:44,569][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:32:45,078][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:32:45,577][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:32:46,076][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:32:46,574][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:32:47,072][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:32:47,575][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:32:48,083][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:32:48,593][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:32:49,094][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:32:49,592][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:32:50,091][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:32:50,590][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:32:51,089][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:32:51,587][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:32:52,085][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:32:52,583][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:32:53,083][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:32:53,580][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:32:54,077][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:32:54,574][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:32:55,072][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:32:55,570][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:32:56,072][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:32:56,574][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:32:57,076][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:32:57,575][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:32:58,073][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:32:58,574][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:32:59,075][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:32:59,576][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:33:00,080][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:33:00,580][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:33:01,076][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:33:01,574][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:33:02,071][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:33:02,569][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:33:03,071][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10134 tokens. +[2026-03-26 02:33:03,827][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.17%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.17%, ΔTime: 00:00:32 +[2026-03-26 02:33:04,570][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:33:04,572][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:33:04,574][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:33:05,266][__main__][INFO] - Iteration 308 took 53s (33.18% Gen, 65.53% Train). Generation: 17s, Training: 34s. Estimated remaining time: 39h 58m 19s. Estimated total time: 44h 30m 36s. Time estimates for 10 more iterations: 8m 54s, 100 more iterations: 1h 29m 1s, 500 more iterations: 7h 25m 6s. +[2026-03-26 02:33:05,267][__main__][INFO] - Starting iteration 308. +[2026-03-26 02:33:05,674][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 30 and human policies 1. +[2026-03-26 02:33:05,674][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:33:23,128][__main__][INFO] - Number of regex retries in iteration 308: 0 +[2026-03-26 02:33:23,129][__main__][INFO] - agents played in iteration 308 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:33:23,906][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:33:23,930][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:33:23,953][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:33:23,976][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:33:23,977][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:33:23,977][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:33:24,686][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:33:25,137][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:33:25,642][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:33:26,138][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:33:26,638][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:33:27,143][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:33:27,643][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:33:28,147][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:33:28,644][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:33:29,143][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:33:29,649][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:33:30,158][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:33:30,657][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:33:31,158][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:33:31,655][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:33:32,155][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:33:32,653][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:33:33,155][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:33:33,655][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:33:34,162][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:33:34,664][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:33:35,168][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:33:35,666][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:33:36,164][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:33:36,667][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:33:37,168][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:33:37,666][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:33:38,168][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:33:38,680][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:33:39,189][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:33:39,689][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:33:40,187][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:33:40,690][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:33:41,189][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:33:41,691][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:33:42,193][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:33:42,694][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:33:43,192][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:33:43,692][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:33:44,200][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:33:44,701][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:33:45,200][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:33:45,702][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:33:46,203][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:33:46,699][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:33:47,200][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:33:47,698][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:33:48,196][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:33:48,697][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:33:49,197][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:33:49,697][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:33:50,196][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:33:50,697][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:33:51,196][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:33:51,693][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:33:52,194][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:33:52,700][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:33:53,201][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:33:53,701][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:33:54,209][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:33:54,710][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:33:55,211][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:33:55,710][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:33:56,208][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:33:56,705][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10336 tokens. +[2026-03-26 02:33:57,455][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.04%, Current % of VRAM taken: 60.47%, Block Peak % of device VRAM: 62.19%, ΔTime: 00:00:32 +[2026-03-26 02:33:58,284][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:33:58,287][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:33:58,288][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:33:59,032][__main__][INFO] - Iteration 309 took 53s (32.71% Gen, 65.89% Train). Generation: 17s, Training: 35s. Estimated remaining time: 39h 54m 45s. Estimated total time: 44h 27m 56s. Time estimates for 10 more iterations: 8m 53s, 100 more iterations: 1h 28m 55s, 500 more iterations: 7h 24m 39s. +[2026-03-26 02:33:59,034][__main__][INFO] - Starting iteration 309. +[2026-03-26 02:33:59,454][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 30 and human policies 1. +[2026-03-26 02:33:59,455][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:34:16,406][__main__][INFO] - Number of regex retries in iteration 309: 0 +[2026-03-26 02:34:16,407][__main__][INFO] - agents played in iteration 309 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:34:17,176][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:34:17,200][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:34:17,223][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:34:17,246][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:34:17,247][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:34:17,248][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:34:17,966][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:34:18,419][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:34:18,930][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:34:19,430][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:34:19,928][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:34:20,428][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:34:20,929][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:34:21,425][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:34:21,921][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:34:22,431][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:34:22,930][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:34:23,431][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:34:23,930][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:34:24,432][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:34:24,930][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:34:25,427][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:34:25,925][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:34:26,423][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:34:26,927][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:34:27,428][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:34:27,926][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:34:28,424][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:34:28,921][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:34:29,418][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:34:29,919][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:34:30,425][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:34:30,927][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:34:31,447][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:34:31,950][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:34:32,451][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:34:32,950][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:34:33,449][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:34:33,947][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:34:34,445][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:34:34,943][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:34:35,440][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:34:35,938][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:34:36,435][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:34:36,932][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:34:37,430][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:34:37,938][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:34:38,436][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:34:38,933][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:34:39,431][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:34:39,928][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:34:40,426][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:34:40,923][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:34:41,422][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:34:41,919][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:34:42,418][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:34:42,914][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:34:43,412][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:34:43,910][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:34:44,412][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:34:44,909][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:34:45,417][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:34:45,916][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:34:46,415][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:34:46,918][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:34:47,418][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:34:47,922][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:34:48,429][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:34:48,935][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:34:49,438][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:34:49,938][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10096 tokens. +[2026-03-26 02:34:50,694][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.08%, ΔTime: 00:00:32 +[2026-03-26 02:34:51,466][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:34:51,468][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:34:51,470][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:34:52,294][__main__][INFO] - Iteration 310 took 52s (32.08% Gen, 66.35% Train). Generation: 16s, Training: 35s. Estimated remaining time: 39h 27m 59s. Estimated total time: 44h 2m 3s. Time estimates for 10 more iterations: 8m 48s, 100 more iterations: 1h 28m 4s, 500 more iterations: 7h 20m 20s. +[2026-03-26 02:34:52,297][__main__][INFO] - Starting iteration 310. +[2026-03-26 02:34:52,707][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 30 and human policies 1. +[2026-03-26 02:34:52,707][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:35:08,963][__main__][INFO] - Number of regex retries in iteration 310: 0 +[2026-03-26 02:35:08,964][__main__][INFO] - agents played in iteration 310 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:35:09,740][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:35:09,764][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:35:09,787][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:35:09,810][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:35:09,811][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:35:09,811][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:35:10,518][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:35:10,975][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:35:11,478][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:35:11,976][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:35:12,474][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:35:12,976][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:35:13,476][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:35:13,975][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:35:14,477][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:35:14,978][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:35:15,477][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:35:15,979][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:35:16,481][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:35:16,981][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:35:17,480][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:35:17,980][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:35:18,479][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:35:18,975][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:35:19,473][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:35:19,973][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:35:20,475][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:35:20,978][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:35:21,477][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:35:21,976][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:35:22,476][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:35:22,972][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:35:23,470][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:35:23,968][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:35:24,467][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:35:24,964][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:35:25,460][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:35:25,956][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:35:26,452][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:35:26,949][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:35:27,447][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:35:27,945][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:35:28,441][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:35:28,937][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:35:29,434][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:35:29,931][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:35:30,431][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:35:30,934][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:35:31,435][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:35:31,934][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:35:32,440][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:35:32,942][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:35:33,445][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:35:33,942][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:35:34,445][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:35:34,941][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:35:35,440][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:35:35,941][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:35:36,438][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:35:36,936][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:35:37,438][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:35:37,935][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:35:38,433][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:35:38,931][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:35:39,431][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:35:39,930][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:35:40,432][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:35:40,935][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:35:41,439][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:35:41,939][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:35:42,443][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10198 tokens. +[2026-03-26 02:35:43,190][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.24%, Current % of VRAM taken: 60.67%, Block Peak % of device VRAM: 62.27%, ΔTime: 00:00:32 +[2026-03-26 02:35:44,013][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:35:44,015][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:35:44,017][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:35:45,510][__main__][INFO] - Iteration 311 took 52s (30.79% Gen, 66.38% Train). Generation: 16s, Training: 35s. Estimated remaining time: 39h 25m 15s. Estimated total time: 44h 0m 12s. Time estimates for 10 more iterations: 8m 48s, 100 more iterations: 1h 28m 0s, 500 more iterations: 7h 20m 2s. +[2026-03-26 02:35:45,512][__main__][INFO] - Starting iteration 311. +[2026-03-26 02:35:45,914][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 31 and human policies 1. +[2026-03-26 02:35:45,915][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:35:50,712][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:35:58,922][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 20 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:36:03,192][__main__][INFO] - Number of regex retries in iteration 311: 2 +[2026-03-26 02:36:03,193][__main__][INFO] - agents played in iteration 311 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:36:03,967][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:36:03,990][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:36:04,014][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:36:04,036][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:36:04,037][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:36:04,038][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:36:04,747][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:36:05,200][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:36:05,704][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:36:06,203][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:36:06,706][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:36:07,202][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:36:07,700][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:36:08,200][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:36:08,702][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:36:09,199][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:36:09,696][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:36:10,193][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:36:10,690][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:36:11,186][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:36:11,689][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:36:12,193][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:36:12,695][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:36:13,190][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:36:13,687][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:36:14,182][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:36:14,679][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:36:15,176][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:36:15,673][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:36:16,170][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:36:16,667][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:36:17,163][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:36:17,667][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:36:18,165][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:36:18,663][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:36:19,160][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:36:19,656][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:36:20,154][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:36:20,654][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:36:21,152][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:36:21,651][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:36:22,152][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:36:22,648][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:36:23,144][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:36:23,640][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:36:24,136][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:36:24,632][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:36:25,131][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:36:25,629][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:36:26,129][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:36:26,629][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:36:27,130][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:36:27,628][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:36:28,127][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:36:28,628][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:36:29,132][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:36:29,632][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:36:30,128][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:36:30,627][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:36:31,127][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:36:31,630][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:36:32,134][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:36:32,636][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:36:33,137][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:36:33,635][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:36:34,136][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:36:34,637][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:36:35,140][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:36:35,640][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:36:36,141][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:36:36,641][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10146 tokens. +[2026-03-26 02:36:37,380][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 62.27%, ΔTime: 00:00:32 +[2026-03-26 02:36:38,095][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:36:38,098][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:36:38,099][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:36:39,238][__main__][INFO] - Iteration 312 took 53s (32.40% Gen, 65.46% Train). Generation: 17s, Training: 34s. Estimated remaining time: 39h 50m 30s. Estimated total time: 44h 26m 20s. Time estimates for 10 more iterations: 8m 53s, 100 more iterations: 1h 28m 52s, 500 more iterations: 7h 24m 23s. +[2026-03-26 02:36:39,241][__main__][INFO] - Starting iteration 312. +[2026-03-26 02:36:39,642][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 31 and human policies 1. +[2026-03-26 02:36:39,643][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:36:45,851][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:36:51,039][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:36:57,561][__main__][INFO] - Number of regex retries in iteration 312: 2 +[2026-03-26 02:36:57,562][__main__][INFO] - agents played in iteration 312 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:36:58,346][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:36:58,372][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:36:58,397][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:36:58,422][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:36:58,423][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:36:58,423][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:36:59,141][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:36:59,601][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:37:00,109][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:37:00,608][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:37:01,108][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:37:01,617][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:37:02,118][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:37:02,623][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:37:03,122][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:37:03,621][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:37:04,118][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:37:04,616][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:37:05,122][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:37:05,630][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:37:06,129][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:37:06,632][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:37:07,131][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:37:07,627][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:37:08,127][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:37:08,625][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:37:09,123][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:37:09,623][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:37:10,128][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:37:10,625][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:37:11,123][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:37:11,621][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:37:12,118][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:37:12,614][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:37:13,109][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:37:13,607][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:37:14,108][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:37:14,606][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:37:15,105][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:37:15,603][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:37:16,102][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:37:16,603][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:37:17,101][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:37:17,601][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:37:18,105][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:37:18,609][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:37:19,117][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:37:19,619][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:37:20,121][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:37:20,622][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:37:21,121][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:37:21,621][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:37:22,119][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:37:22,618][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:37:23,116][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:37:23,614][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:37:24,112][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:37:24,611][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:37:25,112][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:37:25,617][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:37:26,119][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:37:26,620][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:37:27,119][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:37:27,621][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:37:28,123][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:37:28,635][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:37:29,137][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:37:29,640][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:37:30,146][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:37:30,650][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:37:31,151][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10125 tokens. +[2026-03-26 02:37:31,885][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 62.18%, ΔTime: 00:00:32 +[2026-03-26 02:37:32,626][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:37:32,628][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:37:32,630][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:37:33,321][__main__][INFO] - Iteration 313 took 53s (33.38% Gen, 65.33% Train). Generation: 17s, Training: 35s. Estimated remaining time: 40h 7m 14s. Estimated total time: 44h 43m 59s. Time estimates for 10 more iterations: 8m 56s, 100 more iterations: 1h 29m 27s, 500 more iterations: 7h 27m 19s. +[2026-03-26 02:37:33,324][__main__][INFO] - Starting iteration 313. +[2026-03-26 02:37:33,730][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 31 and human policies 1. +[2026-03-26 02:37:33,730][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:37:51,389][__main__][INFO] - Number of regex retries in iteration 313: 0 +[2026-03-26 02:37:51,390][__main__][INFO] - agents played in iteration 313 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:37:52,166][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:37:52,190][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:37:52,213][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:37:52,237][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:37:52,237][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:37:52,238][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:37:52,922][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:37:53,374][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:37:53,876][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:37:54,374][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:37:54,872][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:37:55,370][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:37:55,867][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:37:56,363][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:37:56,862][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:37:57,360][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:37:57,857][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:37:58,356][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:37:58,854][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:37:59,351][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:37:59,856][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:38:00,354][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:38:00,851][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:38:01,354][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:38:01,859][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:38:02,360][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:38:02,857][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:38:03,354][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:38:03,854][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:38:04,352][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:38:04,850][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:38:05,356][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:38:05,857][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:38:06,354][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:38:06,852][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:38:07,350][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:38:07,848][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:38:08,346][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:38:08,843][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:38:09,370][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:38:09,871][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:38:10,370][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:38:10,873][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:38:11,372][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:38:11,873][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:38:12,374][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:38:12,872][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:38:13,373][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:38:13,875][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:38:14,372][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:38:14,870][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:38:15,368][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:38:15,868][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:38:16,373][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:38:16,877][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:38:17,378][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:38:17,880][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:38:18,379][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:38:18,879][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:38:19,383][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:38:19,885][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:38:20,395][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:38:20,897][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:38:21,402][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:38:21,903][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:38:22,403][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:38:22,902][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:38:23,401][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:38:23,899][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:38:24,397][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:38:24,896][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10158 tokens. +[2026-03-26 02:38:25,619][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.15%, Current % of VRAM taken: 60.58%, Block Peak % of device VRAM: 62.21%, ΔTime: 00:00:32 +[2026-03-26 02:38:26,369][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:38:26,372][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:38:26,373][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:38:27,073][__main__][INFO] - Iteration 314 took 53s (33.11% Gen, 65.58% Train). Generation: 17s, Training: 34s. Estimated remaining time: 39h 49m 33s. Estimated total time: 44h 27m 12s. Time estimates for 10 more iterations: 8m 53s, 100 more iterations: 1h 28m 54s, 500 more iterations: 7h 24m 32s. +[2026-03-26 02:38:27,075][__main__][INFO] - Starting iteration 314. +[2026-03-26 02:38:27,481][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 31 and human policies 1. +[2026-03-26 02:38:27,481][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:38:45,409][__main__][INFO] - Number of regex retries in iteration 314: 0 +[2026-03-26 02:38:45,410][__main__][INFO] - agents played in iteration 314 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:38:46,189][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:38:46,213][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:38:46,236][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:38:46,260][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:38:46,260][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:38:46,261][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:38:46,980][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:38:47,436][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:38:47,941][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:38:48,446][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:38:48,945][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:38:49,443][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:38:49,944][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:38:50,442][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:38:50,961][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:38:51,485][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:38:51,995][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:38:52,507][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:38:53,008][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:38:53,506][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:38:54,010][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:38:54,510][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:38:55,011][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:38:55,535][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:38:56,046][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:38:56,557][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:38:57,069][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:38:57,572][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:38:58,071][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:38:58,572][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:38:59,075][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:38:59,580][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:39:00,080][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:39:00,579][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:39:01,078][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:39:01,576][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:39:02,074][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:39:02,574][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:39:03,078][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:39:03,577][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:39:04,077][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:39:04,576][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:39:05,075][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:39:05,573][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:39:06,071][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:39:06,571][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:39:07,080][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:39:07,582][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:39:08,087][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:39:08,591][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:39:09,097][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:39:09,597][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:39:10,098][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:39:10,599][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:39:11,100][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:39:11,602][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:39:12,117][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:39:12,620][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:39:13,124][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:39:13,627][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:39:14,128][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:39:14,629][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:39:15,128][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:39:15,628][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:39:16,127][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:39:16,626][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:39:17,125][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:39:17,629][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:39:18,138][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:39:18,641][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:39:19,139][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10056 tokens. +[2026-03-26 02:39:19,895][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.19%, ΔTime: 00:00:32 +[2026-03-26 02:39:20,637][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:39:20,640][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:39:20,642][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:39:21,390][__main__][INFO] - Iteration 315 took 53s (33.26% Gen, 65.35% Train). Generation: 17s, Training: 35s. Estimated remaining time: 40h 16m 56s. Estimated total time: 44h 55m 29s. Time estimates for 10 more iterations: 8m 59s, 100 more iterations: 1h 29m 50s, 500 more iterations: 7h 29m 14s. +[2026-03-26 02:39:21,392][__main__][INFO] - Starting iteration 315. +[2026-03-26 02:39:21,791][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 31 and human policies 1. +[2026-03-26 02:39:21,792][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:39:38,487][__main__][INFO] - Number of regex retries in iteration 315: 0 +[2026-03-26 02:39:38,488][__main__][INFO] - agents played in iteration 315 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:39:39,257][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:39:39,280][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:39:39,303][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:39:39,326][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:39:39,327][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:39:39,328][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:39:40,018][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:39:40,477][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:39:40,984][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:39:41,484][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:39:41,981][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:39:42,480][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:39:42,976][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:39:43,479][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:39:43,980][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:39:44,480][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:39:44,987][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:39:45,494][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:39:45,995][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:39:46,497][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:39:46,996][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:39:47,498][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:39:47,998][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:39:48,497][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:39:48,996][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:39:49,494][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:39:49,995][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:39:50,493][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:39:50,991][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:39:51,489][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:39:51,989][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:39:52,488][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:39:52,986][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:39:53,488][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:39:53,985][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:39:54,482][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:39:54,979][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:39:55,479][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:39:55,976][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:39:56,474][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:39:56,973][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:39:57,471][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:39:57,972][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:39:58,473][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:39:58,975][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:39:59,481][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:39:59,994][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:40:00,498][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:40:00,998][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:40:01,659][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:40:02,159][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:40:02,660][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:40:03,165][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:40:03,665][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:40:04,163][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:40:04,667][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:40:05,169][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:40:05,671][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:40:06,198][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:40:06,700][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:40:07,202][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:40:07,704][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:40:08,208][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:40:08,706][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:40:09,203][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:40:09,700][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:40:10,198][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:40:10,695][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:40:11,194][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:40:11,693][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:40:12,191][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10155 tokens. +[2026-03-26 02:40:12,929][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.13%, ΔTime: 00:00:32 +[2026-03-26 02:40:13,704][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:40:13,707][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:40:13,709][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:40:14,486][__main__][INFO] - Iteration 316 took 52s (31.68% Gen, 66.84% Train). Generation: 16s, Training: 35s. Estimated remaining time: 39h 15m 18s. Estimated total time: 43h 54m 44s. Time estimates for 10 more iterations: 8m 46s, 100 more iterations: 1h 27m 49s, 500 more iterations: 7h 19m 7s. +[2026-03-26 02:40:14,494][__main__][INFO] - Starting iteration 316. +[2026-03-26 02:40:14,894][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 31 and human policies 1. +[2026-03-26 02:40:14,895][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:40:35,001][__main__][INFO] - Number of regex retries in iteration 316: 0 +[2026-03-26 02:40:35,002][__main__][INFO] - agents played in iteration 316 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:40:35,785][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:40:35,809][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:40:35,833][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:40:35,856][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:40:35,856][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:40:35,857][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:40:36,548][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:40:37,000][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:40:37,521][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:40:38,027][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:40:38,526][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:40:39,024][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:40:39,522][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:40:40,029][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:40:40,529][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:40:41,029][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:40:41,528][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:40:42,026][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:40:42,524][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:40:43,029][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:40:43,530][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:40:44,026][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:40:44,522][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:40:45,021][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:40:45,518][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:40:46,018][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:40:46,520][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:40:47,018][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:40:47,526][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:40:48,027][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:40:48,529][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:40:49,028][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:40:49,529][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:40:50,030][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:40:50,530][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:40:51,037][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:40:51,537][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:40:52,036][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:40:52,534][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:40:53,036][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:40:53,538][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:40:54,043][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:40:54,546][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:40:55,051][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:40:55,561][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:40:56,067][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:40:56,568][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:40:57,068][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:40:57,569][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:40:58,069][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:40:58,571][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:40:59,070][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:40:59,572][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:41:00,071][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:41:00,570][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:41:01,072][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:41:01,571][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:41:02,070][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:41:02,573][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:41:03,071][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:41:03,573][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:41:04,074][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:41:04,587][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:41:05,091][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:41:05,615][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:41:06,112][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:41:06,613][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:41:07,114][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:41:07,615][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:41:08,113][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:41:08,612][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10212 tokens. +[2026-03-26 02:41:09,337][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.60%, Block Peak % of device VRAM: 62.23%, ΔTime: 00:00:32 +[2026-03-26 02:41:10,084][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:41:10,086][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:41:10,088][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:41:10,920][__main__][INFO] - Iteration 317 took 56s (35.89% Gen, 62.62% Train). Generation: 20s, Training: 35s. Estimated remaining time: 42h 0m 58s. Estimated total time: 46h 41m 21s. Time estimates for 10 more iterations: 9m 20s, 100 more iterations: 1h 33m 22s, 500 more iterations: 7h 46m 53s. +[2026-03-26 02:41:10,923][__main__][INFO] - Starting iteration 317. +[2026-03-26 02:41:11,326][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 31 and human policies 1. +[2026-03-26 02:41:11,327][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:41:28,268][__main__][INFO] - Number of regex retries in iteration 317: 0 +[2026-03-26 02:41:28,269][__main__][INFO] - agents played in iteration 317 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:41:29,040][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:41:29,064][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:41:29,087][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:41:29,110][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:41:29,111][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:41:29,111][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:41:29,799][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:41:30,256][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:41:30,762][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:41:31,263][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:41:31,760][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:41:32,261][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:41:32,762][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:41:33,262][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:41:33,757][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:41:34,254][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:41:34,752][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:41:35,257][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:41:35,758][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:41:36,257][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:41:36,755][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:41:37,253][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:41:37,752][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:41:38,251][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:41:38,750][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:41:39,249][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:41:39,748][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:41:40,248][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:41:40,747][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:41:41,245][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:41:41,743][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:41:42,245][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:41:42,748][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:41:43,247][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:41:43,746][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:41:44,245][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:41:44,751][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:41:45,254][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:41:45,753][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:41:46,253][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:41:46,758][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:41:47,263][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:41:47,769][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:41:48,272][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:41:48,774][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:41:49,275][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:41:49,775][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:41:50,276][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:41:50,780][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:41:51,281][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:41:51,781][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:41:52,281][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:41:52,779][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:41:53,277][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:41:53,775][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:41:54,274][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:41:54,779][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:41:55,278][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:41:55,777][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:41:56,276][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:41:56,778][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:41:57,275][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:41:57,774][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:41:58,272][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:41:58,771][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:41:59,270][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:41:59,769][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:42:00,270][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:42:00,768][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:42:01,270][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:42:01,769][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10149 tokens. +[2026-03-26 02:42:02,499][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.20%, ΔTime: 00:00:32 +[2026-03-26 02:42:03,251][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:42:03,253][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:42:03,255][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:42:04,003][__main__][INFO] - Iteration 318 took 52s (32.16% Gen, 66.41% Train). Generation: 16s, Training: 34s. Estimated remaining time: 39h 12m 37s. Estimated total time: 43h 53m 53s. Time estimates for 10 more iterations: 8m 46s, 100 more iterations: 1h 27m 47s, 500 more iterations: 7h 18m 58s. +[2026-03-26 02:42:04,005][__main__][INFO] - Starting iteration 318. +[2026-03-26 02:42:04,408][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 31 and human policies 1. +[2026-03-26 02:42:04,409][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:42:22,051][__main__][INFO] - Number of regex retries in iteration 318: 0 +[2026-03-26 02:42:22,051][__main__][INFO] - agents played in iteration 318 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:42:22,865][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:42:22,889][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:42:22,913][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:42:22,936][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:42:22,936][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:42:22,937][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:42:23,627][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:42:24,087][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:42:24,598][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:42:25,100][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:42:25,603][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:42:26,110][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:42:26,611][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:42:27,112][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:42:27,611][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:42:28,111][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:42:28,614][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:42:29,121][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:42:29,622][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:42:30,123][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:42:30,626][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:42:31,128][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:42:31,639][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:42:32,138][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:42:32,640][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:42:33,137][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:42:33,636][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:42:34,134][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:42:34,633][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:42:35,132][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:42:35,638][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:42:36,136][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:42:36,636][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:42:37,135][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:42:37,635][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:42:38,141][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:42:38,649][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:42:39,150][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:42:39,661][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:42:40,164][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:42:40,664][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:42:41,172][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:42:41,674][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:42:42,175][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:42:42,681][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:42:43,182][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:42:43,683][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:42:44,185][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:42:44,694][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:42:45,193][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:42:45,691][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:42:46,191][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:42:46,693][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:42:47,190][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:42:47,689][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:42:48,187][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:42:48,684][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:42:49,182][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:42:49,681][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:42:50,182][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:42:50,681][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:42:51,183][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:42:51,681][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:42:52,180][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:42:52,678][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:42:53,180][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:42:53,678][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:42:54,174][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:42:54,674][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:42:55,173][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:42:55,671][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10152 tokens. +[2026-03-26 02:42:56,402][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.17%, ΔTime: 00:00:32 +[2026-03-26 02:42:57,220][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:42:57,223][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:42:57,225][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:42:57,912][__main__][INFO] - Iteration 319 took 53s (32.97% Gen, 65.74% Train). Generation: 17s, Training: 35s. Estimated remaining time: 39h 53m 4s. Estimated total time: 44h 35m 13s. Time estimates for 10 more iterations: 8m 55s, 100 more iterations: 1h 29m 10s, 500 more iterations: 7h 25m 52s. +[2026-03-26 02:42:57,914][__main__][INFO] - Starting iteration 319. +[2026-03-26 02:42:58,333][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 31 and human policies 1. +[2026-03-26 02:42:58,334][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:43:15,444][__main__][INFO] - Number of regex retries in iteration 319: 0 +[2026-03-26 02:43:15,445][__main__][INFO] - agents played in iteration 319 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:43:16,216][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:43:16,240][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:43:16,263][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:43:16,286][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:43:16,287][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:43:16,287][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:43:16,982][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:43:17,434][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:43:17,946][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:43:18,450][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:43:18,948][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:43:19,446][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:43:19,944][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:43:20,442][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:43:20,940][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:43:21,442][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:43:21,942][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:43:22,448][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:43:22,949][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:43:23,451][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:43:23,952][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:43:24,453][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:43:24,955][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:43:25,453][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:43:25,951][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:43:26,450][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:43:26,948][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:43:27,445][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:43:27,944][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:43:28,442][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:43:28,942][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:43:29,442][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:43:29,939][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:43:30,440][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:43:30,945][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:43:31,458][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:43:31,963][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:43:32,466][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:43:32,967][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:43:33,473][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:43:33,978][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:43:34,501][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:43:35,006][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:43:35,511][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:43:36,016][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:43:36,520][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:43:37,021][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:43:37,523][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:43:38,021][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:43:38,520][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:43:39,019][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:43:39,517][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:43:40,016][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:43:40,518][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:43:41,016][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:43:41,515][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:43:42,012][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:43:42,513][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:43:43,013][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:43:43,510][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:43:44,007][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:43:44,505][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:43:45,002][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:43:45,498][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:43:45,994][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:43:46,491][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:43:46,989][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:43:47,485][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:43:47,981][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:43:48,479][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:43:48,975][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10129 tokens. +[2026-03-26 02:43:49,697][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.23%, ΔTime: 00:00:32 +[2026-03-26 02:43:50,462][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:43:50,465][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:43:50,466][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:43:51,170][__main__][INFO] - Iteration 320 took 52s (32.38% Gen, 66.28% Train). Generation: 17s, Training: 35s. Estimated remaining time: 39h 18m 49s. Estimated total time: 44h 1m 51s. Time estimates for 10 more iterations: 8m 48s, 100 more iterations: 1h 28m 3s, 500 more iterations: 7h 20m 18s. +[2026-03-26 02:43:51,172][__main__][INFO] - Starting iteration 320. +[2026-03-26 02:43:51,574][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 31 and human policies 1. +[2026-03-26 02:43:51,575][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:44:08,777][__main__][INFO] - Number of regex retries in iteration 320: 0 +[2026-03-26 02:44:08,778][__main__][INFO] - agents played in iteration 320 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:44:09,560][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:44:09,584][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:44:09,608][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:44:09,632][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:44:09,633][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:44:09,633][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:44:10,336][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:44:10,790][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:44:11,295][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:44:11,805][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:44:12,306][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:44:12,807][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:44:13,306][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:44:13,806][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:44:14,316][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:44:14,822][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:44:15,324][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:44:15,821][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:44:16,322][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:44:16,819][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:44:17,319][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:44:17,822][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:44:18,320][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:44:18,819][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:44:19,328][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:44:19,827][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:44:20,329][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:44:20,828][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:44:21,327][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:44:21,825][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:44:22,324][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:44:22,830][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:44:23,332][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:44:23,838][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:44:24,342][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:44:24,846][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:44:25,350][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:44:25,851][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:44:26,354][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:44:26,867][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:44:27,366][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:44:27,868][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:44:28,370][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:44:28,873][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:44:29,373][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:44:29,872][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:44:30,372][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:44:30,870][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:44:31,372][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:44:31,887][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:44:32,387][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:44:32,886][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:44:33,388][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:44:33,891][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:44:34,390][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:44:34,891][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:44:35,393][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:44:35,892][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:44:36,392][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:44:36,894][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:44:37,396][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:44:37,907][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:44:38,409][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:44:38,908][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:44:39,406][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:44:39,910][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:44:40,410][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:44:40,909][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:44:41,408][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:44:41,908][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:44:42,407][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10155 tokens. +[2026-03-26 02:44:43,133][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 62.12%, ΔTime: 00:00:32 +[2026-03-26 02:44:44,054][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:44:44,056][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:44:44,058][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:44:45,312][__main__][INFO] - Iteration 321 took 53s (32.01% Gen, 65.65% Train). Generation: 17s, Training: 35s. Estimated remaining time: 40h 3m 0s. Estimated total time: 44h 46m 57s. Time estimates for 10 more iterations: 8m 57s, 100 more iterations: 1h 29m 33s, 500 more iterations: 7h 27m 49s. +[2026-03-26 02:44:45,315][__main__][INFO] - Starting iteration 321. +[2026-03-26 02:44:45,727][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 32 and human policies 1. +[2026-03-26 02:44:45,728][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:45:02,691][__main__][INFO] - Number of regex retries in iteration 321: 0 +[2026-03-26 02:45:02,692][__main__][INFO] - agents played in iteration 321 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:45:03,470][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:45:03,494][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:45:03,517][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:45:03,541][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:45:03,541][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:45:03,542][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:45:04,241][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:45:04,697][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:45:05,199][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:45:05,697][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:45:06,194][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:45:06,696][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:45:07,196][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:45:07,693][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:45:08,191][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:45:08,688][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:45:09,187][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:45:09,685][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:45:10,192][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:45:10,693][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:45:11,199][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:45:11,699][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:45:12,200][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:45:12,704][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:45:13,204][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:45:13,713][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:45:14,216][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:45:14,721][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:45:15,219][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:45:15,719][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:45:16,222][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:45:16,727][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:45:17,230][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:45:17,733][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:45:18,236][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:45:18,736][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:45:19,266][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:45:19,766][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:45:20,268][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:45:20,772][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:45:21,271][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:45:21,772][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:45:22,272][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:45:22,774][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:45:23,274][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:45:23,774][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:45:24,273][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:45:24,776][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:45:25,275][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:45:25,777][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:45:26,275][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:45:26,773][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:45:27,275][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:45:27,773][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:45:28,275][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:45:28,773][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:45:29,272][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:45:29,773][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:45:30,276][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:45:30,774][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:45:31,273][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:45:31,773][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:45:32,275][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:45:32,776][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:45:33,277][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:45:33,774][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:45:34,275][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:45:34,776][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:45:35,273][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:45:35,774][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:45:36,277][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10253 tokens. +[2026-03-26 02:45:37,006][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.17%, Current % of VRAM taken: 60.60%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 02:45:37,761][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:45:37,763][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:45:37,765][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:45:38,502][__main__][INFO] - Iteration 322 took 52s (32.14% Gen, 66.46% Train). Generation: 16s, Training: 35s. Estimated remaining time: 39h 13m 56s. Estimated total time: 43h 58m 46s. Time estimates for 10 more iterations: 8m 47s, 100 more iterations: 1h 27m 57s, 500 more iterations: 7h 19m 47s. +[2026-03-26 02:45:38,504][__main__][INFO] - Starting iteration 322. +[2026-03-26 02:45:38,946][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 32 and human policies 1. +[2026-03-26 02:45:38,947][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:45:43,657][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:45:56,327][__main__][INFO] - Number of regex retries in iteration 322: 1 +[2026-03-26 02:45:56,328][__main__][INFO] - agents played in iteration 322 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:45:57,105][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:45:57,129][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:45:57,152][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:45:57,175][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:45:57,176][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:45:57,176][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:45:57,859][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:45:58,313][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:45:58,815][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:45:59,315][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:45:59,811][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:46:00,307][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:46:00,803][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:46:01,299][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:46:01,795][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:46:02,293][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:46:02,794][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:46:03,294][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:46:03,794][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:46:04,293][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:46:04,795][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:46:05,297][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:46:05,807][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:46:06,306][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:46:06,807][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:46:07,334][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:46:07,834][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:46:08,334][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:46:08,833][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:46:09,335][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:46:09,837][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:46:10,337][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:46:10,838][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:46:11,338][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:46:11,838][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:46:12,335][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:46:12,833][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:46:13,331][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:46:13,829][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:46:14,326][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:46:14,822][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:46:15,319][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:46:15,820][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:46:16,318][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:46:16,819][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:46:17,317][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:46:17,816][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:46:18,316][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:46:18,814][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:46:19,314][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:46:19,812][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:46:20,319][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:46:20,822][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:46:21,320][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:46:21,821][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:46:22,321][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:46:22,818][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:46:23,316][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:46:23,815][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:46:24,315][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:46:24,816][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:46:25,314][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:46:25,812][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:46:26,310][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:46:26,810][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:46:27,308][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:46:27,808][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:46:28,308][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:46:28,811][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:46:29,311][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:46:29,809][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10156 tokens. +[2026-03-26 02:46:30,538][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.07%, ΔTime: 00:00:32 +[2026-03-26 02:46:31,292][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:46:31,294][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:46:31,296][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:46:32,009][__main__][INFO] - Iteration 323 took 53s (32.75% Gen, 65.90% Train). Generation: 17s, Training: 34s. Estimated remaining time: 39h 27m 25s. Estimated total time: 44h 13m 9s. Time estimates for 10 more iterations: 8m 50s, 100 more iterations: 1h 28m 26s, 500 more iterations: 7h 22m 11s. +[2026-03-26 02:46:32,011][__main__][INFO] - Starting iteration 323. +[2026-03-26 02:46:32,425][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 32 and human policies 1. +[2026-03-26 02:46:32,426][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:46:48,942][__main__][INFO] - Number of regex retries in iteration 323: 0 +[2026-03-26 02:46:48,942][__main__][INFO] - agents played in iteration 323 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:46:49,719][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:46:49,742][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:46:49,765][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:46:49,788][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:46:49,789][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:46:49,789][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:46:50,478][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:46:50,934][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:46:51,434][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:46:51,931][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:46:52,428][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:46:52,926][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:46:53,422][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:46:53,924][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:46:54,420][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:46:54,920][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:46:55,424][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:46:55,924][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:46:56,425][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:46:56,930][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:46:57,430][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:46:57,933][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:46:58,438][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:46:58,939][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:46:59,442][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:46:59,942][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:47:00,448][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:47:00,952][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:47:01,455][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:47:01,959][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:47:02,457][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:47:02,954][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:47:03,452][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:47:03,949][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:47:04,450][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:47:04,951][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:47:05,452][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:47:05,951][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:47:06,448][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:47:06,949][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:47:07,453][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:47:07,951][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:47:08,450][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:47:08,951][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:47:09,453][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:47:09,955][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:47:10,453][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:47:10,949][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:47:11,450][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:47:11,953][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:47:12,451][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:47:12,952][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:47:13,449][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:47:13,946][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:47:14,443][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:47:14,942][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:47:15,443][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:47:15,939][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:47:16,439][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:47:16,938][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:47:17,435][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:47:17,933][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:47:18,433][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:47:18,932][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:47:19,434][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:47:19,939][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:47:20,443][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:47:20,946][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:47:21,446][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:47:21,947][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:47:22,449][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10261 tokens. +[2026-03-26 02:47:23,176][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.61%, Block Peak % of device VRAM: 62.29%, ΔTime: 00:00:32 +[2026-03-26 02:47:23,925][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:47:23,927][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:47:23,929][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:47:24,647][__main__][INFO] - Iteration 324 took 52s (31.63% Gen, 66.99% Train). Generation: 16s, Training: 34s. Estimated remaining time: 38h 44m 30s. Estimated total time: 43h 31m 6s. Time estimates for 10 more iterations: 8m 42s, 100 more iterations: 1h 27m 2s, 500 more iterations: 7h 15m 11s. +[2026-03-26 02:47:24,649][__main__][INFO] - Starting iteration 324. +[2026-03-26 02:47:25,052][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 32 and human policies 1. +[2026-03-26 02:47:25,052][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:47:43,528][__main__][INFO] - Number of regex retries in iteration 324: 0 +[2026-03-26 02:47:43,529][__main__][INFO] - agents played in iteration 324 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:47:44,295][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:47:44,318][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:47:44,342][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:47:44,365][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:47:44,365][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:47:44,366][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:47:45,058][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:47:45,511][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:47:46,015][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:47:46,515][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:47:47,012][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:47:47,510][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:47:48,011][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:47:48,510][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:47:49,008][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:47:49,505][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:47:50,002][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:47:50,504][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:47:51,004][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:47:51,500][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:47:51,997][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:47:52,495][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:47:52,991][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:47:53,489][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:47:53,986][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:47:54,483][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:47:54,980][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:47:55,477][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:47:55,979][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:47:56,478][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:47:56,974][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:47:57,472][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:47:57,971][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:47:58,475][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:47:58,980][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:47:59,481][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:47:59,979][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:48:00,480][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:48:00,981][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:48:01,483][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:48:01,986][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:48:02,487][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:48:02,990][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:48:03,496][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:48:04,014][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:48:04,513][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:48:05,012][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:48:05,512][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:48:06,008][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:48:06,509][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:48:07,005][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:48:07,502][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:48:07,999][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:48:08,496][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:48:08,997][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:48:09,494][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:48:09,996][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:48:10,494][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:48:10,996][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:48:11,497][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:48:12,001][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:48:12,503][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:48:13,001][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:48:13,499][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:48:13,999][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:48:14,497][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:48:15,003][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:48:15,500][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:48:15,999][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:48:16,498][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:48:17,999][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10137 tokens. +[2026-03-26 02:48:17,728][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.23%, Current % of VRAM taken: 60.66%, Block Peak % of device VRAM: 62.21%, ΔTime: 00:00:32 +[2026-03-26 02:48:18,486][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:48:18,489][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:48:18,490][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:48:19,194][__main__][INFO] - Iteration 325 took 54s (34.12% Gen, 64.57% Train). Generation: 18s, Training: 34s. Estimated remaining time: 40h 19m 39s. Estimated total time: 45h 7m 10s. Time estimates for 10 more iterations: 9m 1s, 100 more iterations: 1h 30m 14s, 500 more iterations: 7h 31m 11s. +[2026-03-26 02:48:19,197][__main__][INFO] - Starting iteration 325. +[2026-03-26 02:48:19,596][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 32 and human policies 1. +[2026-03-26 02:48:19,596][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:48:38,448][__main__][INFO] - Number of regex retries in iteration 325: 0 +[2026-03-26 02:48:38,449][__main__][INFO] - agents played in iteration 325 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:48:39,220][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:48:39,243][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:48:39,267][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:48:39,289][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:48:39,290][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:48:39,291][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:48:39,987][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:48:40,443][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:48:40,943][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:48:41,439][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:48:41,935][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:48:42,432][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:48:42,928][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:48:43,423][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:48:43,920][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:48:44,418][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:48:44,917][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:48:45,415][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:48:45,913][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:48:46,412][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:48:46,909][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:48:47,408][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:48:47,909][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:48:48,421][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:48:48,925][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:48:49,429][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:48:49,952][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:48:50,451][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:48:50,952][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:48:51,456][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:48:51,959][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:48:52,457][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:48:52,958][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:48:53,460][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:48:53,958][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:48:54,455][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:48:54,957][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:48:55,458][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:48:55,955][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:48:56,453][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:48:56,951][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:48:57,449][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:48:57,950][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:48:58,449][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:48:58,949][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:48:59,451][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:48:59,949][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:49:00,448][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:49:00,947][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:49:01,444][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:49:01,942][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:49:02,439][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:49:02,937][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:49:03,440][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:49:03,941][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:49:04,448][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:49:04,948][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:49:05,447][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:49:05,944][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:49:06,441][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:49:06,938][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:49:07,436][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:49:07,934][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:49:08,436][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:49:08,937][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:49:09,437][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:49:09,937][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:49:10,437][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:49:10,937][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:49:11,434][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:49:11,935][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10170 tokens. +[2026-03-26 02:49:12,662][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.23%, Current % of VRAM taken: 60.65%, Block Peak % of device VRAM: 62.23%, ΔTime: 00:00:32 +[2026-03-26 02:49:13,544][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:49:13,546][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:49:13,548][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:49:14,376][__main__][INFO] - Iteration 326 took 54s (34.42% Gen, 64.07% Train). Generation: 18s, Training: 35s. Estimated remaining time: 40h 50m 36s. Estimated total time: 45h 39m 2s. Time estimates for 10 more iterations: 9m 7s, 100 more iterations: 1h 31m 18s, 500 more iterations: 7h 36m 30s. +[2026-03-26 02:49:14,378][__main__][INFO] - Starting iteration 326. +[2026-03-26 02:49:14,787][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 32 and human policies 1. +[2026-03-26 02:49:14,788][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:49:33,698][__main__][INFO] - Number of regex retries in iteration 326: 0 +[2026-03-26 02:49:33,699][__main__][INFO] - agents played in iteration 326 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:49:34,520][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:49:34,543][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:49:34,566][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:49:34,589][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:49:34,590][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:49:34,590][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:49:35,283][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:49:35,736][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:49:36,239][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:49:36,737][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:49:37,233][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:49:37,731][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:49:38,231][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:49:38,728][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:49:39,226][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:49:39,729][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:49:40,227][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:49:40,728][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:49:41,226][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:49:41,729][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:49:42,229][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:49:42,729][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:49:43,227][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:49:43,727][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:49:44,227][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:49:44,725][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:49:45,222][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:49:45,721][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:49:46,218][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:49:46,716][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:49:47,214][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:49:47,715][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:49:48,215][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:49:48,719][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:49:49,236][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:49:49,736][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:49:50,238][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:49:50,744][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:49:51,252][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:49:51,755][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:49:52,252][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:49:52,752][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:49:53,253][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:49:53,754][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:49:54,257][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:49:54,761][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:49:55,268][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:49:55,772][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:49:56,271][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:49:56,771][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:49:57,274][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:49:57,772][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:49:58,271][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:49:58,771][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:49:59,273][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:49:59,770][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:50:00,266][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:50:00,766][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:50:01,263][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:50:01,759][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:50:02,259][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:50:02,756][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:50:03,257][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:50:03,754][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:50:04,251][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:50:04,747][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:50:05,243][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:50:05,739][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:50:06,236][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:50:06,734][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:50:07,231][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10206 tokens. +[2026-03-26 02:50:07,953][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.33%, ΔTime: 00:00:32 +[2026-03-26 02:50:08,755][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:50:08,758][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:50:08,760][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:50:09,453][__main__][INFO] - Iteration 327 took 54s (34.59% Gen, 64.14% Train). Generation: 18s, Training: 35s. Estimated remaining time: 40h 43m 59s. Estimated total time: 45h 33m 20s. Time estimates for 10 more iterations: 9m 6s, 100 more iterations: 1h 31m 6s, 500 more iterations: 7h 35m 33s. +[2026-03-26 02:50:09,456][__main__][INFO] - Starting iteration 327. +[2026-03-26 02:50:09,858][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 32 and human policies 1. +[2026-03-26 02:50:09,859][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:50:26,333][__main__][INFO] - Number of regex retries in iteration 327: 0 +[2026-03-26 02:50:26,333][__main__][INFO] - agents played in iteration 327 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:50:27,107][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:50:27,130][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:50:27,153][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:50:27,176][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:50:27,177][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:50:27,177][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:50:27,875][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:50:28,328][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:50:28,831][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:50:29,330][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:50:29,827][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:50:30,330][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:50:30,828][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:50:31,326][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:50:31,825][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:50:32,322][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:50:32,821][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:50:33,318][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:50:33,815][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:50:34,312][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:50:34,809][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:50:35,306][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:50:35,801][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:50:36,302][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:50:36,805][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:50:37,306][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:50:37,807][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:50:38,308][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:50:38,809][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:50:39,309][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:50:39,808][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:50:40,307][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:50:40,805][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:50:41,302][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:50:41,802][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:50:42,300][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:50:42,804][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:50:43,310][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:50:43,814][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:50:44,322][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:50:44,826][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:50:45,328][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:50:45,830][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:50:46,330][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:50:46,832][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:50:47,334][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:50:47,836][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:50:48,337][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:50:48,838][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:50:49,339][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:50:49,842][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:50:50,342][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:50:50,840][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:50:51,337][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:50:51,838][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:50:52,334][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:50:52,830][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:50:53,328][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:50:53,830][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:50:54,328][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:50:54,829][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:50:55,330][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:50:55,828][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:50:56,330][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:50:56,830][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:50:57,332][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:50:57,831][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:50:58,331][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:50:58,832][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:50:59,333][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:50:59,835][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10286 tokens. +[2026-03-26 02:51:00,569][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.61%, Block Peak % of device VRAM: 62.18%, ΔTime: 00:00:32 +[2026-03-26 02:51:01,321][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:51:01,324][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:51:01,325][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:51:01,959][__main__][INFO] - Iteration 328 took 52s (31.62% Gen, 67.16% Train). Generation: 16s, Training: 34s. Estimated remaining time: 38h 34m 53s. Estimated total time: 43h 25m 6s. Time estimates for 10 more iterations: 8m 41s, 100 more iterations: 1h 26m 50s, 500 more iterations: 7h 14m 11s. +[2026-03-26 02:51:01,961][__main__][INFO] - Starting iteration 328. +[2026-03-26 02:51:02,369][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 32 and human policies 1. +[2026-03-26 02:51:02,370][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:51:20,032][__main__][INFO] - Number of regex retries in iteration 328: 0 +[2026-03-26 02:51:20,033][__main__][INFO] - agents played in iteration 328 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:51:20,802][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:51:20,825][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:51:20,848][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:51:20,871][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:51:20,871][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:51:20,872][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:51:21,582][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:51:22,035][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:51:22,538][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:51:23,034][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:51:23,534][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:51:24,031][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:51:24,534][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:51:25,031][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:51:25,528][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:51:26,025][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:51:26,520][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:51:27,016][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:51:27,514][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:51:28,011][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:51:28,508][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:51:29,006][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:51:29,503][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:51:30,000][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:51:30,499][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:51:30,997][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:51:31,494][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:51:31,992][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:51:32,490][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:51:32,990][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:51:33,487][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:51:34,006][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:51:34,512][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:51:35,020][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:51:35,522][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:51:36,020][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:51:36,517][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:51:37,022][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:51:37,533][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:51:38,036][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:51:38,536][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:51:39,038][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:51:39,539][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:51:40,038][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:51:40,542][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:51:41,045][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:51:41,548][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:51:42,052][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:51:42,552][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:51:43,052][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:51:43,550][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:51:44,049][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:51:44,550][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:51:45,050][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:51:45,548][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:51:46,044][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:51:46,542][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:51:47,039][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:51:47,536][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:51:48,034][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:51:48,532][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:51:49,030][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:51:49,528][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:51:50,031][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:51:50,532][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:51:51,031][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:51:51,531][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:51:52,030][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:51:52,530][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:51:53,029][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:51:53,529][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10247 tokens. +[2026-03-26 02:51:54,264][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.25%, Current % of VRAM taken: 60.68%, Block Peak % of device VRAM: 62.33%, ΔTime: 00:00:32 +[2026-03-26 02:51:55,012][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:51:55,014][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:51:55,016][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:51:55,942][__main__][INFO] - Iteration 329 took 53s (32.97% Gen, 65.30% Train). Generation: 17s, Training: 34s. Estimated remaining time: 39h 47m 33s. Estimated total time: 44h 38m 40s. Time estimates for 10 more iterations: 8m 55s, 100 more iterations: 1h 29m 17s, 500 more iterations: 7h 26m 26s. +[2026-03-26 02:51:55,944][__main__][INFO] - Starting iteration 329. +[2026-03-26 02:51:56,348][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 32 and human policies 1. +[2026-03-26 02:51:56,349][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:52:04,592][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:52:13,376][__main__][INFO] - Number of regex retries in iteration 329: 1 +[2026-03-26 02:52:13,377][__main__][INFO] - agents played in iteration 329 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:52:14,145][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:52:14,169][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:52:14,192][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:52:14,215][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:52:14,215][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:52:14,216][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:52:14,909][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:52:15,362][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:52:15,862][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:52:16,359][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:52:16,854][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:52:17,352][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:52:17,853][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:52:18,353][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:52:18,854][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:52:19,352][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:52:19,862][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:52:20,360][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:52:20,861][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:52:21,362][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:52:21,860][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:52:22,356][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:52:22,857][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:52:23,354][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:52:23,850][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:52:24,350][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:52:24,848][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:52:25,348][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:52:25,849][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:52:26,351][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:52:26,881][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:52:27,386][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:52:27,882][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:52:28,378][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:52:28,875][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:52:29,374][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:52:29,878][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:52:30,378][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:52:30,880][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:52:31,382][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:52:31,879][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:52:32,378][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:52:32,881][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:52:33,379][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:52:33,878][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:52:34,383][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:52:34,884][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:52:35,383][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:52:35,882][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:52:36,381][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:52:36,881][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:52:37,383][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:52:37,880][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:52:38,377][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:52:38,874][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:52:39,374][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:52:39,870][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:52:40,367][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:52:40,868][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:52:41,365][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:52:41,862][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:52:42,361][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:52:42,857][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:52:43,354][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:52:43,853][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:52:44,351][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:52:44,849][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:52:45,348][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:52:45,847][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:52:46,344][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:52:46,841][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10175 tokens. +[2026-03-26 02:52:47,565][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.09%, ΔTime: 00:00:32 +[2026-03-26 02:52:48,347][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:52:48,350][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:52:48,352][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:52:49,189][__main__][INFO] - Iteration 330 took 52s (32.22% Gen, 66.19% Train). Generation: 17s, Training: 34s. Estimated remaining time: 39h 10m 5s. Estimated total time: 44h 2m 6s. Time estimates for 10 more iterations: 8m 48s, 100 more iterations: 1h 28m 4s, 500 more iterations: 7h 20m 21s. +[2026-03-26 02:52:49,192][__main__][INFO] - Starting iteration 330. +[2026-03-26 02:52:49,593][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 32 and human policies 1. +[2026-03-26 02:52:49,594][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:53:07,869][__main__][INFO] - Number of regex retries in iteration 330: 0 +[2026-03-26 02:53:07,870][__main__][INFO] - agents played in iteration 330 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:53:08,640][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:53:08,666][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:53:08,689][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:53:08,712][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:53:08,712][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:53:08,713][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:53:09,404][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:53:09,861][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:53:10,364][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:53:10,860][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:53:11,356][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:53:11,857][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:53:12,353][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:53:12,850][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:53:13,350][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:53:13,849][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:53:14,351][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:53:14,858][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:53:15,356][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:53:15,856][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:53:16,359][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:53:16,862][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:53:17,367][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:53:17,870][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:53:18,370][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:53:18,870][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:53:19,369][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:53:19,871][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:53:20,372][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:53:20,874][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:53:21,373][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:53:21,871][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:53:22,372][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:53:22,872][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:53:23,372][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:53:23,872][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:53:24,379][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:53:24,880][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:53:25,384][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:53:25,884][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:53:26,384][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:53:26,883][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:53:27,383][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:53:27,885][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:53:28,384][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:53:28,882][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:53:29,385][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:53:29,882][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:53:30,379][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:53:30,879][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:53:31,375][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:53:31,871][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:53:32,367][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:53:32,868][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:53:33,368][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:53:33,864][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:53:34,366][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:53:34,863][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:53:35,361][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:53:35,858][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:53:36,355][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:53:36,855][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:53:37,352][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:53:37,853][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:53:38,352][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:53:38,850][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:53:39,353][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:53:39,854][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:53:40,352][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:53:40,852][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:53:41,349][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10331 tokens. +[2026-03-26 02:53:42,093][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.34%, ΔTime: 00:00:32 +[2026-03-26 02:53:42,848][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:53:42,850][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:53:42,852][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:53:44,249][__main__][INFO] - Iteration 331 took 54s (33.44% Gen, 64.00% Train). Generation: 18s, Training: 34s. Estimated remaining time: 40h 39m 54s. Estimated total time: 45h 32m 50s. Time estimates for 10 more iterations: 9m 6s, 100 more iterations: 1h 31m 5s, 500 more iterations: 7h 35m 28s. +[2026-03-26 02:53:44,252][__main__][INFO] - Starting iteration 331. +[2026-03-26 02:53:44,654][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 33 and human policies 1. +[2026-03-26 02:53:44,655][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:54:02,206][__main__][INFO] - Number of regex retries in iteration 331: 0 +[2026-03-26 02:54:02,206][__main__][INFO] - agents played in iteration 331 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:54:02,976][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:54:02,999][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:54:03,022][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:54:03,045][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:54:03,046][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:54:03,046][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:54:03,751][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:54:04,209][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:54:04,713][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:54:05,214][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:54:05,718][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:54:06,216][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:54:06,718][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:54:07,219][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:54:07,720][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:54:08,226][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:54:08,731][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:54:09,238][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:54:09,737][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:54:10,235][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:54:10,736][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:54:11,236][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:54:11,733][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:54:12,231][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:54:12,730][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:54:13,230][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:54:13,726][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:54:14,226][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:54:14,728][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:54:15,234][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:54:15,736][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:54:16,238][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:54:16,737][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:54:17,238][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:54:17,738][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:54:18,239][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:54:18,740][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:54:19,243][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:54:19,743][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:54:20,245][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:54:20,748][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:54:21,248][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:54:21,748][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:54:22,250][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:54:22,751][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:54:23,248][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:54:23,750][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:54:24,246][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:54:24,744][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:54:25,246][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:54:25,747][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:54:26,244][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:54:26,745][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:54:27,243][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:54:27,744][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:54:28,241][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:54:28,742][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:54:29,243][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:54:29,740][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:54:30,241][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:54:30,741][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:54:31,244][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:54:31,744][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:54:32,242][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:54:32,739][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:54:33,237][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:54:33,736][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:54:34,236][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:54:34,734][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:54:35,231][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:54:35,728][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10298 tokens. +[2026-03-26 02:54:36,462][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.23%, ΔTime: 00:00:32 +[2026-03-26 02:54:37,258][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:54:37,261][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:54:37,263][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:54:38,099][__main__][INFO] - Iteration 332 took 53s (32.84% Gen, 65.59% Train). Generation: 17s, Training: 35s. Estimated remaining time: 39h 38m 26s. Estimated total time: 44h 32m 15s. Time estimates for 10 more iterations: 8m 54s, 100 more iterations: 1h 29m 4s, 500 more iterations: 7h 25m 22s. +[2026-03-26 02:54:38,102][__main__][INFO] - Starting iteration 332. +[2026-03-26 02:54:38,503][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 33 and human policies 1. +[2026-03-26 02:54:38,504][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:54:43,946][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:54:54,857][__main__][INFO] - Number of regex retries in iteration 332: 1 +[2026-03-26 02:54:54,858][__main__][INFO] - agents played in iteration 332 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:54:55,628][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:54:55,652][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:54:55,675][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:54:55,698][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:54:55,698][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:54:55,699][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:54:56,394][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:54:56,846][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:54:57,348][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:54:57,847][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:54:58,350][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:54:58,857][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:54:59,354][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:54:59,851][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:55:00,348][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:55:00,850][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:55:01,347][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:55:01,850][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:55:02,347][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:55:02,845][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:55:03,342][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:55:03,838][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:55:04,340][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:55:04,836][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:55:05,336][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:55:05,832][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:55:06,328][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:55:06,829][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:55:07,331][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:55:07,830][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:55:08,328][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:55:08,827][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:55:09,327][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:55:09,826][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:55:10,323][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:55:10,823][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:55:11,324][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:55:11,825][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:55:12,326][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:55:12,827][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:55:13,327][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:55:13,825][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:55:14,330][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:55:14,829][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:55:15,325][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:55:15,823][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:55:16,321][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:55:16,819][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:55:17,340][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:55:17,840][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:55:18,340][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:55:18,843][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:55:19,343][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:55:19,841][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:55:20,343][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:55:20,840][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:55:21,337][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:55:21,834][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:55:22,332][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:55:22,829][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:55:23,325][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:55:23,829][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:55:24,327][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:55:24,825][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:55:25,324][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:55:25,826][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:55:26,326][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:55:26,825][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:55:27,324][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:55:27,821][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:55:28,319][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10250 tokens. +[2026-03-26 02:55:29,068][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.15%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.19%, ΔTime: 00:00:32 +[2026-03-26 02:55:29,812][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:55:29,814][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:55:29,816][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:55:30,646][__main__][INFO] - Iteration 333 took 52s (31.36% Gen, 67.04% Train). Generation: 16s, Training: 34s. Estimated remaining time: 38h 32m 30s. Estimated total time: 43h 27m 12s. Time estimates for 10 more iterations: 8m 41s, 100 more iterations: 1h 26m 54s, 500 more iterations: 7h 14m 32s. +[2026-03-26 02:55:30,648][__main__][INFO] - Starting iteration 333. +[2026-03-26 02:55:31,047][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 33 and human policies 1. +[2026-03-26 02:55:31,048][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:55:47,959][__main__][INFO] - Number of regex retries in iteration 333: 0 +[2026-03-26 02:55:47,960][__main__][INFO] - agents played in iteration 333 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:55:48,728][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:55:48,751][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:55:48,774][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:55:48,797][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:55:48,798][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:55:48,798][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:55:49,508][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:55:49,967][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:55:50,484][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:55:50,992][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:55:51,498][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:55:52,002][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:55:52,504][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:55:53,008][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:55:53,510][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:55:54,010][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:55:54,509][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:55:55,010][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:55:55,509][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:55:56,008][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:55:56,508][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:55:57,009][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:55:57,510][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:55:58,010][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:55:58,509][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:55:59,007][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:55:59,511][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:56:00,011][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:56:00,514][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:56:01,017][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:56:01,518][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:56:02,021][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:56:02,527][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:56:03,032][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:56:03,532][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:56:04,035][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:56:04,539][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:56:05,043][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:56:05,544][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:56:06,050][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:56:06,550][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:56:07,053][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:56:07,555][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:56:08,054][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:56:08,553][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:56:09,052][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:56:09,555][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:56:10,057][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:56:10,557][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:56:11,055][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:56:11,556][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:56:12,057][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:56:12,558][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:56:13,059][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:56:13,560][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:56:14,061][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:56:14,560][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:56:15,058][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:56:15,555][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:56:16,057][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:56:16,557][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:56:17,055][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:56:17,553][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:56:18,052][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:56:18,554][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:56:19,054][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:56:19,556][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:56:20,054][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:56:20,557][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:56:21,053][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:56:21,550][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10305 tokens. +[2026-03-26 02:56:22,291][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.39%, ΔTime: 00:00:32 +[2026-03-26 02:56:23,052][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:56:23,054][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:56:23,056][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:56:23,877][__main__][INFO] - Iteration 334 took 52s (32.01% Gen, 66.43% Train). Generation: 16s, Training: 35s. Estimated remaining time: 39h 5m 55s. Estimated total time: 44h 1m 31s. Time estimates for 10 more iterations: 8m 48s, 100 more iterations: 1h 28m 3s, 500 more iterations: 7h 20m 15s. +[2026-03-26 02:56:23,879][__main__][INFO] - Starting iteration 334. +[2026-03-26 02:56:24,278][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 33 and human policies 1. +[2026-03-26 02:56:24,279][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:56:41,399][__main__][INFO] - Number of regex retries in iteration 334: 0 +[2026-03-26 02:56:41,400][__main__][INFO] - agents played in iteration 334 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:56:42,167][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:56:42,191][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:56:42,214][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:56:42,236][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:56:42,237][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:56:42,238][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:56:42,932][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:56:43,383][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:56:43,887][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:56:44,386][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:56:44,884][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:56:45,381][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:56:45,884][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:56:46,381][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:56:46,879][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:56:47,377][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:56:47,884][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:56:48,387][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:56:48,884][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:56:49,383][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:56:49,880][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:56:50,385][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:56:50,885][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:56:51,398][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:56:51,901][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:56:52,401][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:56:52,903][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:56:53,402][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:56:53,902][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:56:54,402][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:56:54,901][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:56:55,405][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:56:55,908][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:56:56,434][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:56:56,937][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:56:57,438][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:56:57,938][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:56:58,439][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:56:58,942][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:56:59,440][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:56:59,939][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:57:00,438][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:57:00,939][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:57:01,441][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:57:01,941][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:57:02,440][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:57:02,941][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:57:03,440][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:57:03,940][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:57:04,440][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:57:04,940][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:57:05,442][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:57:05,943][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:57:06,445][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:57:06,950][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:57:07,451][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:57:07,951][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:57:08,449][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:57:08,947][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:57:09,444][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:57:09,942][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:57:10,442][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:57:10,940][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:57:11,438][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:57:11,936][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:57:12,436][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:57:12,936][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:57:13,434][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:57:13,932][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:57:14,434][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:57:14,932][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10252 tokens. +[2026-03-26 02:57:15,670][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.15%, Current % of VRAM taken: 60.58%, Block Peak % of device VRAM: 62.18%, ΔTime: 00:00:32 +[2026-03-26 02:57:16,424][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:57:16,426][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:57:16,428][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:57:17,128][__main__][INFO] - Iteration 335 took 52s (32.40% Gen, 66.28% Train). Generation: 17s, Training: 35s. Estimated remaining time: 39h 6m 4s. Estimated total time: 44h 2m 32s. Time estimates for 10 more iterations: 8m 48s, 100 more iterations: 1h 28m 5s, 500 more iterations: 7h 20m 25s. +[2026-03-26 02:57:17,130][__main__][INFO] - Starting iteration 335. +[2026-03-26 02:57:17,529][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 33 and human policies 1. +[2026-03-26 02:57:17,529][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:57:27,028][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 02:57:38,131][__main__][INFO] - Number of regex retries in iteration 335: 1 +[2026-03-26 02:57:38,132][__main__][INFO] - agents played in iteration 335 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:57:38,897][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:57:38,920][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:57:38,943][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:57:38,966][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:57:38,966][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:57:38,967][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:57:39,660][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:57:40,115][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:57:40,613][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:57:41,110][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:57:41,609][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:57:42,107][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:57:42,604][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:57:43,103][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:57:43,602][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:57:44,104][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:57:44,601][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:57:45,106][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:57:45,610][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:57:46,110][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:57:46,613][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:57:47,116][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:57:47,624][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:57:48,126][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:57:48,625][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:57:49,129][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:57:49,627][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:57:50,129][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:57:50,628][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:57:51,129][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:57:51,628][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:57:52,128][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:57:52,627][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:57:53,125][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:57:53,622][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:57:54,123][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:57:54,622][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:57:55,120][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:57:55,620][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:57:56,118][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:57:56,616][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:57:57,114][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:57:57,612][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:57:58,111][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:57:58,610][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:57:59,107][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:57:59,605][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:58:00,104][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:58:00,603][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:58:01,101][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:58:01,604][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:58:02,107][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:58:02,631][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:58:03,134][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:58:03,637][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:58:04,136][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:58:04,632][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:58:05,129][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:58:05,625][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:58:06,121][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:58:06,618][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:58:07,116][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:58:07,612][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:58:08,109][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:58:08,604][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:58:09,100][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:58:09,597][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:58:10,093][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:58:10,590][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:58:11,086][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:58:11,581][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10214 tokens. +[2026-03-26 02:58:12,318][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 02:58:13,066][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:58:13,069][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:58:13,070][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:58:13,715][__main__][INFO] - Iteration 336 took 56s (36.67% Gen, 62.18% Train). Generation: 20s, Training: 34s. Estimated remaining time: 41h 51m 55s. Estimated total time: 46h 49m 20s. Time estimates for 10 more iterations: 9m 21s, 100 more iterations: 1h 33m 38s, 500 more iterations: 7h 48m 13s. +[2026-03-26 02:58:13,717][__main__][INFO] - Starting iteration 336. +[2026-03-26 02:58:14,116][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 33 and human policies 1. +[2026-03-26 02:58:14,117][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:58:28,302][__main__][INFO] - Number of regex retries in iteration 336: 0 +[2026-03-26 02:58:28,303][__main__][INFO] - agents played in iteration 336 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:58:29,065][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:58:29,089][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:58:29,112][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:58:29,135][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:58:29,135][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:58:29,136][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:58:29,858][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:58:30,310][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:58:30,812][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:58:31,310][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:58:31,811][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:58:32,310][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:58:32,812][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:58:33,309][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:58:33,806][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:58:34,303][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:58:34,801][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:58:35,299][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:58:35,798][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:58:36,303][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:58:36,806][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:58:37,308][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:58:37,806][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:58:38,309][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:58:38,809][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:58:39,309][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:58:39,809][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:58:40,315][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:58:40,823][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:58:41,329][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:58:41,829][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:58:42,333][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:58:42,831][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:58:43,333][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:58:43,832][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:58:44,332][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:58:44,833][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:58:45,334][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:58:45,831][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:58:46,327][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:58:46,827][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:58:47,328][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:58:47,860][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:58:48,359][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:58:48,858][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:58:49,358][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:58:49,857][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:58:50,360][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:58:50,858][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:58:51,356][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:58:51,858][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:58:52,358][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:58:52,859][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:58:53,359][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:58:53,857][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:58:54,360][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:58:54,864][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:58:55,365][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:58:55,869][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:58:56,369][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:58:56,869][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:58:57,370][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:58:57,870][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:58:58,372][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:58:58,870][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:58:59,371][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:58:59,872][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:59:00,373][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:59:00,869][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:59:01,372][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:59:01,872][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10310 tokens. +[2026-03-26 02:59:02,613][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.22%, Current % of VRAM taken: 60.65%, Block Peak % of device VRAM: 62.17%, ΔTime: 00:00:32 +[2026-03-26 02:59:03,347][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:59:03,350][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:59:03,351][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:59:03,997][__main__][INFO] - Iteration 337 took 49s (28.44% Gen, 70.26% Train). Generation: 14s, Training: 35s. Estimated remaining time: 36h 35m 49s. Estimated total time: 41h 34m 4s. Time estimates for 10 more iterations: 8m 18s, 100 more iterations: 1h 23m 8s, 500 more iterations: 6h 55m 40s. +[2026-03-26 02:59:03,999][__main__][INFO] - Starting iteration 337. +[2026-03-26 02:59:04,398][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 33 and human policies 1. +[2026-03-26 02:59:04,399][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 02:59:19,805][__main__][INFO] - Number of regex retries in iteration 337: 0 +[2026-03-26 02:59:19,806][__main__][INFO] - agents played in iteration 337 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 02:59:20,581][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:59:20,604][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:59:20,627][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:59:20,650][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 02:59:20,651][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 02:59:20,651][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 02:59:21,375][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 02:59:21,828][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 02:59:22,331][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 02:59:22,828][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 02:59:23,328][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 02:59:23,826][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 02:59:24,324][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 02:59:24,821][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 02:59:25,320][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 02:59:25,817][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 02:59:26,320][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 02:59:26,817][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 02:59:27,319][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 02:59:27,822][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 02:59:28,323][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 02:59:28,826][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 02:59:29,323][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 02:59:29,823][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 02:59:30,324][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 02:59:30,828][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 02:59:31,331][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 02:59:31,836][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 02:59:32,340][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 02:59:32,842][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 02:59:33,343][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 02:59:33,840][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 02:59:34,338][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 02:59:34,833][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 02:59:35,332][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 02:59:35,829][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 02:59:36,328][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 02:59:36,824][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 02:59:37,321][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 02:59:37,818][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 02:59:38,316][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 02:59:38,814][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 02:59:39,311][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 02:59:39,809][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 02:59:40,305][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 02:59:40,802][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 02:59:41,300][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 02:59:41,798][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 02:59:42,298][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 02:59:42,795][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 02:59:43,299][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 02:59:43,801][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 02:59:44,300][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 02:59:44,798][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 02:59:45,296][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 02:59:45,795][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 02:59:46,295][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 02:59:46,791][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 02:59:47,287][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 02:59:47,784][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 02:59:48,281][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 02:59:48,777][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 02:59:49,272][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 02:59:49,769][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 02:59:50,269][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 02:59:50,766][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 02:59:51,263][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 02:59:51,760][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 02:59:52,257][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 02:59:52,754][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 02:59:53,253][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10003 tokens. +[2026-03-26 02:59:54,001][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.05%, Current % of VRAM taken: 60.48%, Block Peak % of device VRAM: 62.05%, ΔTime: 00:00:32 +[2026-03-26 02:59:54,758][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 02:59:54,761][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 02:59:54,763][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 02:59:55,405][__main__][INFO] - Iteration 338 took 51s (30.20% Gen, 68.53% Train). Generation: 15s, Training: 34s. Estimated remaining time: 37h 31m 14s. Estimated total time: 42h 30m 21s. Time estimates for 10 more iterations: 8m 30s, 100 more iterations: 1h 25m 0s, 500 more iterations: 7h 5m 3s. +[2026-03-26 02:59:55,407][__main__][INFO] - Starting iteration 338. +[2026-03-26 02:59:55,808][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 33 and human policies 1. +[2026-03-26 02:59:55,809][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:00:11,669][__main__][INFO] - Number of regex retries in iteration 338: 0 +[2026-03-26 03:00:11,670][__main__][INFO] - agents played in iteration 338 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:00:12,435][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:00:12,458][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:00:12,481][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:00:12,505][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:00:12,505][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:00:12,506][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:00:13,220][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:00:13,673][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:00:14,174][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:00:14,672][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:00:15,171][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:00:15,669][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:00:16,166][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:00:16,662][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:00:17,159][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:00:17,654][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:00:18,151][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:00:18,647][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:00:19,149][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:00:19,648][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:00:20,151][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:00:20,649][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:00:21,147][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:00:21,645][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:00:22,144][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:00:22,646][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:00:23,149][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:00:23,655][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:00:24,156][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:00:24,654][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:00:25,151][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:00:25,652][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:00:26,151][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:00:26,650][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:00:27,168][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:00:27,671][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:00:28,170][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:00:28,669][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:00:29,170][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:00:29,669][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:00:30,167][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:00:30,668][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:00:31,166][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:00:31,664][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:00:32,162][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:00:32,659][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:00:33,157][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:00:33,655][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:00:34,151][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:00:34,651][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:00:35,152][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:00:35,650][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:00:36,147][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:00:36,646][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:00:37,142][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:00:37,640][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:00:38,139][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:00:38,638][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:00:39,135][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:00:39,631][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:00:40,127][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:00:40,621][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:00:41,116][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:00:41,616][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:00:42,114][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:00:42,617][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:00:43,115][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:00:43,615][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:00:44,116][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:00:44,615][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:00:45,119][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10069 tokens. +[2026-03-26 03:00:45,869][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.29%, Current % of VRAM taken: 60.71%, Block Peak % of device VRAM: 62.28%, ΔTime: 00:00:32 +[2026-03-26 03:00:46,619][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:00:46,622][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:00:46,624][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:00:47,375][__main__][INFO] - Iteration 339 took 51s (30.76% Gen, 67.78% Train). Generation: 15s, Training: 34s. Estimated remaining time: 37h 58m 23s. Estimated total time: 42h 58m 22s. Time estimates for 10 more iterations: 8m 35s, 100 more iterations: 1h 25m 56s, 500 more iterations: 7h 9m 43s. +[2026-03-26 03:00:47,377][__main__][INFO] - Starting iteration 339. +[2026-03-26 03:00:47,778][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 33 and human policies 1. +[2026-03-26 03:00:47,779][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:01:04,228][__main__][INFO] - Number of regex retries in iteration 339: 0 +[2026-03-26 03:01:04,229][__main__][INFO] - agents played in iteration 339 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:01:04,997][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:01:05,021][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:01:05,043][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:01:05,066][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.58%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:01:05,067][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:01:05,068][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:01:05,779][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:01:06,235][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:01:06,737][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:01:07,234][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:01:07,731][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:01:08,228][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:01:08,724][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:01:09,222][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:01:09,719][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:01:10,219][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:01:10,718][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:01:11,221][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:01:11,724][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:01:12,228][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:01:12,739][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:01:13,248][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:01:13,752][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:01:14,264][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:01:14,764][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:01:15,267][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:01:15,769][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:01:16,273][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:01:16,774][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:01:17,279][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:01:17,783][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:01:18,282][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:01:18,782][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:01:19,279][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:01:19,777][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:01:20,277][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:01:20,778][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:01:21,275][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:01:21,776][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:01:22,273][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:01:22,771][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:01:23,270][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:01:23,767][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:01:24,264][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:01:24,762][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:01:25,261][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:01:25,758][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:01:26,258][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:01:26,758][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:01:27,256][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:01:27,758][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:01:28,259][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:01:28,758][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:01:29,257][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:01:29,757][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:01:30,257][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:01:30,757][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:01:31,258][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:01:31,756][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:01:32,253][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:01:32,751][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:01:33,252][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:01:33,749][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:01:34,246][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:01:34,749][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:01:35,247][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:01:35,747][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:01:36,246][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:01:36,746][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:01:37,243][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:01:37,740][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10202 tokens. +[2026-03-26 03:01:38,479][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.16%, Current % of VRAM taken: 60.58%, Block Peak % of device VRAM: 62.20%, ΔTime: 00:00:32 +[2026-03-26 03:01:39,238][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:01:39,240][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:01:39,242][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:01:40,008][__main__][INFO] - Iteration 340 took 52s (31.50% Gen, 67.03% Train). Generation: 16s, Training: 35s. Estimated remaining time: 38h 30m 41s. Estimated total time: 43h 31m 32s. Time estimates for 10 more iterations: 8m 42s, 100 more iterations: 1h 27m 3s, 500 more iterations: 7h 15m 15s. +[2026-03-26 03:01:40,010][__main__][INFO] - Starting iteration 340. +[2026-03-26 03:01:40,412][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 33 and human policies 1. +[2026-03-26 03:01:40,412][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:01:45,632][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 03:01:57,767][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 03:01:58,563][__main__][INFO] - Number of regex retries in iteration 340: 2 +[2026-03-26 03:01:58,564][__main__][INFO] - agents played in iteration 340 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:01:59,334][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:01:59,357][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:01:59,380][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:01:59,403][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:01:59,404][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:01:59,404][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:02:00,114][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:02:00,569][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:02:01,072][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:02:01,572][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:02:02,071][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:02:02,573][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:02:03,073][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:02:03,579][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:02:04,085][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:02:04,592][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:02:05,107][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:02:05,609][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:02:06,108][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:02:06,606][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:02:07,110][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:02:07,613][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:02:08,118][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:02:08,617][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:02:09,115][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:02:09,614][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:02:10,113][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:02:10,611][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:02:11,112][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:02:11,615][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:02:12,114][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:02:12,610][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:02:13,106][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:02:13,606][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:02:14,101][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:02:14,599][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:02:15,096][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:02:15,592][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:02:16,090][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:02:16,589][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:02:17,090][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:02:17,593][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:02:18,092][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:02:18,597][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:02:19,101][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:02:19,600][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:02:20,100][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:02:20,600][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:02:21,101][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:02:21,603][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:02:22,102][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:02:22,601][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:02:23,099][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:02:23,599][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:02:24,096][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:02:24,594][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:02:25,092][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:02:25,589][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:02:26,089][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:02:26,586][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:02:27,083][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:02:27,579][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:02:28,075][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:02:28,574][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:02:29,070][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:02:29,564][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:02:30,060][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:02:30,555][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:02:31,052][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:02:31,551][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:02:32,050][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10133 tokens. +[2026-03-26 03:02:32,782][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.25%, ΔTime: 00:00:32 +[2026-03-26 03:02:33,554][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:02:33,556][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:02:33,558][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:02:35,659][__main__][INFO] - Iteration 341 took 55s (32.85% Gen, 63.34% Train). Generation: 18s, Training: 34s. Estimated remaining time: 41h 0m 37s. Estimated total time: 46h 2m 24s. Time estimates for 10 more iterations: 9m 12s, 100 more iterations: 1h 32m 4s, 500 more iterations: 7h 40m 24s. +[2026-03-26 03:02:35,662][__main__][INFO] - Starting iteration 341. +[2026-03-26 03:02:36,064][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 34 and human policies 1. +[2026-03-26 03:02:36,064][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:02:40,949][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 03:02:51,649][__main__][INFO] - Number of regex retries in iteration 341: 1 +[2026-03-26 03:02:51,650][__main__][INFO] - agents played in iteration 341 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:02:52,414][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:02:52,437][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:02:52,460][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:02:52,483][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:02:52,483][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:02:52,484][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:02:53,177][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:02:53,632][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:02:54,138][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:02:54,645][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:02:55,153][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:02:55,659][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:02:56,159][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:02:56,658][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:02:57,161][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:02:57,667][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:02:58,168][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:02:58,668][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:02:59,170][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:02:59,670][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:03:00,172][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:03:00,671][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:03:01,170][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:03:01,669][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:03:02,168][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:03:02,666][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:03:03,165][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:03:03,664][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:03:04,162][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:03:04,661][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:03:05,160][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:03:05,659][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:03:06,158][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:03:06,656][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:03:07,154][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:03:07,653][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:03:08,152][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:03:08,651][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:03:09,150][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:03:09,650][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:03:10,147][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:03:10,646][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:03:11,142][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:03:11,641][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:03:12,140][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:03:12,638][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:03:13,135][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:03:13,634][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:03:14,135][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:03:14,631][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:03:15,128][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:03:15,624][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:03:16,122][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:03:16,619][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:03:17,119][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:03:17,618][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:03:18,119][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:03:18,618][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:03:19,116][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:03:19,615][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:03:20,116][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:03:20,616][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:03:21,114][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:03:21,612][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:03:22,111][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:03:22,608][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:03:23,107][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:03:23,607][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:03:24,104][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:03:24,602][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:03:25,100][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10044 tokens. +[2026-03-26 03:03:25,827][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.11%, ΔTime: 00:00:32 +[2026-03-26 03:03:26,584][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:03:26,587][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:03:26,589][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:03:27,292][__main__][INFO] - Iteration 342 took 51s (30.42% Gen, 68.20% Train). Generation: 15s, Training: 34s. Estimated remaining time: 37h 38m 48s. Estimated total time: 42h 41m 26s. Time estimates for 10 more iterations: 8m 32s, 100 more iterations: 1h 25m 22s, 500 more iterations: 7h 6m 54s. +[2026-03-26 03:03:27,294][__main__][INFO] - Starting iteration 342. +[2026-03-26 03:03:27,694][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 34 and human policies 1. +[2026-03-26 03:03:27,695][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:03:42,739][__main__][INFO] - Number of regex retries in iteration 342: 0 +[2026-03-26 03:03:42,739][__main__][INFO] - agents played in iteration 342 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:03:43,499][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:03:43,522][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:03:43,545][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:03:43,568][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:03:43,569][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:03:43,570][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:03:44,287][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:03:44,746][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:03:45,259][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:03:45,767][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:03:46,277][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:03:46,779][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:03:47,280][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:03:47,782][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:03:48,282][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:03:48,782][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:03:49,283][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:03:49,783][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:03:50,284][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:03:50,784][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:03:51,281][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:03:51,779][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:03:52,277][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:03:52,775][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:03:53,273][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:03:53,771][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:03:54,268][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:03:54,765][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:03:55,263][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:03:55,760][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:03:56,258][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:03:56,755][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:03:57,251][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:03:57,747][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:03:58,243][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:03:58,739][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:03:59,235][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:03:59,731][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:04:00,227][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:04:00,728][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:04:01,232][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:04:01,728][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:04:02,229][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:04:02,726][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:04:03,225][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:04:03,724][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:04:04,221][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:04:04,717][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:04:05,215][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:04:05,713][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:04:06,214][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:04:06,709][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:04:07,205][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:04:07,702][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:04:08,199][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:04:08,696][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:04:09,192][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:04:09,690][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:04:10,186][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:04:10,685][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:04:11,180][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:04:11,677][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:04:12,173][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:04:12,673][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:04:13,176][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:04:13,675][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:04:14,176][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:04:14,681][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:04:15,180][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:04:15,677][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:04:16,177][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10038 tokens. +[2026-03-26 03:04:16,930][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.17%, Current % of VRAM taken: 60.60%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 03:04:17,686][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:04:17,689][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:04:17,691][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:04:18,415][__main__][INFO] - Iteration 343 took 50s (29.66% Gen, 68.91% Train). Generation: 15s, Training: 34s. Estimated remaining time: 37h 12m 36s. Estimated total time: 42h 16m 6s. Time estimates for 10 more iterations: 8m 27s, 100 more iterations: 1h 24m 32s, 500 more iterations: 7h 2m 41s. +[2026-03-26 03:04:18,418][__main__][INFO] - Starting iteration 343. +[2026-03-26 03:04:18,821][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 34 and human policies 1. +[2026-03-26 03:04:18,821][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:04:28,114][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 03:04:36,739][__main__][INFO] - Number of regex retries in iteration 343: 1 +[2026-03-26 03:04:36,740][__main__][INFO] - agents played in iteration 343 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:04:37,537][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:04:37,561][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:04:37,584][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:04:37,608][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:04:37,608][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:04:37,609][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:04:38,324][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:04:38,783][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:04:39,288][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:04:39,791][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:04:40,292][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:04:40,794][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:04:41,296][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:04:41,798][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:04:42,299][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:04:42,801][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:04:43,302][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:04:43,804][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:04:44,305][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:04:44,805][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:04:45,303][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:04:45,805][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:04:46,305][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:04:46,803][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:04:47,306][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:04:47,805][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:04:48,306][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:04:48,808][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:04:49,307][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:04:49,806][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:04:50,307][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:04:50,808][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:04:51,308][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:04:51,809][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:04:52,310][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:04:52,807][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:04:53,309][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:04:53,809][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:04:54,313][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:04:54,816][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:04:55,314][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:04:55,813][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:04:56,311][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:04:56,811][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:04:57,313][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:04:57,814][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:04:58,311][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:04:58,810][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:04:59,308][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:04:59,807][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:05:00,304][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:05:00,803][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:05:01,302][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:05:01,800][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:05:02,298][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:05:02,798][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:05:03,296][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:05:03,795][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:05:04,295][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:05:04,792][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:05:05,292][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:05:05,789][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:05:06,289][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:05:06,787][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:05:07,285][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:05:07,783][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:05:08,281][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:05:08,779][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:05:09,276][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:05:09,777][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:05:10,276][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10225 tokens. +[2026-03-26 03:05:11,021][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.20%, ΔTime: 00:00:32 +[2026-03-26 03:05:11,780][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:05:11,783][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:05:11,788][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:05:12,514][__main__][INFO] - Iteration 344 took 53s (33.37% Gen, 65.27% Train). Generation: 17s, Training: 35s. Estimated remaining time: 39h 40m 17s. Estimated total time: 44h 44m 41s. Time estimates for 10 more iterations: 8m 56s, 100 more iterations: 1h 29m 29s, 500 more iterations: 7h 27m 26s. +[2026-03-26 03:05:12,516][__main__][INFO] - Starting iteration 344. +[2026-03-26 03:05:12,918][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 34 and human policies 1. +[2026-03-26 03:05:12,918][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:05:29,001][__main__][INFO] - Number of regex retries in iteration 344: 0 +[2026-03-26 03:05:29,002][__main__][INFO] - agents played in iteration 344 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:05:29,775][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:05:29,799][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:05:29,822][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:05:29,845][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:05:29,846][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:05:29,847][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:05:30,548][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:05:31,009][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:05:31,515][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:05:32,014][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:05:32,515][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:05:33,014][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:05:33,513][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:05:34,011][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:05:34,508][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:05:35,008][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:05:35,508][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:05:36,008][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:05:36,507][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:05:37,007][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:05:37,504][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:05:38,002][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:05:38,504][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:05:39,001][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:05:39,504][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:05:40,002][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:05:40,506][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:05:41,007][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:05:41,510][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:05:42,011][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:05:42,510][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:05:43,010][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:05:43,512][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:05:44,010][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:05:44,508][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:05:45,005][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:05:45,504][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:05:46,001][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:05:46,503][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:05:47,003][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:05:47,504][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:05:48,006][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:05:48,506][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:05:49,010][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:05:49,510][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:05:50,009][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:05:50,509][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:05:51,011][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:05:51,508][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:05:52,006][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:05:52,508][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:05:53,007][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:05:53,507][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:05:54,004][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:05:54,503][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:05:55,000][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:05:55,499][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:05:55,995][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:05:56,494][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:05:56,992][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:05:57,489][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:05:57,986][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:05:58,484][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:05:58,984][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:05:59,481][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:05:59,977][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:06:00,474][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:06:00,973][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:06:01,471][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:06:01,969][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:06:02,468][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10111 tokens. +[2026-03-26 03:06:03,203][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.04%, Current % of VRAM taken: 60.47%, Block Peak % of device VRAM: 62.03%, ΔTime: 00:00:32 +[2026-03-26 03:06:03,961][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:06:03,965][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:06:03,966][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:06:04,683][__main__][INFO] - Iteration 345 took 51s (31.07% Gen, 67.54% Train). Generation: 16s, Training: 34s. Estimated remaining time: 38h 3m 1s. Estimated total time: 43h 8m 17s. Time estimates for 10 more iterations: 8m 37s, 100 more iterations: 1h 26m 16s, 500 more iterations: 7h 11m 22s. +[2026-03-26 03:06:04,686][__main__][INFO] - Starting iteration 345. +[2026-03-26 03:06:05,087][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 34 and human policies 1. +[2026-03-26 03:06:05,087][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:06:21,549][__main__][INFO] - Number of regex retries in iteration 345: 0 +[2026-03-26 03:06:21,550][__main__][INFO] - agents played in iteration 345 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:06:22,335][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:06:22,359][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:06:22,382][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:06:22,404][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:06:22,405][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:06:22,406][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:06:23,108][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:06:23,568][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:06:24,072][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:06:24,577][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:06:25,077][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:06:25,580][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:06:26,080][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:06:26,581][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:06:27,085][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:06:27,588][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:06:28,089][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:06:28,590][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:06:29,088][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:06:29,593][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:06:30,093][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:06:30,593][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:06:31,093][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:06:31,595][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:06:32,099][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:06:32,599][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:06:33,100][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:06:33,602][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:06:34,103][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:06:34,604][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:06:35,107][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:06:35,605][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:06:36,104][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:06:36,602][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:06:37,102][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:06:37,602][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:06:38,102][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:06:38,604][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:06:39,106][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:06:39,603][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:06:40,101][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:06:40,598][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:06:41,098][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:06:41,598][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:06:42,095][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:06:42,591][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:06:43,088][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:06:43,584][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:06:44,084][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:06:44,580][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:06:45,078][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:06:45,575][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:06:46,074][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:06:46,572][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:06:47,069][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:06:47,567][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:06:48,069][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:06:48,569][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:06:49,070][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:06:49,570][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:06:50,071][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:06:50,571][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:06:51,071][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:06:51,568][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:06:52,065][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:06:52,565][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:06:53,061][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:06:53,558][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:06:54,056][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:06:54,555][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:06:55,052][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10195 tokens. +[2026-03-26 03:06:55,785][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.28%, ΔTime: 00:00:32 +[2026-03-26 03:06:56,534][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:06:56,536][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:06:56,538][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:06:57,266][__main__][INFO] - Iteration 346 took 52s (31.55% Gen, 67.05% Train). Generation: 16s, Training: 34s. Estimated remaining time: 38h 22m 52s. Estimated total time: 43h 29m 1s. Time estimates for 10 more iterations: 8m 41s, 100 more iterations: 1h 26m 58s, 500 more iterations: 7h 14m 50s. +[2026-03-26 03:06:57,269][__main__][INFO] - Starting iteration 346. +[2026-03-26 03:06:57,670][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 34 and human policies 1. +[2026-03-26 03:06:57,671][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:07:01,558][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 03:07:13,166][__main__][INFO] - Number of regex retries in iteration 346: 1 +[2026-03-26 03:07:13,167][__main__][INFO] - agents played in iteration 346 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:07:13,959][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:07:13,982][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:07:14,005][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:07:14,028][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:07:14,029][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:07:14,030][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:07:14,729][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:07:15,188][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:07:15,691][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:07:16,192][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:07:16,692][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:07:17,195][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:07:17,695][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:07:18,197][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:07:18,702][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:07:19,202][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:07:19,703][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:07:20,201][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:07:20,699][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:07:21,199][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:07:21,698][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:07:22,196][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:07:22,693][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:07:23,190][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:07:23,688][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:07:24,184][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:07:24,681][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:07:25,178][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:07:25,676][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:07:26,174][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:07:26,672][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:07:27,173][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:07:27,673][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:07:28,170][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:07:28,672][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:07:29,172][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:07:29,671][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:07:30,171][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:07:30,669][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:07:31,173][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:07:31,671][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:07:32,173][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:07:32,672][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:07:33,171][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:07:33,668][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:07:34,165][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:07:34,662][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:07:35,158][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:07:35,654][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:07:36,151][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:07:36,646][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:07:37,145][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:07:37,640][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:07:38,136][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:07:38,632][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:07:39,129][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:07:39,625][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:07:40,122][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:07:40,619][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:07:41,115][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:07:41,612][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:07:42,108][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:07:42,604][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:07:43,100][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:07:43,596][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:07:44,094][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:07:44,591][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:07:45,089][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:07:45,586][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:07:46,082][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:07:46,578][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10095 tokens. +[2026-03-26 03:07:47,310][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.03%, Current % of VRAM taken: 60.46%, Block Peak % of device VRAM: 62.03%, ΔTime: 00:00:32 +[2026-03-26 03:07:48,101][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:07:48,103][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:07:48,107][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:07:48,876][__main__][INFO] - Iteration 347 took 51s (30.26% Gen, 68.23% Train). Generation: 15s, Training: 34s. Estimated remaining time: 37h 33m 20s. Estimated total time: 42h 40m 21s. Time estimates for 10 more iterations: 8m 32s, 100 more iterations: 1h 25m 20s, 500 more iterations: 7h 6m 43s. +[2026-03-26 03:07:48,879][__main__][INFO] - Starting iteration 347. +[2026-03-26 03:07:49,285][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 34 and human policies 1. +[2026-03-26 03:07:49,286][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:08:05,398][__main__][INFO] - Number of regex retries in iteration 347: 0 +[2026-03-26 03:08:05,399][__main__][INFO] - agents played in iteration 347 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:08:06,180][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:08:06,204][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:08:06,228][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:08:06,251][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:08:06,252][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:08:06,252][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:08:06,962][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:08:07,421][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:08:07,926][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:08:08,427][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:08:08,926][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:08:09,425][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:08:09,925][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:08:10,426][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:08:10,925][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:08:11,421][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:08:11,924][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:08:12,421][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:08:12,918][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:08:13,415][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:08:13,913][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:08:14,410][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:08:14,909][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:08:15,406][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:08:15,905][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:08:16,405][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:08:16,904][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:08:17,402][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:08:17,899][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:08:18,399][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:08:18,898][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:08:19,397][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:08:19,895][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:08:20,396][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:08:20,894][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:08:21,392][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:08:21,894][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:08:22,392][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:08:22,896][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:08:23,395][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:08:23,893][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:08:24,393][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:08:24,892][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:08:25,390][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:08:25,889][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:08:26,387][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:08:26,883][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:08:27,379][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:08:27,876][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:08:28,379][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:08:28,877][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:08:29,375][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:08:29,873][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:08:30,372][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:08:30,870][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:08:31,367][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:08:31,865][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:08:32,364][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:08:32,861][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:08:33,360][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:08:33,857][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:08:34,355][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:08:34,853][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:08:35,350][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:08:35,846][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:08:36,342][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:08:36,838][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:08:37,334][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:08:37,830][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:08:38,326][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:08:38,823][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10054 tokens. +[2026-03-26 03:08:39,553][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 62.21%, ΔTime: 00:00:32 +[2026-03-26 03:08:40,308][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:08:40,311][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:08:40,313][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:08:41,188][__main__][INFO] - Iteration 348 took 51s (31.04% Gen, 67.26% Train). Generation: 16s, Training: 34s. Estimated remaining time: 38h 7m 30s. Estimated total time: 43h 15m 22s. Time estimates for 10 more iterations: 8m 39s, 100 more iterations: 1h 26m 30s, 500 more iterations: 7h 12m 33s. +[2026-03-26 03:08:41,190][__main__][INFO] - Starting iteration 348. +[2026-03-26 03:08:41,591][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 34 and human policies 1. +[2026-03-26 03:08:41,591][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:08:59,174][__main__][INFO] - Number of regex retries in iteration 348: 0 +[2026-03-26 03:08:59,174][__main__][INFO] - agents played in iteration 348 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:08:59,963][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:08:59,986][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:09:00,010][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:09:00,033][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:09:00,033][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:09:00,034][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:09:00,749][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:09:01,200][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:09:01,706][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:09:02,212][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:09:02,711][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:09:03,210][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:09:03,709][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:09:04,207][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:09:04,705][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:09:05,205][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:09:05,705][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:09:06,205][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:09:06,705][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:09:07,207][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:09:07,708][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:09:08,210][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:09:08,710][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:09:09,211][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:09:09,710][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:09:10,213][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:09:10,711][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:09:11,207][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:09:11,709][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:09:12,206][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:09:12,704][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:09:13,201][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:09:13,704][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:09:14,204][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:09:14,702][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:09:15,203][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:09:15,702][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:09:16,200][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:09:16,701][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:09:17,198][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:09:17,696][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:09:18,194][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:09:18,690][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:09:19,192][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:09:19,695][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:09:20,196][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:09:20,694][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:09:21,195][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:09:21,692][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:09:22,191][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:09:22,687][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:09:23,183][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:09:23,681][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:09:24,178][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:09:24,678][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:09:25,176][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:09:25,677][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:09:26,177][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:09:26,674][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:09:27,171][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:09:27,672][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:09:28,170][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:09:28,671][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:09:29,169][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:09:29,667][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:09:30,164][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:09:30,660][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:09:31,156][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:09:31,657][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:09:32,153][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:09:32,651][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10190 tokens. +[2026-03-26 03:09:33,394][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 03:09:34,155][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:09:34,157][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:09:34,159][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:09:34,885][__main__][INFO] - Iteration 349 took 53s (32.99% Gen, 65.64% Train). Generation: 17s, Training: 34s. Estimated remaining time: 39h 15m 58s. Estimated total time: 44h 24m 45s. Time estimates for 10 more iterations: 8m 52s, 100 more iterations: 1h 28m 49s, 500 more iterations: 7h 24m 7s. +[2026-03-26 03:09:34,887][__main__][INFO] - Starting iteration 349. +[2026-03-26 03:09:35,290][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 34 and human policies 1. +[2026-03-26 03:09:35,290][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:09:51,936][__main__][INFO] - Number of regex retries in iteration 349: 0 +[2026-03-26 03:09:51,937][__main__][INFO] - agents played in iteration 349 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:09:52,702][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:09:52,725][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:09:52,748][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:09:52,771][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:09:52,772][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:09:52,772][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:09:53,480][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:09:53,933][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:09:54,440][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:09:54,939][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:09:55,437][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:09:55,935][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:09:56,433][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:09:56,932][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:09:57,436][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:09:57,936][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:09:58,442][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:09:58,942][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:09:59,441][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:09:59,940][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:10:00,438][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:10:00,938][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:10:01,435][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:10:01,933][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:10:02,434][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:10:02,936][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:10:03,438][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:10:03,937][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:10:04,434][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:10:04,935][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:10:05,460][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:10:05,958][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:10:06,454][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:10:06,950][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:10:07,446][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:10:07,943][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:10:08,439][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:10:08,937][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:10:09,434][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:10:09,931][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:10:10,429][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:10:10,926][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:10:11,422][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:10:11,922][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:10:12,418][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:10:12,919][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:10:13,414][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:10:13,914][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:10:14,415][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:10:14,915][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:10:15,417][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:10:15,916][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:10:16,416][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:10:16,914][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:10:17,412][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:10:17,919][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:10:18,421][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:10:18,919][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:10:19,420][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:10:19,921][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:10:20,423][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:10:20,923][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:10:21,423][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:10:21,922][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:10:22,419][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:10:22,916][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:10:23,413][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:10:23,911][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:10:24,410][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:10:24,908][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:10:25,409][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10175 tokens. +[2026-03-26 03:10:26,154][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.20%, Current % of VRAM taken: 60.62%, Block Peak % of device VRAM: 62.33%, ΔTime: 00:00:32 +[2026-03-26 03:10:26,912][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:10:26,914][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:10:26,916][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:10:27,676][__main__][INFO] - Iteration 350 took 52s (31.78% Gen, 66.77% Train). Generation: 16s, Training: 34s. Estimated remaining time: 38h 29m 42s. Estimated total time: 43h 39m 21s. Time estimates for 10 more iterations: 8m 43s, 100 more iterations: 1h 27m 18s, 500 more iterations: 7h 16m 33s. +[2026-03-26 03:10:27,678][__main__][INFO] - Starting iteration 350. +[2026-03-26 03:10:28,081][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 34 and human policies 1. +[2026-03-26 03:10:28,081][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:10:37,461][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 20 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 03:10:45,294][__main__][INFO] - Number of regex retries in iteration 350: 1 +[2026-03-26 03:10:45,295][__main__][INFO] - agents played in iteration 350 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:10:46,078][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:10:46,101][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:10:46,124][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:10:46,147][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:10:46,148][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:10:46,148][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:10:46,851][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:10:47,305][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:10:47,813][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:10:48,312][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:10:48,813][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:10:49,315][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:10:49,816][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:10:50,317][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:10:50,819][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:10:51,320][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:10:51,827][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:10:52,330][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:10:52,830][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:10:53,331][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:10:53,836][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:10:54,338][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:10:54,837][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:10:55,338][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:10:55,836][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:10:56,337][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:10:56,836][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:10:57,334][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:10:57,832][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:10:58,330][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:10:58,829][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:10:59,327][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:10:59,823][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:11:00,320][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:11:00,817][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:11:01,315][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:11:01,811][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:11:02,306][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:11:02,800][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:11:03,300][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:11:03,801][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:11:04,301][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:11:04,802][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:11:05,302][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:11:05,820][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:11:06,320][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:11:06,819][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:11:07,315][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:11:07,814][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:11:08,310][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:11:08,806][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:11:09,302][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:11:09,799][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:11:10,298][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:11:10,794][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:11:11,290][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:11:11,788][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:11:12,284][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:11:12,780][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:11:13,277][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:11:13,774][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:11:14,275][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:11:14,771][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:11:15,269][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:11:15,770][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:11:16,269][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:11:16,770][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:11:17,267][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:11:17,764][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:11:18,261][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:11:18,758][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10091 tokens. +[2026-03-26 03:11:19,483][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.15%, Current % of VRAM taken: 60.58%, Block Peak % of device VRAM: 62.29%, ΔTime: 00:00:32 +[2026-03-26 03:11:20,498][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:11:21,679][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:11:21,681][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:11:23,296][__main__][INFO] - Iteration 351 took 55s (31.18% Gen, 65.90% Train). Generation: 17s, Training: 36s. Estimated remaining time: 40h 50m 15s. Estimated total time: 46h 0m 49s. Time estimates for 10 more iterations: 9m 12s, 100 more iterations: 1h 32m 1s, 500 more iterations: 7h 40m 8s. +[2026-03-26 03:11:23,298][__main__][INFO] - Starting iteration 351. +[2026-03-26 03:11:23,697][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 35 and human policies 1. +[2026-03-26 03:11:23,698][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:11:39,725][__main__][INFO] - Number of regex retries in iteration 351: 0 +[2026-03-26 03:11:39,726][__main__][INFO] - agents played in iteration 351 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:11:40,503][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:11:40,526][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:11:40,549][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:11:40,572][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:11:40,573][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:11:40,573][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:11:41,273][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:11:41,731][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:11:42,237][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:11:42,739][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:11:43,237][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:11:43,736][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:11:44,237][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:11:44,745][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:11:45,249][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:11:45,752][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:11:46,250][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:11:46,748][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:11:47,250][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:11:47,751][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:11:48,254][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:11:48,755][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:11:49,253][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:11:49,756][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:11:50,255][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:11:50,755][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:11:51,251][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:11:51,752][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:11:52,252][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:11:52,752][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:11:53,254][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:11:53,751][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:11:54,251][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:11:54,751][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:11:55,250][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:11:55,746][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:11:56,246][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:11:56,742][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:11:57,238][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:11:57,735][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:11:58,232][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:11:58,729][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:11:59,225][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:11:59,725][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:12:00,222][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:12:00,719][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:12:01,217][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:12:01,716][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:12:02,215][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:12:02,712][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:12:03,210][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:12:03,707][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:12:04,209][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:12:04,707][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:12:05,205][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:12:05,703][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:12:06,200][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:12:06,698][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:12:07,195][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:12:07,692][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:12:08,191][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:12:08,689][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:12:09,189][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:12:09,686][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:12:10,184][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:12:10,682][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:12:11,181][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:12:11,678][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:12:12,176][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:12:12,677][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:12:13,179][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10259 tokens. +[2026-03-26 03:12:13,918][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.19%, Current % of VRAM taken: 60.62%, Block Peak % of device VRAM: 62.20%, ΔTime: 00:00:32 +[2026-03-26 03:12:14,679][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:12:14,682][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:12:14,684][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:12:15,458][__main__][INFO] - Iteration 352 took 51s (30.96% Gen, 67.54% Train). Generation: 16s, Training: 34s. Estimated remaining time: 37h 56m 37s. Estimated total time: 43h 8m 4s. Time estimates for 10 more iterations: 8m 37s, 100 more iterations: 1h 26m 16s, 500 more iterations: 7h 11m 20s. +[2026-03-26 03:12:15,460][__main__][INFO] - Starting iteration 352. +[2026-03-26 03:12:15,860][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 35 and human policies 1. +[2026-03-26 03:12:15,861][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:12:31,623][__main__][INFO] - Number of regex retries in iteration 352: 0 +[2026-03-26 03:12:31,624][__main__][INFO] - agents played in iteration 352 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:12:32,389][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:12:32,412][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:12:32,435][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:12:32,458][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:12:32,459][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:12:32,459][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:12:33,162][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:12:33,617][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:12:34,120][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:12:34,619][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:12:35,119][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:12:35,624][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:12:36,122][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:12:36,620][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:12:37,119][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:12:37,622][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:12:38,125][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:12:38,627][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:12:39,130][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:12:39,630][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:12:40,132][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:12:40,638][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:12:41,142][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:12:41,639][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:12:42,136][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:12:42,634][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:12:43,132][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:12:43,630][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:12:44,127][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:12:44,625][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:12:45,122][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:12:45,618][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:12:46,115][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:12:46,616][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:12:47,113][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:12:47,611][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:12:48,109][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:12:48,607][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:12:49,104][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:12:49,604][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:12:50,106][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:12:50,604][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:12:51,102][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:12:51,602][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:12:52,102][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:12:52,601][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:12:53,100][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:12:53,598][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:12:54,096][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:12:54,594][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:12:55,092][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:12:55,592][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:12:56,092][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:12:56,590][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:12:57,092][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:12:57,589][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:12:58,090][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:12:58,586][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:12:59,083][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:12:59,582][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:13:00,078][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:13:00,576][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:13:01,077][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:13:01,574][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:13:02,072][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:13:02,569][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:13:03,064][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:13:03,566][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:13:04,068][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:13:04,568][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:13:05,068][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10083 tokens. +[2026-03-26 03:13:05,804][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.02%, Current % of VRAM taken: 60.45%, Block Peak % of device VRAM: 62.36%, ΔTime: 00:00:32 +[2026-03-26 03:13:06,565][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:13:06,568][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:13:06,569][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:13:07,320][__main__][INFO] - Iteration 353 took 51s (30.63% Gen, 67.91% Train). Generation: 15s, Training: 34s. Estimated remaining time: 37h 40m 43s. Estimated total time: 42h 53m 2s. Time estimates for 10 more iterations: 8m 34s, 100 more iterations: 1h 25m 46s, 500 more iterations: 7h 8m 50s. +[2026-03-26 03:13:07,322][__main__][INFO] - Starting iteration 353. +[2026-03-26 03:13:07,724][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 35 and human policies 1. +[2026-03-26 03:13:07,725][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:13:23,466][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 1, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 03:13:24,240][__main__][INFO] - Number of regex retries in iteration 353: 1 +[2026-03-26 03:13:24,241][__main__][INFO] - agents played in iteration 353 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:13:25,011][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:13:25,034][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:13:25,058][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:13:25,081][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:13:25,081][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:13:25,082][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:13:25,784][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:13:26,236][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:13:26,741][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:13:27,238][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:13:27,737][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:13:28,237][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:13:28,735][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:13:29,236][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:13:29,740][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:13:30,239][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:13:30,740][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:13:31,241][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:13:31,741][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:13:32,240][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:13:32,738][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:13:33,242][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:13:33,741][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:13:34,241][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:13:34,767][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:13:35,269][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:13:35,773][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:13:36,278][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:13:36,777][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:13:37,282][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:13:37,784][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:13:38,281][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:13:38,781][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:13:39,278][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:13:39,777][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:13:40,274][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:13:40,771][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:13:41,267][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:13:41,764][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:13:42,260][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:13:42,756][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:13:43,254][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:13:43,750][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:13:44,247][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:13:44,744][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:13:45,240][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:13:45,739][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:13:46,236][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:13:46,733][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:13:47,231][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:13:47,728][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:13:48,224][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:13:48,724][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:13:49,221][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:13:49,721][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:13:50,222][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:13:50,723][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:13:51,223][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:13:51,728][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:13:52,230][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:13:52,732][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:13:53,238][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:13:53,738][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:13:54,238][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:13:54,739][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:13:55,245][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:13:55,749][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:13:56,250][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:13:56,751][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:13:57,251][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:13:57,751][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10132 tokens. +[2026-03-26 03:13:58,482][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.61%, Block Peak % of device VRAM: 62.39%, ΔTime: 00:00:32 +[2026-03-26 03:13:59,247][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:13:59,249][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:13:59,251][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:14:00,017][__main__][INFO] - Iteration 354 took 52s (31.58% Gen, 66.95% Train). Generation: 16s, Training: 35s. Estimated remaining time: 38h 21m 27s. Estimated total time: 43h 34m 38s. Time estimates for 10 more iterations: 8m 42s, 100 more iterations: 1h 27m 9s, 500 more iterations: 7h 15m 46s. +[2026-03-26 03:14:00,019][__main__][INFO] - Starting iteration 354. +[2026-03-26 03:14:00,418][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 35 and human policies 1. +[2026-03-26 03:14:00,419][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:14:19,107][__main__][INFO] - Number of regex retries in iteration 354: 0 +[2026-03-26 03:14:19,108][__main__][INFO] - agents played in iteration 354 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:14:19,875][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:14:19,898][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:14:19,921][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:14:19,944][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:14:19,945][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:14:19,946][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:14:20,654][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:14:21,108][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:14:21,607][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:14:22,103][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:14:22,600][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:14:23,097][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:14:23,597][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:14:24,099][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:14:24,597][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:14:25,099][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:14:25,601][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:14:26,101][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:14:26,625][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:14:27,131][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:14:27,631][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:14:28,132][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:14:28,633][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:14:29,131][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:14:29,638][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:14:30,139][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:14:30,639][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:14:31,140][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:14:31,641][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:14:32,144][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:14:32,645][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:14:33,146][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:14:33,641][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:14:34,138][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:14:34,635][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:14:35,133][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:14:35,631][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:14:36,128][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:14:36,625][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:14:37,126][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:14:37,627][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:14:38,126][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:14:38,630][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:14:39,130][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:14:39,630][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:14:40,131][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:14:40,631][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:14:41,130][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:14:41,628][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:14:42,125][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:14:42,625][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:14:43,122][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:14:43,619][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:14:44,117][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:14:44,618][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:14:45,121][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:14:45,625][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:14:46,123][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:14:46,619][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:14:47,115][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:14:47,612][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:14:48,112][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:14:48,611][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:14:49,107][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:14:49,604][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:14:50,103][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:14:50,601][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:14:51,099][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:14:51,598][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:14:52,096][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:14:52,595][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10072 tokens. +[2026-03-26 03:14:53,332][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.61%, Block Peak % of device VRAM: 62.40%, ΔTime: 00:00:32 +[2026-03-26 03:14:54,089][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:14:54,092][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:14:54,093][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:14:54,808][__main__][INFO] - Iteration 355 took 54s (34.36% Gen, 64.32% Train). Generation: 18s, Training: 34s. Estimated remaining time: 40h 5m 26s. Estimated total time: 45h 19m 32s. Time estimates for 10 more iterations: 9m 3s, 100 more iterations: 1h 30m 39s, 500 more iterations: 7h 33m 15s. +[2026-03-26 03:14:54,811][__main__][INFO] - Starting iteration 355. +[2026-03-26 03:14:55,210][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 35 and human policies 1. +[2026-03-26 03:14:55,211][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:15:11,860][__main__][INFO] - Number of regex retries in iteration 355: 0 +[2026-03-26 03:15:11,861][__main__][INFO] - agents played in iteration 355 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:15:12,632][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:15:12,655][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:15:12,678][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:15:12,701][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:15:12,702][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:15:12,702][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:15:13,405][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:15:13,863][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:15:14,365][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:15:14,864][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:15:15,360][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:15:15,863][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:15:16,360][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:15:16,857][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:15:17,356][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:15:17,861][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:15:18,361][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:15:18,861][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:15:19,361][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:15:19,860][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:15:20,362][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:15:20,865][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:15:21,364][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:15:21,862][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:15:22,359][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:15:22,856][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:15:23,357][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:15:23,858][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:15:24,354][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:15:24,853][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:15:25,351][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:15:25,848][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:15:26,346][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:15:26,841][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:15:27,338][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:15:27,835][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:15:28,335][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:15:28,831][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:15:29,328][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:15:29,824][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:15:30,323][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:15:30,823][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:15:31,322][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:15:31,822][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:15:32,323][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:15:32,822][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:15:33,322][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:15:33,820][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:15:34,319][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:15:34,817][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:15:35,316][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:15:35,813][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:15:36,313][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:15:36,812][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:15:37,312][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:15:37,814][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:15:38,314][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:15:38,813][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:15:39,310][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:15:39,807][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:15:40,303][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:15:40,799][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:15:41,295][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:15:41,792][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:15:42,289][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:15:42,785][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:15:43,281][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:15:43,777][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:15:44,273][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:15:44,770][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:15:45,267][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10107 tokens. +[2026-03-26 03:15:46,001][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.37%, ΔTime: 00:00:32 +[2026-03-26 03:15:46,755][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:15:46,757][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:15:46,759][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:15:47,482][__main__][INFO] - Iteration 356 took 52s (31.85% Gen, 66.76% Train). Generation: 16s, Training: 34s. Estimated remaining time: 38h 18m 37s. Estimated total time: 43h 33m 36s. Time estimates for 10 more iterations: 8m 42s, 100 more iterations: 1h 27m 7s, 500 more iterations: 7h 15m 36s. +[2026-03-26 03:15:47,484][__main__][INFO] - Starting iteration 356. +[2026-03-26 03:15:47,885][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 35 and human policies 1. +[2026-03-26 03:15:47,886][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:16:03,639][__main__][INFO] - Number of regex retries in iteration 356: 0 +[2026-03-26 03:16:03,640][__main__][INFO] - agents played in iteration 356 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:16:04,407][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:16:04,432][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:16:04,455][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:16:04,481][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:16:04,482][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:16:04,482][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:16:05,188][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:16:05,642][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:16:06,151][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:16:06,652][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:16:07,149][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:16:07,647][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:16:08,144][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:16:08,647][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:16:09,145][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:16:09,643][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:16:10,140][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:16:10,636][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:16:11,132][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:16:11,632][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:16:12,132][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:16:12,628][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:16:13,126][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:16:13,624][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:16:14,121][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:16:14,618][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:16:15,116][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:16:15,619][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:16:16,120][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:16:16,617][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:16:17,113][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:16:17,611][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:16:18,113][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:16:18,613][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:16:19,111][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:16:19,611][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:16:20,112][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:16:20,610][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:16:21,108][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:16:21,606][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:16:22,104][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:16:22,603][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:16:23,101][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:16:23,598][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:16:24,096][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:16:24,595][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:16:25,093][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:16:25,591][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:16:26,090][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:16:26,590][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:16:27,087][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:16:27,588][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:16:28,093][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:16:28,594][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:16:29,094][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:16:29,597][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:16:30,100][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:16:30,598][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:16:31,095][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:16:31,595][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:16:32,093][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:16:32,592][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:16:33,092][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:16:33,592][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:16:34,090][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:16:34,585][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:16:35,081][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:16:35,576][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:16:36,077][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:16:36,574][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:16:37,071][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10062 tokens. +[2026-03-26 03:16:37,806][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.02%, Current % of VRAM taken: 60.45%, Block Peak % of device VRAM: 62.12%, ΔTime: 00:00:32 +[2026-03-26 03:16:38,564][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:16:38,567][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:16:38,568][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:16:39,337][__main__][INFO] - Iteration 357 took 51s (30.62% Gen, 67.89% Train). Generation: 15s, Training: 34s. Estimated remaining time: 37h 36m 45s. Estimated total time: 42h 52m 36s. Time estimates for 10 more iterations: 8m 34s, 100 more iterations: 1h 25m 45s, 500 more iterations: 7h 8m 46s. +[2026-03-26 03:16:39,339][__main__][INFO] - Starting iteration 357. +[2026-03-26 03:16:39,740][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 35 and human policies 1. +[2026-03-26 03:16:39,740][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:16:45,622][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 03:16:56,534][__main__][INFO] - Number of regex retries in iteration 357: 1 +[2026-03-26 03:16:56,535][__main__][INFO] - agents played in iteration 357 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:16:57,305][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:16:57,329][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:16:57,352][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:16:57,375][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:16:57,375][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:16:57,376][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:16:58,087][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:16:58,543][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:16:59,049][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:16:59,550][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:17:00,048][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:17:00,550][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:17:01,050][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:17:01,549][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:17:02,050][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:17:02,547][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:17:03,051][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:17:03,551][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:17:04,049][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:17:04,546][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:17:05,044][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:17:05,548][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:17:06,050][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:17:06,550][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:17:07,047][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:17:07,548][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:17:08,046][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:17:08,548][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:17:09,044][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:17:09,542][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:17:10,042][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:17:10,538][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:17:11,034][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:17:11,530][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:17:12,030][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:17:12,525][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:17:13,020][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:17:13,516][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:17:14,011][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:17:14,509][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:17:15,007][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:17:15,505][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:17:16,004][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:17:16,503][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:17:17,000][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:17:17,498][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:17:18,001][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:17:18,504][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:17:19,005][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:17:19,503][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:17:20,001][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:17:20,501][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:17:20,998][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:17:21,496][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:17:21,994][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:17:22,492][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:17:22,990][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:17:23,488][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:17:23,985][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:17:24,482][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:17:24,979][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:17:25,475][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:17:25,977][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:17:26,473][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:17:26,970][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:17:27,467][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:17:27,967][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:17:28,464][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:17:28,962][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:17:29,460][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:17:29,957][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10089 tokens. +[2026-03-26 03:17:30,694][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.16%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.29%, ΔTime: 00:00:32 +[2026-03-26 03:17:31,439][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:17:31,441][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:17:31,443][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:17:32,213][__main__][INFO] - Iteration 358 took 52s (32.01% Gen, 66.52% Train). Generation: 16s, Training: 34s. Estimated remaining time: 38h 26m 58s. Estimated total time: 43h 43m 41s. Time estimates for 10 more iterations: 8m 44s, 100 more iterations: 1h 27m 27s, 500 more iterations: 7h 17m 16s. +[2026-03-26 03:17:32,215][__main__][INFO] - Starting iteration 358. +[2026-03-26 03:17:32,618][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 35 and human policies 1. +[2026-03-26 03:17:32,619][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:17:49,430][__main__][INFO] - Number of regex retries in iteration 358: 0 +[2026-03-26 03:17:49,431][__main__][INFO] - agents played in iteration 358 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:17:50,206][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:17:50,229][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:17:50,252][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:17:50,276][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:17:50,276][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:17:50,277][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:17:50,988][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:17:51,462][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:17:51,967][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:17:52,465][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:17:52,962][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:17:53,460][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:17:53,961][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:17:54,458][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:17:54,960][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:17:55,457][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:17:55,958][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:17:56,465][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:17:56,966][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:17:57,467][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:17:57,968][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:17:58,464][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:17:58,965][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:17:59,462][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:17:59,960][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:18:00,459][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:18:00,957][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:18:01,455][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:18:01,953][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:18:02,453][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:18:02,974][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:18:03,473][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:18:03,973][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:18:04,477][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:18:04,973][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:18:05,469][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:18:05,971][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:18:06,472][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:18:06,969][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:18:07,469][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:18:07,969][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:18:08,475][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:18:08,978][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:18:09,482][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:18:09,983][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:18:10,478][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:18:10,975][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:18:11,472][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:18:11,968][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:18:12,461][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:18:12,956][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:18:13,451][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:18:13,948][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:18:14,446][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:18:14,942][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:18:15,445][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:18:15,942][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:18:16,438][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:18:16,937][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:18:17,436][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:18:17,932][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:18:18,431][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:18:18,931][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:18:19,429][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:18:19,926][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:18:20,424][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:18:20,922][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:18:21,419][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:18:21,916][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:18:22,413][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:18:22,910][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10144 tokens. +[2026-03-26 03:18:23,651][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.23%, ΔTime: 00:00:32 +[2026-03-26 03:18:24,403][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:18:24,405][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:18:24,407][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:18:25,152][__main__][INFO] - Iteration 359 took 52s (32.00% Gen, 66.58% Train). Generation: 16s, Training: 34s. Estimated remaining time: 38h 29m 8s. Estimated total time: 43h 46m 44s. Time estimates for 10 more iterations: 8m 45s, 100 more iterations: 1h 27m 33s, 500 more iterations: 7h 17m 47s. +[2026-03-26 03:18:25,155][__main__][INFO] - Starting iteration 359. +[2026-03-26 03:18:25,557][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 35 and human policies 1. +[2026-03-26 03:18:25,557][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:18:42,501][__main__][INFO] - Number of regex retries in iteration 359: 0 +[2026-03-26 03:18:42,502][__main__][INFO] - agents played in iteration 359 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:18:43,272][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:18:43,295][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:18:43,318][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:18:43,341][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:18:43,342][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:18:43,342][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:18:44,065][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:18:44,519][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:18:45,022][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:18:45,521][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:18:46,020][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:18:46,519][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:18:47,022][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:18:47,520][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:18:48,017][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:18:48,516][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:18:49,014][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:18:49,520][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:18:50,022][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:18:50,524][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:18:51,026][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:18:51,526][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:18:52,026][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:18:52,528][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:18:53,026][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:18:53,525][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:18:54,022][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:18:54,520][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:18:55,017][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:18:55,518][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:18:56,018][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:18:56,515][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:18:57,013][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:18:57,512][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:18:58,010][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:18:58,508][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:18:59,006][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:18:59,511][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:19:00,014][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:19:00,520][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:19:01,018][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:19:01,515][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:19:02,013][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:19:02,511][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:19:03,008][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:19:03,507][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:19:04,005][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:19:04,502][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:19:05,999][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:19:05,498][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:19:05,996][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:19:06,494][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:19:06,992][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:19:07,491][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:19:07,990][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:19:08,489][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:19:08,987][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:19:09,486][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:19:09,982][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:19:10,481][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:19:10,977][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:19:11,474][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:19:11,971][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:19:12,467][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:19:12,963][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:19:13,460][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:19:13,955][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:19:14,451][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:19:14,947][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:19:15,444][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:19:15,940][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10084 tokens. +[2026-03-26 03:19:16,686][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.02%, Current % of VRAM taken: 60.45%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 03:19:17,436][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:19:17,438][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:19:17,440][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:19:18,160][__main__][INFO] - Iteration 360 took 52s (32.21% Gen, 66.42% Train). Generation: 16s, Training: 34s. Estimated remaining time: 38h 31m 43s. Estimated total time: 43h 50m 13s. Time estimates for 10 more iterations: 8m 46s, 100 more iterations: 1h 27m 40s, 500 more iterations: 7h 18m 22s. +[2026-03-26 03:19:18,163][__main__][INFO] - Starting iteration 360. +[2026-03-26 03:19:18,563][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 35 and human policies 1. +[2026-03-26 03:19:18,563][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:19:24,609][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 03:19:33,323][__main__][INFO] - Number of regex retries in iteration 360: 1 +[2026-03-26 03:19:33,323][__main__][INFO] - agents played in iteration 360 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:19:34,086][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:19:34,110][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:19:34,133][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:19:34,156][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:19:34,157][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:19:34,157][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:19:34,877][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:19:35,334][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:19:35,839][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:19:36,338][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:19:36,841][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:19:37,341][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:19:37,841][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:19:38,341][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:19:38,841][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:19:39,338][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:19:39,836][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:19:40,334][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:19:40,832][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:19:41,331][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:19:41,830][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:19:42,328][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:19:42,827][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:19:43,328][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:19:43,831][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:19:44,337][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:19:44,836][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:19:45,336][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:19:45,835][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:19:46,335][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:19:46,836][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:19:47,333][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:19:47,830][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:19:48,327][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:19:48,829][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:19:49,334][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:19:49,834][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:19:50,333][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:19:50,831][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:19:51,329][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:19:51,825][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:19:52,322][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:19:52,819][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:19:53,316][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:19:53,813][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:19:54,309][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:19:54,806][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:19:55,301][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:19:55,799][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:19:56,295][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:19:56,792][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:19:57,290][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:19:57,786][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:19:58,286][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:19:58,786][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:19:59,282][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:19:59,779][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:20:00,277][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:20:00,777][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:20:01,274][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:20:01,772][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:20:02,272][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:20:02,771][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:20:03,269][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:20:03,770][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:20:04,268][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:20:04,769][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:20:05,266][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:20:05,766][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:20:06,266][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:20:06,764][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10147 tokens. +[2026-03-26 03:20:07,534][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.33%, ΔTime: 00:00:32 +[2026-03-26 03:20:08,309][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:20:08,314][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:20:08,317][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:20:10,490][__main__][INFO] - Iteration 361 took 51s (28.42% Gen, 67.39% Train). Generation: 14s, Training: 34s. Estimated remaining time: 37h 57m 3s. Estimated total time: 43h 16m 25s. Time estimates for 10 more iterations: 8m 39s, 100 more iterations: 1h 26m 32s, 500 more iterations: 7h 12m 44s. +[2026-03-26 03:20:10,493][__main__][INFO] - Starting iteration 361. +[2026-03-26 03:20:10,894][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 36 and human policies 1. +[2026-03-26 03:20:10,895][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:20:27,731][__main__][INFO] - Number of regex retries in iteration 361: 0 +[2026-03-26 03:20:27,732][__main__][INFO] - agents played in iteration 361 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:20:28,498][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:20:28,522][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:20:28,545][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:20:28,568][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:20:28,568][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:20:28,569][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:20:29,276][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:20:29,736][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:20:30,242][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:20:30,742][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:20:31,240][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:20:31,738][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:20:32,241][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:20:32,741][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:20:33,240][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:20:33,743][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:20:34,243][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:20:34,747][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:20:35,247][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:20:35,749][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:20:36,245][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:20:36,742][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:20:37,240][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:20:37,737][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:20:38,238][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:20:38,735][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:20:39,233][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:20:39,737][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:20:40,243][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:20:40,747][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:20:41,251][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:20:41,747][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:20:42,245][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:20:42,744][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:20:43,241][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:20:43,739][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:20:44,237][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:20:44,737][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:20:45,238][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:20:45,736][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:20:46,233][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:20:46,731][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:20:47,228][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:20:47,725][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:20:48,223][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:20:48,725][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:20:49,221][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:20:49,722][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:20:50,222][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:20:50,722][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:20:51,221][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:20:51,717][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:20:52,214][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:20:52,714][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:20:53,210][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:20:53,707][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:20:54,207][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:20:54,706][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:20:55,203][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:20:55,701][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:20:56,198][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:20:56,697][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:20:57,197][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:20:57,695][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:20:58,195][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:20:58,696][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:20:59,196][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:20:59,698][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:21:00,199][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:21:00,697][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:21:01,219][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10167 tokens. +[2026-03-26 03:21:01,962][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.25%, Current % of VRAM taken: 60.68%, Block Peak % of device VRAM: 62.15%, ΔTime: 00:00:32 +[2026-03-26 03:21:02,728][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:21:02,730][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:21:02,732][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:21:03,440][__main__][INFO] - Iteration 362 took 52s (32.04% Gen, 66.61% Train). Generation: 16s, Training: 34s. Estimated remaining time: 38h 27m 4s. Estimated total time: 43h 47m 18s. Time estimates for 10 more iterations: 8m 45s, 100 more iterations: 1h 27m 34s, 500 more iterations: 7h 17m 53s. +[2026-03-26 03:21:03,442][__main__][INFO] - Starting iteration 362. +[2026-03-26 03:21:03,841][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 36 and human policies 1. +[2026-03-26 03:21:03,842][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:21:19,657][__main__][INFO] - Number of regex retries in iteration 362: 0 +[2026-03-26 03:21:19,658][__main__][INFO] - agents played in iteration 362 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:21:20,423][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:21:20,446][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:21:20,469][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:21:20,493][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:21:20,493][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:21:20,494][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:21:21,214][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:21:21,668][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:21:22,170][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:21:22,667][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:21:23,163][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:21:23,665][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:21:24,164][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:21:24,661][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:21:25,160][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:21:25,660][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:21:26,159][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:21:26,657][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:21:27,156][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:21:27,654][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:21:28,151][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:21:28,653][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:21:29,154][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:21:29,661][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:21:30,159][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:21:30,657][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:21:31,154][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:21:31,651][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:21:32,153][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:21:32,651][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:21:33,148][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:21:33,647][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:21:34,147][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:21:34,645][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:21:35,140][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:21:35,637][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:21:36,134][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:21:36,630][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:21:37,126][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:21:37,624][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:21:38,120][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:21:38,617][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:21:39,115][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:21:39,613][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:21:40,111][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:21:40,608][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:21:41,108][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:21:41,604][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:21:42,100][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:21:42,596][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:21:43,093][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:21:43,590][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:21:44,086][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:21:44,586][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:21:45,082][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:21:45,579][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:21:46,077][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:21:46,576][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:21:47,074][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:21:47,572][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:21:48,071][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:21:48,574][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:21:49,071][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:21:49,572][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:21:50,069][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:21:50,565][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:21:51,062][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:21:51,558][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:21:52,054][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:21:52,550][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:21:53,050][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10032 tokens. +[2026-03-26 03:21:53,800][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.15%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.02%, ΔTime: 00:00:32 +[2026-03-26 03:21:54,568][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:21:54,570][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:21:54,572][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:21:55,292][__main__][INFO] - Iteration 363 took 51s (30.74% Gen, 67.86% Train). Generation: 15s, Training: 34s. Estimated remaining time: 37h 31m 30s. Estimated total time: 42h 52m 36s. Time estimates for 10 more iterations: 8m 34s, 100 more iterations: 1h 25m 45s, 500 more iterations: 7h 8m 46s. +[2026-03-26 03:21:55,297][__main__][INFO] - Starting iteration 363. +[2026-03-26 03:21:55,697][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 36 and human policies 1. +[2026-03-26 03:21:55,698][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:22:12,225][__main__][INFO] - Number of regex retries in iteration 363: 0 +[2026-03-26 03:22:12,226][__main__][INFO] - agents played in iteration 363 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:22:13,000][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:22:13,023][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:22:13,046][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:22:13,070][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:22:13,070][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:22:13,071][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:22:13,781][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:22:14,241][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:22:14,742][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:22:15,245][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:22:15,745][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:22:16,242][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:22:16,740][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:22:17,237][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:22:17,746][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:22:18,248][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:22:18,754][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:22:19,257][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:22:19,757][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:22:20,256][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:22:20,760][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:22:21,258][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:22:21,757][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:22:22,255][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:22:22,757][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:22:23,257][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:22:23,755][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:22:24,258][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:22:24,759][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:22:25,259][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:22:25,756][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:22:26,254][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:22:26,750][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:22:27,247][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:22:27,744][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:22:28,241][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:22:28,738][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:22:29,236][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:22:29,734][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:22:30,235][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:22:30,735][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:22:31,234][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:22:31,730][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:22:32,227][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:22:32,723][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:22:33,219][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:22:33,715][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:22:34,215][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:22:34,714][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:22:35,213][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:22:35,713][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:22:36,211][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:22:36,709][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:22:37,206][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:22:37,704][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:22:38,201][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:22:38,699][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:22:39,197][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:22:39,695][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:22:40,194][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:22:40,694][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:22:41,196][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:22:41,693][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:22:42,192][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:22:42,691][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:22:43,190][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:22:43,688][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:22:44,185][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:22:44,682][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:22:45,180][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:22:45,678][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10083 tokens. +[2026-03-26 03:22:46,445][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.07%, ΔTime: 00:00:32 +[2026-03-26 03:22:47,258][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:22:47,261][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:22:47,263][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:22:48,193][__main__][INFO] - Iteration 364 took 52s (31.48% Gen, 66.74% Train). Generation: 16s, Training: 35s. Estimated remaining time: 38h 22m 51s. Estimated total time: 43h 44m 50s. Time estimates for 10 more iterations: 8m 44s, 100 more iterations: 1h 27m 29s, 500 more iterations: 7h 17m 28s. +[2026-03-26 03:22:48,195][__main__][INFO] - Starting iteration 364. +[2026-03-26 03:22:48,598][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 36 and human policies 1. +[2026-03-26 03:22:48,598][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:23:15,775][__main__][INFO] - Number of regex retries in iteration 364: 0 +[2026-03-26 03:23:15,776][__main__][INFO] - agents played in iteration 364 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:23:16,542][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:23:16,565][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:23:16,588][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:23:16,611][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:23:16,612][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:23:16,612][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:23:17,344][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:23:17,798][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:23:18,298][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:23:18,796][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:23:19,293][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:23:19,791][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:23:20,289][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:23:20,787][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:23:21,284][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:23:21,783][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:23:22,282][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:23:22,779][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:23:23,276][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:23:23,774][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:23:24,274][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:23:24,773][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:23:25,269][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:23:25,772][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:23:26,273][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:23:26,772][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:23:27,272][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:23:27,773][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:23:28,271][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:23:28,768][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:23:29,268][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:23:29,764][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:23:30,262][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:23:30,758][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:23:31,254][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:23:31,751][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:23:32,246][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:23:32,741][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:23:33,238][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:23:33,742][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:23:34,243][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:23:34,744][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:23:35,245][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:23:35,745][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:23:36,266][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:23:36,772][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:23:37,273][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:23:37,772][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:23:38,270][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:23:38,769][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:23:39,267][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:23:39,765][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:23:40,266][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:23:40,764][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:23:41,260][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:23:41,760][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:23:42,257][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:23:42,756][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:23:43,253][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:23:43,749][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:23:44,246][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:23:44,743][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:23:45,239][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:23:45,736][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:23:46,234][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:23:46,732][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:23:47,232][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:23:47,738][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:23:48,242][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:23:48,743][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:23:49,244][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10114 tokens. +[2026-03-26 03:23:50,008][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.27%, Current % of VRAM taken: 60.70%, Block Peak % of device VRAM: 62.42%, ΔTime: 00:00:32 +[2026-03-26 03:23:50,803][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:23:50,805][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:23:50,807][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:23:51,730][__main__][INFO] - Iteration 365 took 1m 3s (43.05% Gen, 55.49% Train). Generation: 27s, Training: 35s. Estimated remaining time: 47h 13m 37s. Estimated total time: 52h 36m 40s. Time estimates for 10 more iterations: 10m 31s, 100 more iterations: 1h 45m 13s, 500 more iterations: 8h 46m 6s. +[2026-03-26 03:23:51,733][__main__][INFO] - Starting iteration 365. +[2026-03-26 03:23:52,131][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 36 and human policies 1. +[2026-03-26 03:23:52,132][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:24:09,277][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 03:24:16,390][__main__][INFO] - Number of regex retries in iteration 365: 1 +[2026-03-26 03:24:16,390][__main__][INFO] - agents played in iteration 365 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:24:17,170][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.60%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:24:17,194][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.60%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:24:17,217][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.60%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:24:17,241][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.60%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:24:17,242][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:24:17,242][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:24:17,950][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:24:18,402][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:24:18,906][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:24:19,404][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:24:19,901][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:24:20,402][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:24:20,900][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:24:21,399][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:24:21,897][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:24:22,398][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:24:22,900][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:24:23,395][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:24:23,893][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:24:24,393][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:24:24,897][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:24:25,395][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:24:25,893][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:24:26,391][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:24:26,892][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:24:27,388][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:24:27,884][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:24:28,381][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:24:28,878][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:24:29,377][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:24:29,873][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:24:30,370][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:24:30,873][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:24:31,373][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:24:31,871][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:24:32,374][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:24:32,876][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:24:33,373][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:24:33,870][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:24:34,367][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:24:35,028][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:24:35,525][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:24:36,025][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:24:36,524][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:24:37,022][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:24:37,519][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:24:38,016][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:24:38,516][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:24:39,014][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:24:39,513][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:24:40,013][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:24:40,514][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:24:41,012][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:24:41,510][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:24:42,010][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:24:42,505][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:24:43,002][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:24:43,500][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:24:44,003][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:24:44,507][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:24:45,009][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:24:45,507][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:24:46,006][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:24:46,506][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:24:47,006][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:24:47,506][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:24:48,007][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:24:48,510][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:24:49,006][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:24:49,506][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:24:50,007][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10105 tokens. +[2026-03-26 03:24:50,772][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 62.09%, ΔTime: 00:00:32 +[2026-03-26 03:24:51,533][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:24:51,536][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:24:51,537][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:24:52,264][__main__][INFO] - Iteration 366 took 1m 0s (40.34% Gen, 58.45% Train). Generation: 24s, Training: 35s. Estimated remaining time: 44h 42m 36s. Estimated total time: 50h 6m 40s. Time estimates for 10 more iterations: 10m 1s, 100 more iterations: 1h 40m 13s, 500 more iterations: 8h 21m 6s. +[2026-03-26 03:24:52,266][__main__][INFO] - Starting iteration 366. +[2026-03-26 03:24:52,669][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 36 and human policies 1. +[2026-03-26 03:24:52,670][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:25:15,005][__main__][INFO] - Number of regex retries in iteration 366: 0 +[2026-03-26 03:25:15,006][__main__][INFO] - agents played in iteration 366 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:25:15,779][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:25:15,803][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:25:15,826][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:25:15,849][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:25:15,849][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:25:15,850][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:25:16,556][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:25:17,018][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:25:17,519][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:25:18,025][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:25:18,524][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:25:19,022][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:25:19,526][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:25:20,027][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:25:20,527][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:25:21,032][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:25:21,532][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:25:22,032][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:25:22,533][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:25:23,033][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:25:23,530][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:25:24,031][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:25:24,533][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:25:25,035][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:25:25,532][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:25:26,036][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:25:26,535][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:25:27,035][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:25:27,534][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:25:28,032][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:25:28,531][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:25:29,034][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:25:29,533][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:25:30,035][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:25:30,532][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:25:31,031][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:25:31,532][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:25:32,032][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:25:32,532][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:25:33,030][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:25:33,528][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:25:34,029][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:25:34,530][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:25:35,035][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:25:35,536][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:25:36,033][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:25:36,532][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:25:37,030][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:25:37,528][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:25:38,032][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:25:38,531][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:25:39,030][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:25:39,538][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:25:40,036][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:25:40,536][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:25:41,042][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:25:41,549][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:25:42,055][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:25:42,560][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:25:43,065][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:25:43,569][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:25:44,074][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:25:44,579][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:25:45,087][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:25:45,591][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:25:46,092][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:25:46,594][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:25:47,094][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:25:47,594][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:25:48,095][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:25:48,597][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10309 tokens. +[2026-03-26 03:25:49,347][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.19%, Current % of VRAM taken: 60.62%, Block Peak % of device VRAM: 62.47%, ΔTime: 00:00:32 +[2026-03-26 03:25:50,097][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:25:50,100][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:25:50,101][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:25:50,894][__main__][INFO] - Iteration 367 took 58s (38.36% Gen, 60.28% Train). Generation: 22s, Training: 35s. Estimated remaining time: 43h 6m 14s. Estimated total time: 48h 31m 16s. Time estimates for 10 more iterations: 9m 42s, 100 more iterations: 1h 37m 2s, 500 more iterations: 8h 5m 12s. +[2026-03-26 03:25:50,896][__main__][INFO] - Starting iteration 367. +[2026-03-26 03:25:51,336][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 36 and human policies 1. +[2026-03-26 03:25:51,337][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:26:08,795][__main__][INFO] - Number of regex retries in iteration 367: 0 +[2026-03-26 03:26:08,796][__main__][INFO] - agents played in iteration 367 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:26:09,562][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:26:09,586][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:26:09,609][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:26:09,632][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:26:09,633][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:26:09,633][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:26:10,352][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:26:10,809][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:26:11,309][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:26:11,808][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:26:12,306][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:26:12,806][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:26:13,303][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:26:13,799][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:26:14,303][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:26:14,801][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:26:15,301][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:26:15,800][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:26:16,303][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:26:16,802][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:26:17,303][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:26:17,803][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:26:18,305][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:26:18,805][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:26:19,305][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:26:19,802][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:26:20,304][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:26:20,804][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:26:21,310][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:26:21,814][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:26:22,314][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:26:22,810][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:26:23,307][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:26:23,804][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:26:24,300][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:26:24,798][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:26:25,296][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:26:25,793][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:26:26,291][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:26:26,792][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:26:27,292][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:26:27,794][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:26:28,293][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:26:28,794][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:26:29,293][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:26:29,791][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:26:30,288][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:26:30,787][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:26:31,292][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:26:31,793][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:26:32,292][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:26:32,790][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:26:33,287][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:26:33,787][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:26:34,285][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:26:34,785][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:26:35,282][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:26:35,782][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:26:36,279][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:26:36,777][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:26:37,278][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:26:37,780][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:26:38,284][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:26:38,791][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:26:39,297][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:26:39,795][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:26:40,295][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:26:40,799][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:26:41,302][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:26:41,805][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:26:42,308][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10254 tokens. +[2026-03-26 03:26:43,076][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.21%, Current % of VRAM taken: 60.63%, Block Peak % of device VRAM: 62.32%, ΔTime: 00:00:32 +[2026-03-26 03:26:43,832][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:26:43,834][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:26:43,836][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:26:44,551][__main__][INFO] - Iteration 368 took 53s (32.81% Gen, 65.85% Train). Generation: 17s, Training: 35s. Estimated remaining time: 38h 54m 49s. Estimated total time: 44h 20m 45s. Time estimates for 10 more iterations: 8m 52s, 100 more iterations: 1h 28m 41s, 500 more iterations: 7h 23m 27s. +[2026-03-26 03:26:44,553][__main__][INFO] - Starting iteration 368. +[2026-03-26 03:26:44,955][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 36 and human policies 1. +[2026-03-26 03:26:44,955][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:26:49,596][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 03:27:01,338][__main__][INFO] - Number of regex retries in iteration 368: 1 +[2026-03-26 03:27:01,339][__main__][INFO] - agents played in iteration 368 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:27:02,106][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:27:02,129][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:27:02,153][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:27:02,175][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:27:02,176][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:27:02,177][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:27:02,900][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:27:03,351][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:27:03,856][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:27:04,352][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:27:04,851][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:27:05,347][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:27:05,843][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:27:06,340][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:27:06,839][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:27:07,336][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:27:07,853][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:27:08,349][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:27:08,852][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:27:09,355][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:27:09,855][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:27:10,361][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:27:10,861][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:27:11,368][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:27:11,868][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:27:12,365][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:27:12,862][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:27:13,362][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:27:13,859][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:27:14,358][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:27:14,855][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:27:15,355][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:27:15,855][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:27:16,355][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:27:16,854][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:27:17,354][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:27:17,853][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:27:18,353][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:27:18,849][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:27:19,350][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:27:19,846][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:27:20,346][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:27:20,843][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:27:21,340][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:27:21,838][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:27:22,336][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:27:22,833][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:27:23,332][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:27:23,829][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:27:24,329][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:27:24,826][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:27:25,326][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:27:25,826][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:27:26,327][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:27:26,823][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:27:27,320][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:27:27,818][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:27:28,316][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:27:28,812][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:27:29,313][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:27:29,809][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:27:30,311][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:27:30,808][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:27:31,308][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:27:31,808][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:27:32,309][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:27:32,812][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:27:33,315][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:27:33,822][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:27:34,325][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:27:34,823][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10079 tokens. +[2026-03-26 03:27:35,611][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.09%, ΔTime: 00:00:32 +[2026-03-26 03:27:36,479][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:27:36,481][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:27:36,483][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:27:37,249][__main__][INFO] - Iteration 369 took 52s (31.33% Gen, 67.20% Train). Generation: 16s, Training: 35s. Estimated remaining time: 38h 7m 57s. Estimated total time: 43h 34m 46s. Time estimates for 10 more iterations: 8m 42s, 100 more iterations: 1h 27m 9s, 500 more iterations: 7h 15m 47s. +[2026-03-26 03:27:37,252][__main__][INFO] - Starting iteration 369. +[2026-03-26 03:27:37,651][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 36 and human policies 1. +[2026-03-26 03:27:37,652][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:28:04,319][__main__][INFO] - Number of regex retries in iteration 369: 0 +[2026-03-26 03:28:04,320][__main__][INFO] - agents played in iteration 369 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:28:05,100][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:28:05,123][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:28:05,147][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:28:05,170][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:28:05,171][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:28:05,171][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:28:05,900][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:28:06,352][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:28:06,858][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:28:07,357][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:28:07,854][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:28:08,359][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:28:08,858][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:28:09,357][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:28:09,855][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:28:10,353][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:28:10,859][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:28:11,361][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:28:11,859][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:28:12,357][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:28:12,854][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:28:13,351][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:28:13,848][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:28:14,345][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:28:14,848][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:28:15,346][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:28:15,844][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:28:16,340][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:28:16,837][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:28:17,335][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:28:17,839][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:28:18,338][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:28:18,836][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:28:19,335][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:28:19,835][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:28:20,335][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:28:20,834][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:28:21,333][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:28:21,832][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:28:22,336][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:28:22,836][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:28:23,341][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:28:23,841][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:28:24,340][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:28:24,847][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:28:25,351][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:28:25,852][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:28:26,354][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:28:26,856][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:28:27,352][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:28:27,854][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:28:28,354][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:28:28,855][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:28:29,352][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:28:29,855][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:28:30,355][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:28:30,857][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:28:31,357][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:28:31,861][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:28:32,360][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:28:32,862][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:28:33,363][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:28:33,863][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:28:34,360][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:28:34,857][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:28:35,354][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:28:35,850][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:28:36,347][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:28:36,846][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:28:37,343][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:28:37,839][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10206 tokens. +[2026-03-26 03:28:38,598][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.33%, ΔTime: 00:00:32 +[2026-03-26 03:28:39,360][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:28:39,363][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:28:39,364][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:28:40,088][__main__][INFO] - Iteration 370 took 1m 2s (42.71% Gen, 56.13% Train). Generation: 26s, Training: 35s. Estimated remaining time: 46h 34m 2s. Estimated total time: 52h 1m 53s. Time estimates for 10 more iterations: 10m 24s, 100 more iterations: 1h 44m 3s, 500 more iterations: 8h 40m 18s. +[2026-03-26 03:28:40,090][__main__][INFO] - Starting iteration 370. +[2026-03-26 03:28:40,493][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 36 and human policies 1. +[2026-03-26 03:28:40,494][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:29:01,896][__main__][INFO] - Number of regex retries in iteration 370: 0 +[2026-03-26 03:29:01,897][__main__][INFO] - agents played in iteration 370 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:29:02,705][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:29:02,729][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:29:02,752][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:29:02,775][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:29:02,776][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:29:02,776][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:29:03,481][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:29:03,933][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:29:04,437][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:29:04,943][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:29:05,441][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:29:05,940][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:29:06,439][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:29:06,937][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:29:07,435][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:29:07,936][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:29:08,433][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:29:08,935][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:29:09,432][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:29:09,931][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:29:10,429][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:29:10,925][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:29:11,421][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:29:11,925][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:29:12,422][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:29:12,918][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:29:13,414][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:29:13,912][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:29:14,408][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:29:14,910][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:29:15,411][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:29:15,921][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:29:16,425][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:29:16,931][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:29:17,431][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:29:17,932][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:29:18,463][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:29:18,961][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:29:19,462][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:29:19,962][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:29:20,460][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:29:20,961][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:29:21,457][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:29:21,954][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:29:22,457][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:29:22,959][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:29:23,461][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:29:23,958][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:29:24,455][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:29:24,953][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:29:25,450][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:29:25,946][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:29:26,442][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:29:26,940][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:29:27,439][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:29:27,944][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:29:28,442][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:29:28,941][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:29:29,441][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:29:29,940][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:29:30,438][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:29:30,940][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:29:31,441][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:29:31,941][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:29:32,441][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:29:32,941][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:29:33,442][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:29:33,947][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:29:34,447][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:29:34,946][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:29:35,446][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10214 tokens. +[2026-03-26 03:29:36,191][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.21%, Current % of VRAM taken: 60.63%, Block Peak % of device VRAM: 62.36%, ΔTime: 00:00:32 +[2026-03-26 03:29:36,958][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:29:36,961][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:29:36,962][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:29:38,503][__main__][INFO] - Iteration 371 took 58s (36.90% Gen, 60.45% Train). Generation: 21s, Training: 35s. Estimated remaining time: 42h 51m 41s. Estimated total time: 48h 20m 31s. Time estimates for 10 more iterations: 9m 40s, 100 more iterations: 1h 36m 41s, 500 more iterations: 8h 3m 25s. +[2026-03-26 03:29:38,505][__main__][INFO] - Starting iteration 371. +[2026-03-26 03:29:38,904][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 37 and human policies 1. +[2026-03-26 03:29:38,904][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:29:55,356][__main__][INFO] - Number of regex retries in iteration 371: 0 +[2026-03-26 03:29:55,357][__main__][INFO] - agents played in iteration 371 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:29:56,145][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:29:56,169][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:29:56,192][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:29:56,215][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:29:56,216][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:29:56,216][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:29:56,950][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:29:57,405][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:29:57,912][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:29:58,415][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:29:58,918][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:29:59,417][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:29:59,917][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:30:00,419][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:30:00,917][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:30:01,428][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:30:01,932][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:30:02,433][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:30:02,932][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:30:03,430][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:30:03,930][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:30:04,431][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:30:04,931][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:30:05,435][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:30:05,933][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:30:06,432][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:30:06,940][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:30:07,442][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:30:07,948][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:30:08,445][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:30:08,945][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:30:09,446][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:30:09,953][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:30:10,455][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:30:10,957][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:30:11,458][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:30:11,957][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:30:12,458][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:30:12,957][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:30:13,455][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:30:13,952][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:30:14,449][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:30:14,949][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:30:15,456][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:30:15,954][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:30:16,453][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:30:16,951][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:30:17,450][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:30:17,947][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:30:18,445][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:30:18,942][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:30:19,440][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:30:19,938][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:30:20,438][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:30:20,935][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:30:21,436][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:30:21,943][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:30:22,443][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:30:22,941][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:30:23,442][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:30:23,942][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:30:24,441][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:30:24,941][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:30:25,440][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:30:25,939][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:30:26,437][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:30:26,936][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:30:27,436][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:30:27,936][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:30:28,434][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:30:28,932][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10192 tokens. +[2026-03-26 03:30:29,704][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.23%, ΔTime: 00:00:32 +[2026-03-26 03:30:30,472][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:30:30,474][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:30:30,476][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:30:31,199][__main__][INFO] - Iteration 372 took 52s (31.46% Gen, 67.15% Train). Generation: 16s, Training: 35s. Estimated remaining time: 38h 5m 5s. Estimated total time: 43h 34m 48s. Time estimates for 10 more iterations: 8m 42s, 100 more iterations: 1h 27m 9s, 500 more iterations: 7h 15m 48s. +[2026-03-26 03:30:31,201][__main__][INFO] - Starting iteration 372. +[2026-03-26 03:30:31,601][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 37 and human policies 1. +[2026-03-26 03:30:31,601][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:30:50,037][__main__][INFO] - Number of regex retries in iteration 372: 0 +[2026-03-26 03:30:50,038][__main__][INFO] - agents played in iteration 372 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:30:50,806][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:30:50,830][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:30:50,853][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:30:50,876][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:30:50,877][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:30:50,878][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:30:51,595][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:30:52,053][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:30:52,552][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:30:53,054][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:30:53,550][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:30:54,052][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:30:54,560][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:30:55,062][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:30:55,564][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:30:56,063][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:30:56,560][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:30:57,059][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:30:57,557][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:30:58,055][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:30:58,552][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:30:59,054][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:30:59,558][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:31:00,055][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:31:00,556][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:31:01,056][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:31:01,553][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:31:02,056][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:31:02,559][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:31:03,057][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:31:03,555][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:31:04,057][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:31:04,554][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:31:05,052][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:31:05,548][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:31:06,046][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:31:06,548][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:31:07,047][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:31:07,548][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:31:08,046][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:31:08,543][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:31:09,040][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:31:09,537][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:31:10,034][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:31:10,535][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:31:11,031][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:31:11,530][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:31:12,027][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:31:12,525][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:31:13,022][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:31:13,520][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:31:14,021][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:31:14,522][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:31:15,022][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:31:15,521][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:31:16,023][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:31:16,531][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:31:17,028][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:31:17,530][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:31:18,036][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:31:18,538][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:31:19,039][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:31:19,539][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:31:20,038][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:31:20,538][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:31:21,036][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:31:21,535][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:31:22,035][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:31:22,533][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:31:23,041][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:31:23,538][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10268 tokens. +[2026-03-26 03:31:24,296][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.17%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.25%, ΔTime: 00:00:32 +[2026-03-26 03:31:25,046][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:31:25,048][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:31:25,050][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:31:25,805][__main__][INFO] - Iteration 373 took 54s (34.01% Gen, 64.59% Train). Generation: 18s, Training: 35s. Estimated remaining time: 39h 39m 38s. Estimated total time: 45h 10m 15s. Time estimates for 10 more iterations: 9m 2s, 100 more iterations: 1h 30m 20s, 500 more iterations: 7h 31m 42s. +[2026-03-26 03:31:25,807][__main__][INFO] - Starting iteration 373. +[2026-03-26 03:31:26,205][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 37 and human policies 1. +[2026-03-26 03:31:26,206][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:31:43,756][__main__][INFO] - Number of regex retries in iteration 373: 0 +[2026-03-26 03:31:43,757][__main__][INFO] - agents played in iteration 373 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:31:44,526][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:31:44,549][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:31:44,573][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:31:44,596][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:31:44,596][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:31:44,597][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:31:45,325][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:31:45,779][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:31:46,289][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:31:46,786][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:31:47,287][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:31:47,786][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:31:48,291][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:31:48,789][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:31:49,287][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:31:49,784][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:31:50,284][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:31:50,784][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:31:51,283][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:31:51,786][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:31:52,284][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:31:52,782][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:31:53,280][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:31:53,779][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:31:54,278][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:31:54,777][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:31:55,281][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:31:55,777][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:31:56,275][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:31:56,776][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:31:57,274][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:31:57,772][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:31:58,272][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:31:58,770][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:31:59,268][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:31:59,765][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:32:00,262][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:32:00,760][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:32:01,259][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:32:01,760][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:32:02,259][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:32:02,758][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:32:03,257][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:32:03,755][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:32:04,253][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:32:04,765][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:32:05,268][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:32:05,772][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:32:06,268][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:32:06,766][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:32:07,270][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:32:07,772][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:32:08,270][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:32:08,773][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:32:09,274][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:32:09,770][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:32:10,268][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:32:10,764][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:32:11,260][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:32:11,758][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:32:12,255][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:32:12,752][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:32:13,251][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:32:13,751][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:32:14,248][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:32:14,748][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:32:15,245][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:32:15,743][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:32:16,241][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:32:16,738][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:32:17,236][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10055 tokens. +[2026-03-26 03:32:18,002][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 62.07%, ΔTime: 00:00:32 +[2026-03-26 03:32:18,824][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:32:18,827][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:32:18,828][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:32:19,551][__main__][INFO] - Iteration 374 took 53s (32.90% Gen, 65.74% Train). Generation: 17s, Training: 35s. Estimated remaining time: 38h 55m 49s. Estimated total time: 44h 27m 19s. Time estimates for 10 more iterations: 8m 53s, 100 more iterations: 1h 28m 54s, 500 more iterations: 7h 24m 33s. +[2026-03-26 03:32:19,553][__main__][INFO] - Starting iteration 374. +[2026-03-26 03:32:19,960][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 37 and human policies 1. +[2026-03-26 03:32:19,961][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:32:37,447][__main__][INFO] - Number of regex retries in iteration 374: 0 +[2026-03-26 03:32:37,447][__main__][INFO] - agents played in iteration 374 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:32:38,217][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:32:38,241][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:32:38,264][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:32:38,288][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:32:38,288][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:32:38,289][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:32:39,020][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:32:39,474][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:32:39,975][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:32:40,483][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:32:40,981][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:32:41,479][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:32:41,977][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:32:42,480][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:32:42,977][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:32:43,475][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:32:43,974][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:32:44,477][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:32:44,980][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:32:45,478][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:32:45,974][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:32:46,470][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:32:46,974][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:32:47,472][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:32:47,974][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:32:48,475][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:32:48,973][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:32:49,471][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:32:49,970][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:32:50,471][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:32:50,970][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:32:51,472][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:32:51,974][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:32:52,495][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:32:52,999][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:32:53,501][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:32:54,004][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:32:54,508][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:32:55,014][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:32:55,516][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:32:56,017][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:32:56,514][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:32:57,011][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:32:57,509][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:32:58,005][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:32:58,507][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:32:59,006][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:32:59,507][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:33:00,010][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:33:00,505][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:33:01,006][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:33:01,506][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:33:02,005][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:33:02,505][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:33:03,005][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:33:03,504][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:33:04,001][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:33:04,504][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:33:05,002][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:33:05,502][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:33:06,000][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:33:06,501][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:33:07,001][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:33:07,501][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:33:08,001][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:33:08,502][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:33:09,002][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:33:09,502][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:33:10,002][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:33:10,501][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:33:11,002][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10300 tokens. +[2026-03-26 03:33:11,742][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.23%, Current % of VRAM taken: 60.66%, Block Peak % of device VRAM: 62.23%, ΔTime: 00:00:32 +[2026-03-26 03:33:12,561][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:33:12,564][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:33:12,566][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:33:13,277][__main__][INFO] - Iteration 375 took 53s (32.80% Gen, 65.87% Train). Generation: 17s, Training: 35s. Estimated remaining time: 38h 53m 26s. Estimated total time: 44h 25m 51s. Time estimates for 10 more iterations: 8m 53s, 100 more iterations: 1h 28m 51s, 500 more iterations: 7h 24m 18s. +[2026-03-26 03:33:13,279][__main__][INFO] - Starting iteration 375. +[2026-03-26 03:33:13,678][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 37 and human policies 1. +[2026-03-26 03:33:13,679][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:33:18,905][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 03:33:31,522][__main__][INFO] - Number of regex retries in iteration 375: 1 +[2026-03-26 03:33:31,523][__main__][INFO] - agents played in iteration 375 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:33:32,294][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:33:32,318][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:33:32,341][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:33:32,364][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:33:32,365][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:33:32,366][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:33:33,088][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:33:33,543][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:33:34,043][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:33:34,542][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:33:35,039][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:33:35,535][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:33:36,035][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:33:36,542][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:33:37,039][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:33:37,537][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:33:38,035][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:33:38,538][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:33:39,039][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:33:39,538][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:33:40,037][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:33:40,540][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:33:41,037][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:33:41,560][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:33:42,057][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:33:42,554][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:33:43,055][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:33:43,554][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:33:44,052][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:33:44,552][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:33:45,062][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:33:45,564][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:33:46,068][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:33:46,565][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:33:47,064][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:33:47,562][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:33:48,060][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:33:48,562][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:33:49,064][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:33:49,561][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:33:50,064][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:33:50,563][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:33:51,061][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:33:51,558][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:33:52,056][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:33:52,554][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:33:53,053][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:33:53,554][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:33:54,053][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:33:54,553][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:33:55,052][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:33:55,553][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:33:56,053][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:33:56,553][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:33:57,052][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:33:57,553][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:33:58,051][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:33:58,548][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:33:59,046][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:33:59,547][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:34:00,046][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:34:00,547][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:34:01,048][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:34:01,545][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:34:02,042][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:34:02,540][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:34:03,038][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:34:03,537][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:34:04,034][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:34:04,532][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:34:05,031][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10116 tokens. +[2026-03-26 03:34:05,782][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.16%, Current % of VRAM taken: 60.58%, Block Peak % of device VRAM: 62.13%, ΔTime: 00:00:32 +[2026-03-26 03:34:06,584][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:34:06,586][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:34:06,588][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:34:07,316][__main__][INFO] - Iteration 376 took 53s (33.27% Gen, 65.37% Train). Generation: 17s, Training: 35s. Estimated remaining time: 39h 8m 38s. Estimated total time: 44h 41m 56s. Time estimates for 10 more iterations: 8m 56s, 100 more iterations: 1h 29m 23s, 500 more iterations: 7h 26m 59s. +[2026-03-26 03:34:07,318][__main__][INFO] - Starting iteration 376. +[2026-03-26 03:34:07,719][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 37 and human policies 1. +[2026-03-26 03:34:07,720][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:34:22,450][__main__][INFO] - Number of regex retries in iteration 376: 0 +[2026-03-26 03:34:22,450][__main__][INFO] - agents played in iteration 376 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:34:23,216][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:34:23,240][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:34:23,263][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:34:23,286][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:34:23,287][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:34:23,287][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:34:23,999][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:34:24,452][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:34:24,953][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:34:25,452][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:34:25,952][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:34:26,454][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:34:26,953][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:34:27,452][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:34:27,950][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:34:28,455][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:34:28,953][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:34:29,454][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:34:29,971][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:34:30,474][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:34:30,977][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:34:31,476][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:34:31,977][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:34:32,475][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:34:32,977][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:34:33,481][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:34:33,980][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:34:34,493][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:34:34,995][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:34:35,495][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:34:35,996][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:34:36,495][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:34:36,994][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:34:37,495][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:34:37,995][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:34:38,497][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:34:38,995][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:34:39,495][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:34:39,993][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:34:40,491][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:34:40,988][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:34:41,485][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:34:41,982][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:34:42,479][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:34:42,976][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:34:43,474][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:34:43,971][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:34:44,473][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:34:44,974][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:34:45,474][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:34:45,977][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:34:46,474][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:34:46,978][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:34:47,480][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:34:47,981][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:34:48,478][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:34:48,976][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:34:49,478][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:34:49,977][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:34:50,475][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:34:50,973][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:34:51,471][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:34:51,968][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:34:52,464][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:34:52,963][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:34:53,459][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:34:53,956][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:34:54,453][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:34:54,954][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:34:55,451][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:34:55,947][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10148 tokens. +[2026-03-26 03:34:56,710][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.29%, ΔTime: 00:00:32 +[2026-03-26 03:34:57,480][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:34:57,482][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:34:57,484][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:34:58,165][__main__][INFO] - Iteration 377 took 50s (29.20% Gen, 69.45% Train). Generation: 14s, Training: 35s. Estimated remaining time: 36h 28m 8s. Estimated total time: 42h 2m 18s. Time estimates for 10 more iterations: 8m 24s, 100 more iterations: 1h 24m 4s, 500 more iterations: 7h 0m 23s. +[2026-03-26 03:34:58,167][__main__][INFO] - Starting iteration 377. +[2026-03-26 03:34:58,566][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 37 and human policies 1. +[2026-03-26 03:34:58,566][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:35:27,990][__main__][INFO] - Number of regex retries in iteration 377: 0 +[2026-03-26 03:35:27,991][__main__][INFO] - agents played in iteration 377 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:35:28,757][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:35:28,781][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:35:28,804][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:35:28,828][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:35:28,829][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:35:28,829][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:35:29,560][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:35:30,013][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:35:30,521][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:35:31,019][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:35:31,525][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:35:32,027][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:35:32,525][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:35:33,022][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:35:33,519][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:35:34,018][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:35:34,525][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:35:35,028][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:35:35,528][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:35:36,028][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:35:36,524][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:35:37,024][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:35:37,529][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:35:38,025][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:35:38,528][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:35:39,028][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:35:39,527][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:35:40,027][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:35:40,528][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:35:41,025][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:35:41,529][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:35:42,029][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:35:42,550][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:35:43,052][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:35:43,554][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:35:44,052][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:35:44,550][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:35:45,049][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:35:45,550][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:35:46,049][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:35:46,546][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:35:47,049][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:35:47,549][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:35:48,049][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:35:48,548][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:35:49,045][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:35:49,541][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:35:50,038][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:35:50,536][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:35:51,034][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:35:51,537][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:35:52,033][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:35:52,531][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:35:53,027][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:35:53,525][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:35:54,022][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:35:54,519][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:35:55,018][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:35:55,520][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:35:56,016][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:35:56,514][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:35:57,011][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:35:57,510][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:35:58,015][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:35:58,523][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:35:59,027][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:35:59,528][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:36:00,027][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:36:00,528][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:36:01,026][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:36:01,524][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10210 tokens. +[2026-03-26 03:36:02,274][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.16%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.20%, ΔTime: 00:00:32 +[2026-03-26 03:36:03,118][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:36:03,121][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:36:03,122][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:36:03,978][__main__][INFO] - Iteration 378 took 1m 5s (44.98% Gen, 53.71% Train). Generation: 29s, Training: 35s. Estimated remaining time: 48h 55m 23s. Estimated total time: 54h 30m 38s. Time estimates for 10 more iterations: 10m 54s, 100 more iterations: 1h 49m 1s, 500 more iterations: 9h 5m 6s. +[2026-03-26 03:36:03,980][__main__][INFO] - Starting iteration 378. +[2026-03-26 03:36:04,378][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 37 and human policies 1. +[2026-03-26 03:36:04,379][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:36:30,147][__main__][INFO] - Number of regex retries in iteration 378: 0 +[2026-03-26 03:36:30,148][__main__][INFO] - agents played in iteration 378 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:36:30,930][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:36:30,953][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:36:30,976][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:36:31,000][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:36:31,001][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:36:31,001][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:36:31,711][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:36:32,165][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:36:32,666][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:36:33,163][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:36:33,662][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:36:34,160][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:36:34,657][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:36:35,153][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:36:35,650][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:36:36,151][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:36:36,651][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:36:37,152][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:36:37,677][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:36:38,179][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:36:38,679][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:36:39,179][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:36:39,679][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:36:40,183][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:36:40,682][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:36:41,180][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:36:41,681][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:36:42,179][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:36:42,677][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:36:43,184][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:36:43,683][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:36:44,186][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:36:44,688][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:36:45,188][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:36:45,693][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:36:46,196][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:36:46,697][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:36:47,200][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:36:47,701][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:36:48,203][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:36:48,701][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:36:49,200][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:36:49,697][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:36:50,196][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:36:50,703][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:36:51,200][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:36:51,698][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:36:52,198][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:36:52,699][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:36:53,197][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:36:53,700][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:36:54,200][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:36:54,700][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:36:55,207][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:36:55,708][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:36:56,205][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:36:56,703][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:36:57,200][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:36:57,698][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:36:58,195][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:36:58,693][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:36:59,190][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:36:59,692][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:37:00,191][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:37:00,688][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:37:01,186][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:37:01,683][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:37:02,179][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:37:02,680][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:37:03,176][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:37:03,675][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10155 tokens. +[2026-03-26 03:37:04,416][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.42%, ΔTime: 00:00:32 +[2026-03-26 03:37:05,184][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:37:05,187][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:37:05,188][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:37:05,922][__main__][INFO] - Iteration 379 took 1m 1s (41.87% Gen, 56.93% Train). Generation: 25s, Training: 35s. Estimated remaining time: 45h 40m 54s. Estimated total time: 51h 17m 11s. Time estimates for 10 more iterations: 10m 15s, 100 more iterations: 1h 42m 34s, 500 more iterations: 8h 32m 51s. +[2026-03-26 03:37:05,924][__main__][INFO] - Starting iteration 379. +[2026-03-26 03:37:06,419][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 37 and human policies 1. +[2026-03-26 03:37:06,419][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:37:13,181][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 03:37:22,706][__main__][INFO] - Number of regex retries in iteration 379: 1 +[2026-03-26 03:37:22,707][__main__][INFO] - agents played in iteration 379 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:37:23,493][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:37:23,516][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:37:23,540][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:37:23,563][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:37:23,563][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:37:23,564][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:37:24,282][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:37:24,734][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:37:25,243][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:37:25,743][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:37:26,243][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:37:26,743][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:37:27,242][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:37:27,743][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:37:28,241][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:37:28,738][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:37:29,236][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:37:29,734][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:37:30,232][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:37:30,729][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:37:31,226][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:37:31,723][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:37:32,221][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:37:32,725][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:37:33,226][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:37:33,726][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:37:34,226][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:37:34,723][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:37:35,224][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:37:35,724][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:37:36,225][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:37:36,722][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:37:37,219][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:37:37,717][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:37:38,218][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:37:38,715][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:37:39,213][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:37:39,710][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:37:40,208][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:37:40,703][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:37:41,201][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:37:41,697][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:37:42,193][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:37:42,691][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:37:43,186][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:37:43,683][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:37:44,179][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:37:44,678][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:37:45,178][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:37:45,678][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:37:46,177][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:37:46,677][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:37:47,177][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:37:47,675][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:37:48,176][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:37:48,697][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:37:49,200][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:37:49,699][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:37:50,200][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:37:50,701][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:37:51,200][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:37:51,699][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:37:52,198][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:37:52,698][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:37:53,198][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:37:53,697][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:37:54,196][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:37:54,696][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:37:55,196][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:37:55,705][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:37:56,210][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10224 tokens. +[2026-03-26 03:37:56,954][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.21%, Current % of VRAM taken: 60.63%, Block Peak % of device VRAM: 62.43%, ΔTime: 00:00:32 +[2026-03-26 03:37:57,723][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:37:57,725][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:37:57,727][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:37:58,466][__main__][INFO] - Iteration 380 took 52s (31.29% Gen, 67.28% Train). Generation: 16s, Training: 35s. Estimated remaining time: 37h 45m 15s. Estimated total time: 43h 22m 24s. Time estimates for 10 more iterations: 8m 40s, 100 more iterations: 1h 26m 44s, 500 more iterations: 7h 13m 44s. +[2026-03-26 03:37:58,468][__main__][INFO] - Starting iteration 380. +[2026-03-26 03:37:58,867][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 37 and human policies 1. +[2026-03-26 03:37:58,868][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:38:15,778][__main__][INFO] - Number of regex retries in iteration 380: 0 +[2026-03-26 03:38:15,779][__main__][INFO] - agents played in iteration 380 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:38:16,547][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:38:16,570][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:38:16,594][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:38:16,617][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:38:16,618][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:38:16,618][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:38:17,346][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:38:17,800][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:38:18,307][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:38:18,810][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:38:19,316][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:38:19,815][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:38:20,317][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:38:20,816][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:38:21,318][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:38:21,817][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:38:22,314][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:38:22,812][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:38:23,310][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:38:23,808][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:38:24,312][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:38:24,810][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:38:25,312][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:38:25,812][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:38:26,309][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:38:26,808][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:38:27,306][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:38:27,807][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:38:28,309][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:38:28,810][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:38:29,309][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:38:29,810][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:38:30,311][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:38:30,810][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:38:31,310][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:38:31,810][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:38:32,310][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:38:32,811][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:38:33,311][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:38:33,810][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:38:34,307][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:38:34,805][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:38:35,302][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:38:35,803][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:38:36,307][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:38:36,809][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:38:37,307][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:38:37,804][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:38:38,301][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:38:38,798][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:38:39,300][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:38:39,798][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:38:40,295][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:38:40,796][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:38:41,294][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:38:41,800][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:38:42,298][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:38:42,796][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:38:43,294][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:38:43,795][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:38:44,293][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:38:44,791][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:38:45,288][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:38:45,785][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:38:46,281][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:38:46,778][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:38:47,277][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:38:47,784][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:38:48,284][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:38:48,781][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:38:49,278][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10107 tokens. +[2026-03-26 03:38:50,035][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.02%, Current % of VRAM taken: 60.45%, Block Peak % of device VRAM: 62.02%, ΔTime: 00:00:32 +[2026-03-26 03:38:50,863][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:38:50,865][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:38:50,867][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:38:52,341][__main__][INFO] - Iteration 381 took 53s (31.62% Gen, 65.62% Train). Generation: 16s, Training: 35s. Estimated remaining time: 38h 55m 41s. Estimated total time: 44h 33m 44s. Time estimates for 10 more iterations: 8m 54s, 100 more iterations: 1h 29m 7s, 500 more iterations: 7h 25m 37s. +[2026-03-26 03:38:52,344][__main__][INFO] - Starting iteration 381. +[2026-03-26 03:38:52,775][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 38 and human policies 1. +[2026-03-26 03:38:52,776][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:39:08,824][__main__][INFO] - Number of regex retries in iteration 381: 0 +[2026-03-26 03:39:08,825][__main__][INFO] - agents played in iteration 381 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:39:09,591][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:39:09,615][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:39:09,638][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:39:09,662][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:39:09,663][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:39:09,663][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:39:10,384][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:39:10,839][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:39:11,341][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:39:11,838][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:39:12,336][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:39:12,833][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:39:13,331][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:39:13,829][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:39:14,326][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:39:14,826][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:39:15,323][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:39:15,821][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:39:16,319][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:39:16,819][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:39:17,325][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:39:17,823][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:39:18,322][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:39:18,826][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:39:19,323][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:39:19,820][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:39:20,320][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:39:20,817][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:39:21,315][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:39:21,812][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:39:22,309][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:39:22,809][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:39:23,309][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:39:23,809][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:39:24,311][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:39:24,811][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:39:25,311][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:39:25,811][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:39:26,310][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:39:26,808][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:39:27,308][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:39:27,804][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:39:28,305][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:39:28,802][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:39:29,301][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:39:29,802][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:39:30,299][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:39:30,799][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:39:31,321][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:39:31,823][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:39:32,324][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:39:32,825][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:39:33,325][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:39:33,825][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:39:34,326][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:39:34,823][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:39:35,321][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:39:35,821][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:39:36,318][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:39:36,822][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:39:37,321][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:39:37,820][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:39:38,319][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:39:38,817][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:39:39,316][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:39:39,812][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:39:40,309][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:39:40,806][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:39:41,302][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:39:41,799][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:39:42,296][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10140 tokens. +[2026-03-26 03:39:43,041][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.30%, ΔTime: 00:00:32 +[2026-03-26 03:39:44,086][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:39:44,088][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:39:44,090][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:39:45,135][__main__][INFO] - Iteration 382 took 52s (30.65% Gen, 67.35% Train). Generation: 16s, Training: 35s. Estimated remaining time: 37h 59m 6s. Estimated total time: 43h 38m 3s. Time estimates for 10 more iterations: 8m 43s, 100 more iterations: 1h 27m 16s, 500 more iterations: 7h 16m 20s. +[2026-03-26 03:39:45,137][__main__][INFO] - Starting iteration 382. +[2026-03-26 03:39:45,539][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 38 and human policies 1. +[2026-03-26 03:39:45,539][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:40:02,241][__main__][INFO] - Number of regex retries in iteration 382: 0 +[2026-03-26 03:40:02,242][__main__][INFO] - agents played in iteration 382 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:40:03,003][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:40:03,027][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:40:03,050][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:40:03,074][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:40:03,075][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:40:03,075][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:40:03,797][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:40:04,253][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:40:04,758][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:40:05,261][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:40:05,761][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:40:06,261][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:40:06,760][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:40:07,259][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:40:07,759][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:40:08,258][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:40:08,758][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:40:09,258][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:40:09,758][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:40:10,260][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:40:10,759][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:40:11,259][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:40:11,759][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:40:12,255][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:40:12,754][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:40:13,251][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:40:13,748][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:40:14,246][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:40:14,742][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:40:15,238][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:40:15,738][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:40:16,236][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:40:16,734][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:40:17,234][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:40:17,735][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:40:18,240][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:40:18,738][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:40:19,236][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:40:19,735][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:40:20,235][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:40:20,736][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:40:21,237][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:40:21,739][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:40:22,237][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:40:22,738][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:40:23,238][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:40:23,738][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:40:24,238][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:40:24,738][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:40:25,239][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:40:25,738][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:40:26,238][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:40:26,738][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:40:27,239][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:40:27,740][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:40:28,241][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:40:28,739][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:40:29,237][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:40:29,735][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:40:30,235][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:40:30,733][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:40:31,233][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:40:31,732][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:40:32,232][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:40:32,728][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:40:33,225][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:40:33,725][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:40:34,225][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:40:34,722][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:40:35,221][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:40:35,726][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10223 tokens. +[2026-03-26 03:40:36,479][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.25%, Current % of VRAM taken: 60.68%, Block Peak % of device VRAM: 62.29%, ΔTime: 00:00:32 +[2026-03-26 03:40:37,220][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:40:37,222][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:40:37,224][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:40:37,982][__main__][INFO] - Iteration 383 took 52s (31.85% Gen, 66.70% Train). Generation: 16s, Training: 34s. Estimated remaining time: 38h 2m 23s. Estimated total time: 43h 42m 12s. Time estimates for 10 more iterations: 8m 44s, 100 more iterations: 1h 27m 24s, 500 more iterations: 7h 17m 2s. +[2026-03-26 03:40:37,984][__main__][INFO] - Starting iteration 383. +[2026-03-26 03:40:38,393][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 38 and human policies 1. +[2026-03-26 03:40:38,394][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:40:42,670][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 03:40:42,896][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 03:40:54,396][__main__][INFO] - Number of regex retries in iteration 383: 2 +[2026-03-26 03:40:54,397][__main__][INFO] - agents played in iteration 383 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:40:55,178][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:40:55,201][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:40:55,225][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:40:55,248][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:40:55,250][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:40:55,250][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:40:55,971][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:40:56,428][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:40:56,928][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:40:57,426][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:40:57,923][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:40:58,420][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:40:58,917][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:40:59,420][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:40:59,917][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:41:00,418][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:41:00,914][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:41:01,412][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:41:01,910][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:41:02,408][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:41:02,907][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:41:03,404][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:41:03,901][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:41:04,400][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:41:04,895][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:41:05,393][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:41:05,889][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:41:06,388][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:41:06,883][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:41:07,378][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:41:07,877][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:41:08,374][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:41:08,872][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:41:09,369][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:41:09,867][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:41:10,363][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:41:10,860][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:41:11,360][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:41:11,861][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:41:12,356][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:41:12,856][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:41:13,352][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:41:13,847][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:41:14,347][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:41:14,845][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:41:15,340][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:41:15,837][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:41:16,340][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:41:16,839][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:41:17,338][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:41:17,842][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:41:18,339][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:41:18,835][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:41:19,333][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:41:19,833][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:41:20,333][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:41:20,831][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:41:21,329][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:41:21,825][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:41:22,322][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:41:22,820][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:41:23,318][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:41:23,816][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:41:24,314][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:41:24,811][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:41:25,309][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:41:25,806][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:41:26,303][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:41:26,799][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:41:27,297][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:41:27,795][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10143 tokens. +[2026-03-26 03:41:28,542][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.18%, ΔTime: 00:00:32 +[2026-03-26 03:41:29,313][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:41:29,315][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:41:29,317][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:41:30,081][__main__][INFO] - Iteration 384 took 51s (30.96% Gen, 67.56% Train). Generation: 16s, Training: 34s. Estimated remaining time: 37h 23m 49s. Estimated total time: 43h 4m 30s. Time estimates for 10 more iterations: 8m 36s, 100 more iterations: 1h 26m 9s, 500 more iterations: 7h 10m 45s. +[2026-03-26 03:41:30,083][__main__][INFO] - Starting iteration 384. +[2026-03-26 03:41:30,497][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 38 and human policies 1. +[2026-03-26 03:41:30,498][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:41:37,990][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 03:41:48,410][__main__][INFO] - Number of regex retries in iteration 384: 1 +[2026-03-26 03:41:48,411][__main__][INFO] - agents played in iteration 384 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:41:49,178][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:41:49,201][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:41:49,225][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:41:49,248][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:41:49,248][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:41:49,249][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:41:49,976][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:41:50,433][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:41:50,934][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:41:51,432][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:41:51,927][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:41:52,427][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:41:52,925][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:41:53,424][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:41:53,929][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:41:54,426][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:41:54,924][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:41:55,424][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:41:55,921][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:41:56,417][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:41:56,914][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:41:57,414][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:41:57,914][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:41:58,412][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:41:58,910][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:41:59,407][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:41:59,903][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:42:00,400][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:42:00,898][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:42:01,398][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:42:01,895][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:42:02,391][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:42:02,887][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:42:03,384][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:42:03,883][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:42:04,382][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:42:04,879][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:42:05,375][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:42:05,873][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:42:06,375][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:42:06,875][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:42:07,372][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:42:07,870][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:42:08,371][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:42:08,872][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:42:09,372][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:42:09,869][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:42:10,369][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:42:10,866][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:42:11,366][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:42:11,863][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:42:12,359][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:42:12,857][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:42:13,354][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:42:13,853][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:42:14,351][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:42:14,846][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:42:15,342][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:42:15,839][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:42:16,339][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:42:16,835][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:42:17,331][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:42:17,827][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:42:18,324][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:42:18,821][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:42:19,317][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:42:19,815][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:42:20,315][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:42:20,825][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:42:21,324][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:42:21,825][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10132 tokens. +[2026-03-26 03:42:22,588][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 62.17%, ΔTime: 00:00:32 +[2026-03-26 03:42:23,377][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:42:23,380][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:42:23,382][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:42:24,061][__main__][INFO] - Iteration 385 took 53s (33.44% Gen, 65.29% Train). Generation: 17s, Training: 34s. Estimated remaining time: 38h 56m 38s. Estimated total time: 44h 38m 13s. Time estimates for 10 more iterations: 8m 55s, 100 more iterations: 1h 29m 16s, 500 more iterations: 7h 26m 22s. +[2026-03-26 03:42:24,063][__main__][INFO] - Starting iteration 385. +[2026-03-26 03:42:24,471][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 38 and human policies 1. +[2026-03-26 03:42:24,472][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:42:40,940][__main__][INFO] - Number of regex retries in iteration 385: 0 +[2026-03-26 03:42:40,941][__main__][INFO] - agents played in iteration 385 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:42:41,704][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:42:41,728][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:42:41,751][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:42:41,774][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:42:41,774][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:42:41,775][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:42:42,485][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:42:42,937][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:42:43,442][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:42:43,941][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:42:44,439][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:42:44,937][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:42:45,435][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:42:45,937][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:42:46,434][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:42:46,938][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:42:47,440][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:42:47,956][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:42:48,454][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:42:48,959][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:42:49,457][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:42:49,959][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:42:50,458][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:42:50,954][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:42:51,452][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:42:51,948][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:42:52,444][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:42:52,939][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:42:53,438][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:42:53,935][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:42:54,433][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:42:54,935][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:42:55,433][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:42:55,929][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:42:56,427][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:42:56,924][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:42:57,426][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:42:57,927][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:42:58,426][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:42:58,931][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:42:59,430][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:42:59,929][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:43:00,432][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:43:00,930][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:43:01,433][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:43:01,931][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:43:02,433][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:43:02,931][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:43:03,427][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:43:03,925][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:43:04,422][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:43:04,920][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:43:05,414][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:43:05,909][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:43:06,404][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:43:06,898][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:43:07,394][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:43:07,893][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:43:08,389][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:43:08,884][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:43:09,380][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:43:09,876][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:43:10,373][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:43:10,874][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:43:11,373][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:43:11,870][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:43:12,368][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:43:12,866][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:43:13,368][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:43:13,870][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:43:14,367][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10125 tokens. +[2026-03-26 03:43:15,114][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.17%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 03:43:15,939][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:43:15,941][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:43:15,943][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:43:16,836][__main__][INFO] - Iteration 386 took 52s (31.45% Gen, 66.84% Train). Generation: 16s, Training: 35s. Estimated remaining time: 37h 55m 49s. Estimated total time: 43h 38m 17s. Time estimates for 10 more iterations: 8m 43s, 100 more iterations: 1h 27m 16s, 500 more iterations: 7h 16m 22s. +[2026-03-26 03:43:16,838][__main__][INFO] - Starting iteration 386. +[2026-03-26 03:43:17,239][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 38 and human policies 1. +[2026-03-26 03:43:17,240][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:43:32,071][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 03:43:33,820][__main__][INFO] - Number of regex retries in iteration 386: 1 +[2026-03-26 03:43:33,820][__main__][INFO] - agents played in iteration 386 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:43:34,582][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:43:34,605][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:43:34,628][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:43:34,651][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:43:34,652][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:43:34,652][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:43:35,362][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:43:35,817][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:43:36,320][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:43:36,819][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:43:37,319][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:43:37,821][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:43:38,318][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:43:38,815][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:43:39,314][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:43:39,812][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:43:40,311][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:43:40,807][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:43:41,304][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:43:41,800][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:43:42,298][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:43:42,798][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:43:43,295][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:43:43,793][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:43:44,291][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:43:44,789][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:43:45,287][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:43:45,785][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:43:46,282][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:43:46,780][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:43:47,278][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:43:47,777][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:43:48,278][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:43:48,777][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:43:49,275][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:43:49,773][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:43:50,273][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:43:50,775][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:43:51,274][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:43:51,775][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:43:52,277][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:43:52,777][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:43:53,279][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:43:53,779][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:43:54,282][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:43:54,786][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:43:55,289][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:43:55,786][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:43:56,283][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:43:56,779][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:43:57,275][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:43:57,771][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:43:58,267][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:43:58,763][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:43:59,260][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:43:59,757][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:44:00,255][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:44:00,753][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:44:01,253][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:44:01,751][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:44:02,248][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:44:02,747][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:44:03,244][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:44:03,740][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:44:04,242][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:44:04,743][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:44:05,244][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:44:05,741][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:44:06,238][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:44:06,738][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:44:07,236][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10156 tokens. +[2026-03-26 03:44:07,998][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.03%, Current % of VRAM taken: 60.46%, Block Peak % of device VRAM: 62.37%, ΔTime: 00:00:32 +[2026-03-26 03:44:08,822][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:44:08,824][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:44:08,826][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:44:09,810][__main__][INFO] - Iteration 387 took 52s (31.54% Gen, 66.59% Train). Generation: 16s, Training: 35s. Estimated remaining time: 38h 5m 13s. Estimated total time: 43h 48m 34s. Time estimates for 10 more iterations: 8m 45s, 100 more iterations: 1h 27m 37s, 500 more iterations: 7h 18m 5s. +[2026-03-26 03:44:09,815][__main__][INFO] - Starting iteration 387. +[2026-03-26 03:44:10,219][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 38 and human policies 1. +[2026-03-26 03:44:10,220][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:44:28,476][__main__][INFO] - Number of regex retries in iteration 387: 0 +[2026-03-26 03:44:28,477][__main__][INFO] - agents played in iteration 387 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:44:29,243][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:44:29,266][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:44:29,289][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:44:29,312][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:44:29,313][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:44:29,313][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:44:30,030][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:44:30,485][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:44:30,993][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:44:31,497][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:44:31,997][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:44:32,495][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:44:32,997][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:44:33,495][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:44:33,996][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:44:34,493][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:44:34,993][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:44:35,491][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:44:35,989][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:44:36,486][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:44:36,984][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:44:37,480][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:44:37,979][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:44:38,483][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:44:38,985][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:44:39,486][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:44:39,986][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:44:40,486][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:44:40,986][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:44:41,483][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:44:41,983][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:44:42,485][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:44:42,986][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:44:43,482][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:44:43,983][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:44:44,480][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:44:44,978][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:44:45,479][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:44:45,980][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:44:46,479][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:44:46,976][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:44:47,477][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:44:47,978][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:44:48,477][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:44:48,975][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:44:49,474][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:44:49,972][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:44:50,474][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:44:50,975][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:44:51,474][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:44:51,981][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:44:52,482][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:44:52,984][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:44:53,485][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:44:53,985][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:44:54,482][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:44:54,983][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:44:55,482][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:44:55,982][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:44:56,479][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:44:56,980][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:44:57,479][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:44:57,981][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:44:58,482][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:44:58,982][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:44:59,480][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:44:59,979][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:45:00,480][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:45:00,981][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:45:01,482][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:45:01,982][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10271 tokens. +[2026-03-26 03:45:02,741][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.22%, ΔTime: 00:00:32 +[2026-03-26 03:45:03,450][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:45:03,452][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:45:03,454][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:45:04,202][__main__][INFO] - Iteration 388 took 53s (33.82% Gen, 64.79% Train). Generation: 18s, Training: 34s. Estimated remaining time: 39h 14m 55s. Estimated total time: 44h 59m 10s. Time estimates for 10 more iterations: 8m 59s, 100 more iterations: 1h 29m 58s, 500 more iterations: 7h 29m 51s. +[2026-03-26 03:45:04,204][__main__][INFO] - Starting iteration 388. +[2026-03-26 03:45:04,615][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 38 and human policies 1. +[2026-03-26 03:45:04,616][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:45:11,160][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 03:45:22,725][__main__][INFO] - Number of regex retries in iteration 388: 1 +[2026-03-26 03:45:22,726][__main__][INFO] - agents played in iteration 388 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:45:23,488][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:45:23,512][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:45:23,535][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:45:23,558][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:45:23,558][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:45:23,559][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:45:24,272][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:45:24,728][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:45:25,231][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:45:25,734][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:45:26,234][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:45:26,733][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:45:27,234][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:45:27,732][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:45:28,229][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:45:28,732][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:45:29,236][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:45:29,739][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:45:30,244][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:45:30,741][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:45:31,241][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:45:31,739][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:45:32,237][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:45:32,734][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:45:33,233][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:45:33,731][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:45:34,231][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:45:34,726][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:45:35,225][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:45:35,720][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:45:36,217][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:45:36,715][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:45:37,216][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:45:37,715][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:45:38,216][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:45:38,716][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:45:39,217][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:45:39,717][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:45:40,220][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:45:40,716][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:45:41,217][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:45:41,737][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:45:42,242][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:45:42,741][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:45:43,243][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:45:43,744][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:45:44,245][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:45:44,741][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:45:45,241][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:45:45,738][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:45:46,235][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:45:46,733][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:45:47,232][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:45:47,729][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:45:48,227][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:45:48,725][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:45:49,221][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:45:49,720][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:45:50,222][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:45:50,725][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:45:51,223][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:45:51,724][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:45:52,221][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:45:52,719][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:45:53,220][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:45:53,725][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:45:54,229][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:45:54,743][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:45:55,247][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:45:55,748][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:45:56,252][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10296 tokens. +[2026-03-26 03:45:57,010][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.60%, Block Peak % of device VRAM: 62.26%, ΔTime: 00:00:32 +[2026-03-26 03:45:57,938][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:45:57,941][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:45:57,943][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:45:59,352][__main__][INFO] - Iteration 389 took 54s (33.08% Gen, 64.34% Train). Generation: 18s, Training: 35s. Estimated remaining time: 39h 51m 42s. Estimated total time: 45h 36m 53s. Time estimates for 10 more iterations: 9m 7s, 100 more iterations: 1h 31m 13s, 500 more iterations: 7h 36m 8s. +[2026-03-26 03:45:59,354][__main__][INFO] - Starting iteration 389. +[2026-03-26 03:45:59,754][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 38 and human policies 1. +[2026-03-26 03:45:59,755][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:46:15,955][__main__][INFO] - Number of regex retries in iteration 389: 0 +[2026-03-26 03:46:15,957][__main__][INFO] - agents played in iteration 389 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:46:16,718][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:46:16,742][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:46:16,766][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:46:16,790][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:46:16,791][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:46:16,791][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:46:17,511][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:46:17,966][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:46:18,469][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:46:18,968][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:46:19,470][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:46:19,977][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:46:20,479][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:46:20,978][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:46:21,478][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:46:21,978][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:46:22,479][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:46:22,982][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:46:23,483][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:46:23,985][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:46:24,485][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:46:24,983][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:46:25,483][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:46:25,982][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:46:26,480][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:46:26,978][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:46:27,477][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:46:27,975][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:46:28,473][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:46:28,970][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:46:29,471][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:46:29,967][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:46:30,464][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:46:30,962][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:46:31,462][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:46:31,966][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:46:32,468][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:46:32,964][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:46:33,464][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:46:33,960][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:46:34,458][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:46:34,956][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:46:35,454][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:46:35,954][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:46:36,451][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:46:36,947][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:46:37,443][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:46:37,941][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:46:38,438][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:46:38,937][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:46:39,441][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:46:39,960][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:46:40,459][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:46:40,960][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:46:41,459][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:46:41,957][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:46:42,454][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:46:42,953][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:46:43,451][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:46:43,952][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:46:44,457][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:46:44,959][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:46:45,458][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:46:45,957][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:46:46,459][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:46:46,959][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:46:47,459][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:46:47,959][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:46:48,460][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:46:48,960][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:46:49,461][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10172 tokens. +[2026-03-26 03:46:50,288][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.15%, ΔTime: 00:00:32 +[2026-03-26 03:46:51,127][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:46:51,129][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:46:51,131][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:46:52,145][__main__][INFO] - Iteration 390 took 52s (30.92% Gen, 67.14% Train). Generation: 16s, Training: 35s. Estimated remaining time: 37h 53m 30s. Estimated total time: 43h 39m 34s. Time estimates for 10 more iterations: 8m 43s, 100 more iterations: 1h 27m 19s, 500 more iterations: 7h 16m 35s. +[2026-03-26 03:46:52,147][__main__][INFO] - Starting iteration 390. +[2026-03-26 03:46:52,542][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 38 and human policies 1. +[2026-03-26 03:46:52,542][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:47:10,360][__main__][INFO] - Number of regex retries in iteration 390: 0 +[2026-03-26 03:47:10,361][__main__][INFO] - agents played in iteration 390 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:47:11,126][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:47:11,150][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:47:11,173][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:47:11,196][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:47:11,197][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:47:11,198][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:47:11,921][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:47:12,382][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:47:12,887][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:47:13,386][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:47:13,885][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:47:14,384][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:47:14,892][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:47:15,394][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:47:15,896][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:47:16,395][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:47:16,897][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:47:17,400][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:47:17,900][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:47:18,404][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:47:18,902][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:47:19,402][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:47:19,901][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:47:20,406][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:47:20,926][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:47:21,428][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:47:21,927][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:47:22,426][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:47:22,925][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:47:23,428][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:47:23,930][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:47:24,429][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:47:24,928][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:47:25,427][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:47:25,925][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:47:26,424][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:47:26,926][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:47:27,424][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:47:27,930][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:47:28,428][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:47:28,927][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:47:29,430][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:47:29,929][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:47:30,426][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:47:30,924][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:47:31,422][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:47:31,921][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:47:32,420][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:47:32,920][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:47:33,422][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:47:33,922][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:47:34,424][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:47:34,927][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:47:35,425][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:47:35,923][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:47:36,426][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:47:36,931][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:47:37,433][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:47:37,934][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:47:38,434][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:47:38,937][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:47:39,440][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:47:39,942][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:47:40,446][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:47:40,949][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:47:41,457][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:47:41,963][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:47:42,467][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:47:42,981][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:47:43,487][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:47:43,989][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10166 tokens. +[2026-03-26 03:47:44,764][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.32%, ΔTime: 00:00:32 +[2026-03-26 03:47:45,572][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:47:45,575][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:47:45,576][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:47:47,262][__main__][INFO] - Iteration 391 took 54s (32.56% Gen, 64.35% Train). Generation: 17s, Training: 35s. Estimated remaining time: 39h 49m 5s. Estimated total time: 45h 36m 3s. Time estimates for 10 more iterations: 9m 7s, 100 more iterations: 1h 31m 12s, 500 more iterations: 7h 36m 0s. +[2026-03-26 03:47:47,264][__main__][INFO] - Starting iteration 391. +[2026-03-26 03:47:47,670][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 39 and human policies 1. +[2026-03-26 03:47:47,671][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:48:02,322][__main__][INFO] - Number of regex retries in iteration 391: 0 +[2026-03-26 03:48:02,323][__main__][INFO] - agents played in iteration 391 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:48:03,084][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:48:03,107][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:48:03,130][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:48:03,153][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:48:03,154][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:48:03,155][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:48:03,877][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:48:04,329][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:48:04,829][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:48:05,328][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:48:05,827][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:48:06,325][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:48:06,825][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:48:07,331][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:48:07,853][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:48:08,352][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:48:08,849][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:48:09,346][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:48:09,844][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:48:10,340][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:48:10,839][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:48:11,338][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:48:11,836][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:48:12,334][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:48:12,838][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:48:13,335][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:48:13,834][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:48:14,332][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:48:14,829][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:48:15,327][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:48:15,826][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:48:16,321][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:48:16,819][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:48:17,316][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:48:17,813][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:48:18,308][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:48:18,814][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:48:19,311][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:48:19,806][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:48:20,305][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:48:20,802][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:48:21,301][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:48:21,797][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:48:22,294][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:48:22,790][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:48:23,286][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:48:23,782][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:48:24,280][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:48:24,779][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:48:25,279][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:48:25,778][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:48:26,276][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:48:26,773][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:48:27,271][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:48:27,769][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:48:28,266][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:48:28,763][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:48:29,260][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:48:29,761][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:48:30,261][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:48:30,761][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:48:31,259][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:48:31,758][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:48:32,259][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:48:32,762][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:48:33,263][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:48:33,765][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:48:34,264][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:48:34,765][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:48:35,266][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:48:35,766][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9996 tokens. +[2026-03-26 03:48:36,533][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 03:48:37,314][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:48:37,316][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:48:37,318][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:48:38,237][__main__][INFO] - Iteration 392 took 50s (28.98% Gen, 69.20% Train). Generation: 14s, Training: 34s. Estimated remaining time: 36h 20m 33s. Estimated total time: 42h 8m 23s. Time estimates for 10 more iterations: 8m 25s, 100 more iterations: 1h 24m 16s, 500 more iterations: 7h 1m 23s. +[2026-03-26 03:48:38,239][__main__][INFO] - Starting iteration 392. +[2026-03-26 03:48:38,640][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 39 and human policies 1. +[2026-03-26 03:48:38,641][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:48:53,022][__main__][INFO] - Number of regex retries in iteration 392: 0 +[2026-03-26 03:48:53,023][__main__][INFO] - agents played in iteration 392 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:48:53,781][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:48:53,805][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:48:53,828][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:48:53,851][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:48:53,852][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:48:53,852][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:48:54,566][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:48:55,019][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:48:55,523][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:48:56,019][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:48:56,518][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:48:57,016][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:48:57,515][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:48:58,013][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:48:58,508][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:48:59,011][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:48:59,510][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:49:00,009][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:49:00,508][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:49:01,006][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:49:01,505][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:49:02,002][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:49:02,500][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:49:03,019][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:49:03,518][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:49:04,021][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:49:04,525][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:49:05,022][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:49:05,523][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:49:06,023][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:49:06,523][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:49:07,019][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:49:07,519][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:49:08,015][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:49:08,511][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:49:09,007][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:49:09,503][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:49:10,000][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:49:10,499][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:49:11,007][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:49:11,506][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:49:12,005][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:49:12,506][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:49:13,003][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:49:13,501][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:49:13,999][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:49:14,496][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:49:14,994][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:49:15,492][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:49:15,992][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:49:16,489][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:49:16,986][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:49:17,482][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:49:17,983][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:49:18,481][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:49:18,983][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:49:19,483][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:49:19,981][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:49:20,486][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:49:20,995][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:49:21,496][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:49:21,996][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:49:22,500][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:49:23,005][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:49:23,508][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:49:24,008][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:49:24,512][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:49:25,014][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:49:25,515][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:49:26,016][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:49:26,517][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10194 tokens. +[2026-03-26 03:49:27,266][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.19%, Current % of VRAM taken: 60.62%, Block Peak % of device VRAM: 62.23%, ΔTime: 00:00:32 +[2026-03-26 03:49:28,020][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:49:28,022][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:49:28,024][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:49:28,840][__main__][INFO] - Iteration 393 took 50s (28.65% Gen, 69.72% Train). Generation: 14s, Training: 35s. Estimated remaining time: 36h 1m 20s. Estimated total time: 41h 50m 0s. Time estimates for 10 more iterations: 8m 22s, 100 more iterations: 1h 23m 40s, 500 more iterations: 6h 58m 20s. +[2026-03-26 03:49:28,842][__main__][INFO] - Starting iteration 393. +[2026-03-26 03:49:29,247][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 39 and human policies 1. +[2026-03-26 03:49:29,248][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:49:44,770][__main__][INFO] - Number of regex retries in iteration 393: 0 +[2026-03-26 03:49:44,771][__main__][INFO] - agents played in iteration 393 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:49:45,536][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:49:45,560][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:49:45,583][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:49:45,606][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:49:45,607][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:49:45,608][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:49:46,324][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:49:46,782][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:49:47,282][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:49:47,781][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:49:48,279][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:49:48,776][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:49:49,273][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:49:49,771][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:49:50,269][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:49:50,766][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:49:51,262][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:49:51,759][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:49:52,256][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:49:52,756][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:49:53,253][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:49:53,751][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:49:54,249][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:49:54,751][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:49:55,253][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:49:55,753][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:49:56,252][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:49:56,752][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:49:57,252][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:49:57,753][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:49:58,250][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:49:58,750][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:49:59,247][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:49:59,744][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:50:00,243][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:50:00,744][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:50:01,244][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:50:01,742][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:50:02,243][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:50:02,740][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:50:03,237][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:50:03,734][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:50:04,231][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:50:04,727][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:50:05,224][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:50:05,722][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:50:06,219][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:50:06,723][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:50:07,238][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:50:07,738][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:50:08,238][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:50:08,739][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:50:09,239][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:50:09,736][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:50:10,236][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:50:10,735][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:50:11,233][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:50:11,733][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:50:12,239][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:50:12,743][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:50:13,243][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:50:13,741][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:50:14,244][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:50:14,742][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:50:15,242][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:50:15,743][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:50:16,246][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:50:16,747][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:50:17,248][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:50:17,747][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:50:18,248][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10279 tokens. +[2026-03-26 03:50:18,997][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.16%, Current % of VRAM taken: 60.58%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 03:50:19,721][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:50:19,723][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:50:19,725][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:50:20,405][__main__][INFO] - Iteration 394 took 51s (30.34% Gen, 68.32% Train). Generation: 15s, Training: 34s. Estimated remaining time: 36h 48m 28s. Estimated total time: 42h 37m 59s. Time estimates for 10 more iterations: 8m 31s, 100 more iterations: 1h 25m 15s, 500 more iterations: 7h 6m 19s. +[2026-03-26 03:50:20,407][__main__][INFO] - Starting iteration 394. +[2026-03-26 03:50:20,823][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 39 and human policies 1. +[2026-03-26 03:50:20,824][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:50:26,355][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 03:50:29,676][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 03:50:37,885][__main__][INFO] - Number of regex retries in iteration 394: 2 +[2026-03-26 03:50:37,886][__main__][INFO] - agents played in iteration 394 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:50:38,653][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:50:38,677][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:50:38,700][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:50:38,723][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:50:38,724][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:50:38,724][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:50:39,438][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:50:39,895][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:50:40,400][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:50:40,900][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:50:41,400][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:50:41,899][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:50:42,399][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:50:42,899][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:50:43,396][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:50:43,894][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:50:44,391][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:50:44,888][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:50:45,385][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:50:45,882][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:50:46,378][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:50:46,879][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:50:47,379][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:50:47,879][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:50:48,378][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:50:48,878][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:50:49,378][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:50:49,875][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:50:50,374][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:50:50,873][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:50:51,370][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:50:51,868][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:50:52,370][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:50:52,869][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:50:53,369][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:50:53,870][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:50:54,366][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:50:54,866][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:50:55,364][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:50:55,861][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:50:56,362][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:50:56,863][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:50:57,362][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:50:57,861][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:50:58,362][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:50:58,862][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:50:59,359][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:50:59,859][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:51:00,359][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:51:00,859][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:51:01,359][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:51:01,859][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:51:02,357][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:51:02,854][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:51:03,354][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:51:03,853][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:51:04,355][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:51:04,856][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:51:05,357][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:51:05,860][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:51:06,359][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:51:06,859][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:51:07,359][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:51:07,859][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:51:08,363][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:51:08,865][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:51:09,364][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:51:09,868][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:51:10,368][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:51:10,868][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:51:11,368][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10308 tokens. +[2026-03-26 03:51:12,112][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.61%, Block Peak % of device VRAM: 62.09%, ΔTime: 00:00:32 +[2026-03-26 03:51:12,881][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:51:12,883][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:51:12,885][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:51:13,897][__main__][INFO] - Iteration 395 took 53s (32.15% Gen, 65.94% Train). Generation: 17s, Training: 34s. Estimated remaining time: 38h 23m 18s. Estimated total time: 44h 13m 43s. Time estimates for 10 more iterations: 8m 50s, 100 more iterations: 1h 28m 27s, 500 more iterations: 7h 22m 17s. +[2026-03-26 03:51:13,899][__main__][INFO] - Starting iteration 395. +[2026-03-26 03:51:14,303][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 39 and human policies 1. +[2026-03-26 03:51:14,303][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:51:18,600][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 03:51:28,980][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 1 y book, 10 z balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 03:51:30,527][__main__][INFO] - Number of regex retries in iteration 395: 2 +[2026-03-26 03:51:30,528][__main__][INFO] - agents played in iteration 395 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:51:31,293][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:51:31,316][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:51:31,339][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:51:31,362][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:51:31,363][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:51:31,363][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:51:32,085][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:51:32,538][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:51:33,043][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:51:33,541][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:51:34,039][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:51:34,538][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:51:35,036][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:51:35,535][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:51:36,033][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:51:36,532][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:51:37,035][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:51:37,534][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:51:38,029][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:51:38,527][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:51:39,030][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:51:39,527][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:51:40,024][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:51:40,521][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:51:41,019][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:51:41,517][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:51:42,014][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:51:42,513][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:51:43,014][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:51:43,512][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:51:44,010][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:51:44,509][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:51:45,008][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:51:45,510][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:51:46,007][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:51:46,508][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:51:47,011][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:51:47,512][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:51:48,014][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:51:48,512][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:51:49,009][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:51:49,508][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:51:50,006][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:51:50,506][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:51:51,003][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:51:51,500][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:51:51,997][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:51:52,495][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:51:52,992][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:51:53,489][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:51:53,986][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:51:54,484][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:51:54,988][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:51:55,491][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:51:55,992][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:51:56,495][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:51:56,999][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:51:57,503][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:51:58,007][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:51:58,512][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:51:59,016][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:51:59,520][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:52:00,024][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:52:00,523][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:52:01,020][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:52:01,519][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:52:02,018][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:52:02,517][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:52:03,017][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:52:03,515][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:52:04,013][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10185 tokens. +[2026-03-26 03:52:04,772][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 03:52:05,505][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:52:05,508][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:52:05,510][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:52:06,169][__main__][INFO] - Iteration 396 took 51s (31.28% Gen, 67.44% Train). Generation: 16s, Training: 34s. Estimated remaining time: 37h 22m 5s. Estimated total time: 43h 13m 22s. Time estimates for 10 more iterations: 8m 38s, 100 more iterations: 1h 26m 26s, 500 more iterations: 7h 12m 13s. +[2026-03-26 03:52:06,172][__main__][INFO] - Starting iteration 396. +[2026-03-26 03:52:06,586][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 39 and human policies 1. +[2026-03-26 03:52:06,587][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:52:23,704][__main__][INFO] - Number of regex retries in iteration 396: 0 +[2026-03-26 03:52:23,705][__main__][INFO] - agents played in iteration 396 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:52:24,473][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:52:24,497][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:52:24,520][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:52:24,543][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:52:24,544][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:52:24,545][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:52:25,268][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:52:25,723][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:52:26,224][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:52:26,722][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:52:27,224][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:52:27,722][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:52:28,218][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:52:28,716][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:52:29,215][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:52:29,718][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:52:30,219][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:52:30,718][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:52:31,222][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:52:31,721][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:52:32,222][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:52:32,723][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:52:33,221][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:52:33,719][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:52:34,218][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:52:34,716][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:52:35,213][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:52:35,709][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:52:36,204][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:52:36,703][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:52:37,204][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:52:37,701][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:52:38,197][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:52:38,694][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:52:39,190][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:52:39,690][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:52:40,185][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:52:40,681][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:52:41,180][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:52:41,681][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:52:42,177][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:52:42,678][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:52:43,179][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:52:43,680][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:52:44,177][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:52:44,679][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:52:45,180][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:52:45,681][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:52:46,184][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:52:46,691][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:52:47,194][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:52:47,693][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:52:48,193][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:52:48,699][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:52:49,201][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:52:49,706][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:52:50,207][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:52:50,708][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:52:51,208][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:52:51,711][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:52:52,211][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:52:52,708][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:52:53,206][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:52:53,702][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:52:54,203][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:52:54,701][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:52:55,199][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:52:55,697][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:52:56,195][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:52:56,693][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:52:57,190][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10291 tokens. +[2026-03-26 03:52:57,948][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 03:52:58,672][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:52:58,674][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:52:58,676][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:52:59,489][__main__][INFO] - Iteration 397 took 52s (32.36% Gen, 66.10% Train). Generation: 17s, Training: 34s. Estimated remaining time: 38h 12m 59s. Estimated total time: 44h 5m 9s. Time estimates for 10 more iterations: 8m 49s, 100 more iterations: 1h 28m 10s, 500 more iterations: 7h 20m 51s. +[2026-03-26 03:52:59,491][__main__][INFO] - Starting iteration 397. +[2026-03-26 03:52:59,900][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 39 and human policies 1. +[2026-03-26 03:52:59,901][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:53:15,126][__main__][INFO] - Number of regex retries in iteration 397: 0 +[2026-03-26 03:53:15,127][__main__][INFO] - agents played in iteration 397 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:53:15,898][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:53:15,922][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:53:15,945][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:53:15,968][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:53:15,969][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:53:15,969][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:53:16,684][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:53:17,141][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:53:17,647][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:53:18,147][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:53:18,646][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:53:19,146][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:53:19,648][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:53:20,147][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:53:20,648][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:53:21,148][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:53:21,648][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:53:22,145][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:53:22,645][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:53:23,145][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:53:23,645][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:53:24,145][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:53:24,644][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:53:25,143][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:53:25,639][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:53:26,140][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:53:26,638][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:53:27,137][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:53:27,634][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:53:28,131][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:53:28,629][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:53:29,127][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:53:29,624][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:53:30,122][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:53:30,620][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:53:31,119][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:53:31,618][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:53:32,117][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:53:32,615][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:53:33,113][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:53:33,610][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:53:34,107][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:53:34,604][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:53:35,101][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:53:35,599][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:53:36,096][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:53:36,594][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:53:37,096][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:53:37,595][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:53:38,095][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:53:38,604][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:53:39,106][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:53:39,609][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:53:40,107][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:53:40,612][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:53:41,112][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:53:41,613][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:53:42,114][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:53:42,615][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:53:43,116][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:53:43,617][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:53:44,116][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:53:44,620][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:53:45,121][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:53:45,618][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:53:46,116][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:53:46,612][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:53:47,109][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:53:47,606][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:53:48,105][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:53:48,604][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10139 tokens. +[2026-03-26 03:53:49,361][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.29%, ΔTime: 00:00:32 +[2026-03-26 03:53:50,324][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:53:50,327][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:53:50,328][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:53:51,112][__main__][INFO] - Iteration 398 took 51s (29.73% Gen, 68.74% Train). Generation: 15s, Training: 35s. Estimated remaining time: 36h 47m 36s. Estimated total time: 42h 40m 39s. Time estimates for 10 more iterations: 8m 32s, 100 more iterations: 1h 25m 21s, 500 more iterations: 7h 6m 46s. +[2026-03-26 03:53:51,115][__main__][INFO] - Starting iteration 398. +[2026-03-26 03:53:51,524][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 39 and human policies 1. +[2026-03-26 03:53:51,525][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:54:07,570][__main__][INFO] - Number of regex retries in iteration 398: 0 +[2026-03-26 03:54:07,571][__main__][INFO] - agents played in iteration 398 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:54:08,340][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:54:08,364][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:54:08,387][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:54:08,410][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:54:08,411][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:54:08,411][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:54:09,140][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:54:09,593][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:54:10,096][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:54:10,595][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:54:11,094][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:54:11,593][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:54:12,094][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:54:12,597][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:54:13,096][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:54:13,594][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:54:14,094][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:54:14,593][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:54:15,092][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:54:15,590][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:54:16,088][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:54:16,585][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:54:17,082][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:54:17,584][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:54:18,083][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:54:18,579][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:54:19,079][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:54:19,577][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:54:20,079][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:54:20,580][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:54:21,078][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:54:21,575][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:54:22,076][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:54:22,573][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:54:23,069][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:54:23,565][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:54:24,065][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:54:24,563][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:54:25,063][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:54:25,563][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:54:26,063][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:54:26,572][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:54:27,070][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:54:27,572][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:54:28,074][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:54:28,575][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:54:29,080][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:54:29,578][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:54:30,076][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:54:30,575][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:54:31,075][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:54:31,582][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:54:32,085][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:54:32,588][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:54:33,089][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:54:33,587][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:54:34,088][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:54:34,591][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:54:35,096][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:54:35,597][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:54:36,098][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:54:36,600][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:54:37,101][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:54:37,600][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:54:38,099][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:54:38,598][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:54:39,097][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:54:39,598][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:54:40,098][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:54:40,595][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:54:41,094][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10183 tokens. +[2026-03-26 03:54:41,864][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 62.32%, ΔTime: 00:00:32 +[2026-03-26 03:54:42,690][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:54:42,692][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:54:42,694][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:54:43,596][__main__][INFO] - Iteration 399 took 52s (30.81% Gen, 67.45% Train). Generation: 16s, Training: 35s. Estimated remaining time: 37h 29m 43s. Estimated total time: 43h 23m 38s. Time estimates for 10 more iterations: 8m 40s, 100 more iterations: 1h 26m 47s, 500 more iterations: 7h 13m 56s. +[2026-03-26 03:54:43,599][__main__][INFO] - Starting iteration 399. +[2026-03-26 03:54:44,009][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 39 and human policies 1. +[2026-03-26 03:54:44,009][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:54:49,213][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 03:55:00,786][__main__][INFO] - Number of regex retries in iteration 399: 1 +[2026-03-26 03:55:00,786][__main__][INFO] - agents played in iteration 399 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:55:01,554][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:55:01,577][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:55:01,601][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:55:01,624][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:55:01,624][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:55:01,625][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:55:02,351][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:55:02,806][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:55:03,308][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:55:03,806][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:55:04,304][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:55:04,801][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:55:05,299][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:55:05,797][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:55:06,295][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:55:06,790][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:55:07,287][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:55:07,782][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:55:08,277][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:55:08,771][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:55:09,270][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:55:09,768][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:55:10,263][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:55:10,760][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:55:11,256][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:55:11,761][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:55:12,260][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:55:12,759][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:55:13,264][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:55:13,761][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:55:14,258][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:55:14,758][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:55:15,268][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:55:15,769][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:55:16,270][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:55:16,769][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:55:17,270][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:55:17,773][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:55:18,276][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:55:18,795][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:55:19,294][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:55:19,797][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:55:20,297][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:55:20,794][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:55:21,297][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:55:21,804][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:55:22,304][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:55:22,803][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:55:23,303][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:55:23,803][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:55:24,303][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:55:24,804][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:55:25,304][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:55:25,805][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:55:26,313][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:55:26,817][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:55:27,322][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:55:27,830][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:55:28,334][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:55:28,835][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:55:29,335][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:55:29,836][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:55:30,336][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:55:30,833][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:55:31,337][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:55:31,835][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:55:32,335][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:55:32,833][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:55:33,331][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:55:33,827][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:55:34,325][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10064 tokens. +[2026-03-26 03:55:35,089][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.33%, ΔTime: 00:00:32 +[2026-03-26 03:55:35,951][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:55:35,953][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:55:35,955][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:55:37,051][__main__][INFO] - Iteration 400 took 53s (31.63% Gen, 66.30% Train). Generation: 16s, Training: 35s. Estimated remaining time: 38h 17m 22s. Estimated total time: 44h 12m 10s. Time estimates for 10 more iterations: 8m 50s, 100 more iterations: 1h 28m 24s, 500 more iterations: 7h 22m 1s. +[2026-03-26 03:55:37,054][__main__][INFO] - Starting iteration 400. +[2026-03-26 03:55:37,462][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 39 and human policies 1. +[2026-03-26 03:55:37,463][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:55:41,893][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 03:55:53,273][__main__][INFO] - Number of regex retries in iteration 400: 1 +[2026-03-26 03:55:53,274][__main__][INFO] - agents played in iteration 400 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:55:54,041][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:55:54,065][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:55:54,088][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:55:54,111][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:55:54,112][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:55:54,113][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:55:54,845][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:55:55,298][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:55:55,817][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:55:56,318][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:55:56,819][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:55:57,319][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:55:57,818][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:55:58,317][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:55:58,824][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:55:59,326][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:55:59,825][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:56:00,323][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:56:00,820][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:56:01,319][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:56:01,826][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:56:02,326][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:56:02,827][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:56:03,327][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:56:03,825][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:56:04,323][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:56:04,821][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:56:05,320][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:56:05,828][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:56:06,327][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:56:06,827][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:56:07,325][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:56:07,821][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:56:08,320][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:56:08,820][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:56:09,319][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:56:09,816][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:56:10,314][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:56:10,811][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:56:11,308][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:56:11,806][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:56:12,303][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:56:12,801][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:56:13,299][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:56:13,798][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:56:14,301][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:56:14,799][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:56:15,299][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:56:15,798][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:56:16,310][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:56:16,816][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:56:17,319][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:56:17,821][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:56:18,324][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:56:18,829][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:56:19,361][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:56:19,866][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:56:20,375][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:56:20,881][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:56:21,386][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:56:21,890][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:56:22,397][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:56:22,898][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:56:23,398][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:56:23,897][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:56:24,399][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:56:24,898][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:56:25,396][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:56:25,896][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:56:26,395][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:56:26,894][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10104 tokens. +[2026-03-26 03:56:27,663][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 03:56:28,428][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:56:28,430][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:56:28,434][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:56:30,118][__main__][INFO] - Iteration 401 took 52s (30.03% Gen, 66.77% Train). Generation: 15s, Training: 35s. Estimated remaining time: 37h 57m 8s. Estimated total time: 43h 52m 49s. Time estimates for 10 more iterations: 8m 46s, 100 more iterations: 1h 27m 45s, 500 more iterations: 7h 18m 48s. +[2026-03-26 03:56:30,121][__main__][INFO] - Starting iteration 401. +[2026-03-26 03:56:30,559][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 40 and human policies 1. +[2026-03-26 03:56:30,560][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:56:45,886][__main__][INFO] - Number of regex retries in iteration 401: 0 +[2026-03-26 03:56:45,887][__main__][INFO] - agents played in iteration 401 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:56:46,648][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:56:46,671][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:56:46,694][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:56:46,717][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:56:46,718][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:56:46,718][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:56:47,449][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:56:47,903][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:56:48,420][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:56:48,920][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:56:49,420][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:56:49,919][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:56:50,421][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:56:50,924][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:56:51,422][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:56:51,926][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:56:52,424][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:56:52,924][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:56:53,424][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:56:53,922][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:56:54,421][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:56:54,920][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:56:55,419][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:56:55,916][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:56:56,413][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:56:56,911][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:56:57,407][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:56:57,904][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:56:58,408][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:56:58,905][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:56:59,402][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:56:59,900][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:57:00,398][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:57:00,901][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:57:01,404][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:57:01,905][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:57:02,404][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:57:02,906][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:57:03,405][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:57:03,902][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:57:04,400][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:57:04,898][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:57:05,395][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:57:05,893][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:57:06,389][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:57:06,890][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:57:07,386][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:57:07,890][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:57:08,394][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:57:08,894][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:57:09,395][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:57:09,892][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:57:10,392][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:57:10,896][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:57:11,397][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:57:11,898][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:57:12,398][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:57:12,898][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:57:13,398][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:57:13,898][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:57:14,396][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:57:14,894][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:57:15,392][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:57:15,890][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:57:16,387][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:57:16,884][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:57:17,385][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:57:17,881][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:57:18,379][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:57:18,876][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:57:19,377][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10011 tokens. +[2026-03-26 03:57:20,141][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.21%, Current % of VRAM taken: 60.63%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 03:57:21,418][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:57:21,420][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:57:21,422][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:57:24,068][__main__][INFO] - Iteration 402 took 53s (28.64% Gen, 66.41% Train). Generation: 15s, Training: 35s. Estimated remaining time: 38h 38m 52s. Estimated total time: 44h 35m 28s. Time estimates for 10 more iterations: 8m 55s, 100 more iterations: 1h 29m 10s, 500 more iterations: 7h 25m 54s. +[2026-03-26 03:57:24,070][__main__][INFO] - Starting iteration 402. +[2026-03-26 03:57:24,472][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 40 and human policies 1. +[2026-03-26 03:57:24,472][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:57:39,557][__main__][INFO] - Number of regex retries in iteration 402: 0 +[2026-03-26 03:57:39,558][__main__][INFO] - agents played in iteration 402 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:57:40,321][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:57:40,345][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:57:40,368][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:57:40,391][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:57:40,391][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:57:40,392][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:57:41,113][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:57:41,581][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:57:42,086][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:57:42,588][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:57:43,089][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:57:43,592][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:57:44,100][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:57:44,602][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:57:45,103][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:57:45,600][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:57:46,098][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:57:46,603][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:57:47,102][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:57:47,603][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:57:48,101][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:57:48,599][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:57:49,096][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:57:49,595][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:57:50,099][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:57:50,599][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:57:51,099][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:57:51,597][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:57:52,096][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:57:52,595][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:57:53,096][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:57:53,595][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:57:54,094][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:57:54,593][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:57:55,093][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:57:55,591][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:57:56,092][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:57:56,589][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:57:57,091][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:57:57,592][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:57:58,088][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:57:58,586][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:57:59,086][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:57:59,591][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:58:00,092][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:58:00,590][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:58:01,089][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:58:01,588][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:58:02,094][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:58:02,594][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:58:03,093][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:58:03,593][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:58:04,092][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:58:04,593][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:58:05,092][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:58:05,589][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:58:06,085][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:58:06,582][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:58:07,080][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:58:07,580][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:58:08,081][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:58:08,580][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:58:09,078][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:58:09,576][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:58:10,074][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:58:10,572][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:58:11,069][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:58:11,572][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:58:12,079][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:58:12,577][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:58:13,079][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10026 tokens. +[2026-03-26 03:58:13,846][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.17%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.26%, ΔTime: 00:00:32 +[2026-03-26 03:58:14,582][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:58:14,584][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:58:14,586][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:58:15,324][__main__][INFO] - Iteration 403 took 50s (29.67% Gen, 68.88% Train). Generation: 15s, Training: 35s. Estimated remaining time: 36h 25m 12s. Estimated total time: 42h 22m 39s. Time estimates for 10 more iterations: 8m 28s, 100 more iterations: 1h 24m 45s, 500 more iterations: 7h 3m 46s. +[2026-03-26 03:58:15,327][__main__][INFO] - Starting iteration 403. +[2026-03-26 03:58:15,732][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 40 and human policies 1. +[2026-03-26 03:58:15,733][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:58:25,660][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 03:58:31,201][__main__][INFO] - Number of regex retries in iteration 403: 1 +[2026-03-26 03:58:31,202][__main__][INFO] - agents played in iteration 403 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:58:31,968][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:58:31,992][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:58:32,015][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:58:32,038][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:58:32,039][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:58:32,039][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:58:32,768][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:58:33,221][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:58:33,733][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:58:34,233][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:58:34,730][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:58:35,227][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:58:35,723][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:58:36,220][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:58:36,718][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:58:37,218][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:58:37,717][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:58:38,216][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:58:38,715][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:58:39,213][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:58:39,711][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:58:40,209][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:58:40,707][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:58:41,205][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:58:41,702][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:58:42,199][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:58:42,697][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:58:43,195][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:58:43,699][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:58:44,198][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:58:44,703][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:58:45,200][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:58:45,702][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:58:46,199][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:58:46,696][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:58:47,193][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:58:47,690][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:58:48,186][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:58:48,682][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:58:49,179][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:58:49,677][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:58:50,174][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:58:50,675][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:58:51,172][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:58:51,668][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:58:52,164][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:58:52,660][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:58:53,156][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:58:53,659][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:58:54,161][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:58:54,660][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:58:55,159][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:58:55,659][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:58:56,158][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:58:56,658][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:58:57,159][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:58:57,660][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:58:58,161][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:58:58,660][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:58:59,161][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:58:59,660][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:59:00,158][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:59:00,656][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:59:01,156][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:59:01,653][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:59:02,152][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:59:02,649][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:59:03,147][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:59:03,650][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:59:04,151][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:59:04,650][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9987 tokens. +[2026-03-26 03:59:05,420][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.04%, Current % of VRAM taken: 60.47%, Block Peak % of device VRAM: 61.92%, ΔTime: 00:00:32 +[2026-03-26 03:59:06,183][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:59:06,186][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:59:06,188][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:59:06,932][__main__][INFO] - Iteration 404 took 51s (30.21% Gen, 68.33% Train). Generation: 15s, Training: 34s. Estimated remaining time: 36h 41m 43s. Estimated total time: 42h 40m 2s. Time estimates for 10 more iterations: 8m 32s, 100 more iterations: 1h 25m 20s, 500 more iterations: 7h 6m 40s. +[2026-03-26 03:59:06,934][__main__][INFO] - Starting iteration 404. +[2026-03-26 03:59:07,346][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 40 and human policies 1. +[2026-03-26 03:59:07,347][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 03:59:16,819][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 03:59:23,698][__main__][INFO] - Number of regex retries in iteration 404: 1 +[2026-03-26 03:59:23,699][__main__][INFO] - agents played in iteration 404 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 03:59:24,468][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:59:24,491][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:59:24,514][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:59:24,537][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 03:59:24,538][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 03:59:24,538][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 03:59:25,275][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 03:59:25,727][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 03:59:26,230][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 03:59:26,728][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 03:59:27,225][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 03:59:27,722][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 03:59:28,220][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 03:59:28,719][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 03:59:29,223][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 03:59:29,720][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 03:59:30,218][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 03:59:30,716][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 03:59:31,214][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 03:59:31,714][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 03:59:32,212][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 03:59:32,709][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 03:59:33,206][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 03:59:33,704][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 03:59:34,201][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 03:59:34,698][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 03:59:35,206][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 03:59:35,704][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 03:59:36,201][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 03:59:36,700][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 03:59:37,199][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 03:59:37,701][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 03:59:38,199][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 03:59:38,701][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 03:59:39,198][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 03:59:39,695][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 03:59:40,196][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 03:59:40,698][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 03:59:41,199][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 03:59:41,696][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 03:59:42,195][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 03:59:42,694][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 03:59:43,190][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 03:59:43,687][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 03:59:44,187][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 03:59:44,688][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 03:59:45,190][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 03:59:45,693][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 03:59:46,194][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 03:59:46,698][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 03:59:47,200][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 03:59:47,701][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 03:59:48,205][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 03:59:48,704][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 03:59:49,202][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 03:59:49,715][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 03:59:50,220][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 03:59:50,721][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 03:59:51,221][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 03:59:51,719][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 03:59:52,222][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 03:59:52,720][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 03:59:53,217][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 03:59:53,714][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 03:59:54,214][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 03:59:54,713][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 03:59:55,211][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 03:59:55,710][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 03:59:56,208][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 03:59:56,709][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 03:59:57,211][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10044 tokens. +[2026-03-26 03:59:57,983][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.05%, ΔTime: 00:00:32 +[2026-03-26 03:59:58,735][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 03:59:58,738][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 03:59:58,739][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 03:59:59,628][__main__][INFO] - Iteration 405 took 52s (31.27% Gen, 67.02% Train). Generation: 16s, Training: 35s. Estimated remaining time: 37h 34m 56s. Estimated total time: 43h 34m 7s. Time estimates for 10 more iterations: 8m 42s, 100 more iterations: 1h 27m 8s, 500 more iterations: 7h 15m 41s. +[2026-03-26 03:59:59,630][__main__][INFO] - Starting iteration 405. +[2026-03-26 04:00:00,037][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 40 and human policies 1. +[2026-03-26 04:00:00,038][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:00:25,220][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:00:36,722][__main__][INFO] - Number of regex retries in iteration 405: 1 +[2026-03-26 04:00:36,723][__main__][INFO] - agents played in iteration 405 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:00:37,487][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:00:37,511][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:00:37,534][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:00:37,557][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:00:37,557][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:00:37,558][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:00:38,288][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:00:38,741][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:00:39,244][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:00:39,741][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:00:40,240][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:00:40,737][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:00:41,235][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:00:41,730][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:00:42,227][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:00:42,726][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:00:43,226][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:00:43,725][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:00:44,230][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:00:44,727][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:00:45,225][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:00:45,726][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:00:46,227][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:00:46,724][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:00:47,221][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:00:47,719][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:00:48,217][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:00:48,714][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:00:49,215][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:00:49,712][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:00:50,212][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:00:50,713][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:00:51,212][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:00:51,728][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:00:52,230][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:00:52,729][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:00:53,228][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:00:53,739][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:00:54,239][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:00:54,739][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:00:55,240][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:00:55,739][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:00:56,238][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:00:56,737][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:00:57,239][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:00:57,747][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:00:58,248][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:00:58,746][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:00:59,243][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:00:59,739][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:01:00,237][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:01:00,735][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:01:01,233][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:01:01,731][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:01:02,237][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:01:02,739][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:01:03,238][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:01:03,742][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:01:04,241][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:01:04,741][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:01:05,240][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:01:05,738][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:01:06,238][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:01:06,736][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:01:07,235][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:01:07,733][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:01:08,231][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:01:08,734][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:01:09,231][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:01:09,729][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:01:10,227][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10019 tokens. +[2026-03-26 04:01:10,993][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.10%, ΔTime: 00:00:32 +[2026-03-26 04:01:11,765][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:01:11,767][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:01:11,769][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:01:12,586][__main__][INFO] - Iteration 406 took 1m 12s (50.57% Gen, 48.31% Train). Generation: 36s, Training: 35s. Estimated remaining time: 54h 27m 5s. Estimated total time: 60h 27m 28s. Time estimates for 10 more iterations: 12m 5s, 100 more iterations: 2h 0m 54s, 500 more iterations: 10h 4m 34s. +[2026-03-26 04:01:12,588][__main__][INFO] - Starting iteration 406. +[2026-03-26 04:01:12,993][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 40 and human policies 1. +[2026-03-26 04:01:12,994][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:01:43,941][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 20 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:01:49,760][__main__][INFO] - Number of regex retries in iteration 406: 1 +[2026-03-26 04:01:49,761][__main__][INFO] - agents played in iteration 406 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:01:50,554][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:01:50,578][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:01:50,601][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:01:50,624][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:01:50,625][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:01:50,625][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:01:51,371][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:01:51,835][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:01:52,344][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:01:52,844][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:01:53,344][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:01:53,845][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:01:54,343][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:01:54,841][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:01:55,340][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:01:55,836][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:01:56,338][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:01:56,836][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:01:57,334][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:01:57,833][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:01:58,331][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:01:58,828][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:01:59,325][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:01:59,836][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:02:00,334][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:02:00,833][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:02:01,333][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:02:01,831][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:02:02,332][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:02:02,836][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:02:03,343][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:02:03,843][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:02:04,345][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:02:04,842][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:02:05,343][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:02:05,840][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:02:06,340][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:02:06,838][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:02:07,336][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:02:07,837][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:02:08,335][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:02:08,834][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:02:09,332][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:02:09,830][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:02:10,328][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:02:10,828][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:02:11,328][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:02:11,825][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:02:12,323][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:02:12,830][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:02:13,334][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:02:13,833][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:02:14,334][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:02:14,834][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:02:15,345][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:02:15,843][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:02:16,339][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:02:16,835][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:02:17,332][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:02:17,827][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:02:18,327][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:02:18,823][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:02:19,319][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:02:19,818][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:02:20,320][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:02:20,819][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:02:21,342][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:02:21,844][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:02:22,344][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:02:22,844][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:02:23,342][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10077 tokens. +[2026-03-26 04:02:24,106][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.19%, ΔTime: 00:00:32 +[2026-03-26 04:02:24,836][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:02:24,838][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:02:24,840][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:02:25,585][__main__][INFO] - Iteration 407 took 1m 12s (50.65% Gen, 48.32% Train). Generation: 36s, Training: 35s. Estimated remaining time: 54h 28m 1s. Estimated total time: 60h 29m 38s. Time estimates for 10 more iterations: 12m 5s, 100 more iterations: 2h 0m 59s, 500 more iterations: 10h 4m 56s. +[2026-03-26 04:02:25,587][__main__][INFO] - Starting iteration 407. +[2026-03-26 04:02:26,012][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 40 and human policies 1. +[2026-03-26 04:02:26,013][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:02:37,960][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:02:42,796][__main__][INFO] - Number of regex retries in iteration 407: 1 +[2026-03-26 04:02:42,797][__main__][INFO] - agents played in iteration 407 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:02:43,568][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:02:43,592][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:02:43,616][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:02:43,639][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:02:43,640][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:02:43,640][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:02:44,371][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:02:44,827][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:02:45,333][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:02:45,833][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:02:46,338][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:02:46,846][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:02:47,348][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:02:47,851][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:02:48,356][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:02:48,859][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:02:49,369][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:02:49,868][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:02:50,372][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:02:50,871][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:02:51,370][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:02:51,877][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:02:52,380][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:02:52,888][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:02:53,390][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:02:53,901][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:02:54,407][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:02:54,914][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:02:55,414][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:02:55,912][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:02:56,409][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:02:56,907][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:02:57,405][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:02:57,902][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:02:58,400][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:02:58,899][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:02:59,397][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:02:59,896][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:03:00,397][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:03:00,903][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:03:01,402][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:03:01,901][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:03:02,404][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:03:02,906][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:03:03,404][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:03:03,904][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:03:04,403][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:03:04,901][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:03:05,397][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:03:05,895][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:03:06,394][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:03:06,890][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:03:07,388][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:03:07,885][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:03:08,386][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:03:08,887][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:03:09,385][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:03:09,884][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:03:10,382][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:03:10,880][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:03:11,379][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:03:11,876][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:03:12,374][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:03:12,874][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:03:13,375][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:03:13,873][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:03:14,375][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:03:14,877][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:03:15,378][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:03:15,878][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:03:16,376][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10173 tokens. +[2026-03-26 04:03:17,150][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.31%, ΔTime: 00:00:32 +[2026-03-26 04:03:17,882][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:03:17,884][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:03:17,885][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:03:18,556][__main__][INFO] - Iteration 408 took 52s (31.94% Gen, 66.78% Train). Generation: 16s, Training: 35s. Estimated remaining time: 37h 44m 45s. Estimated total time: 43h 47m 15s. Time estimates for 10 more iterations: 8m 45s, 100 more iterations: 1h 27m 34s, 500 more iterations: 7h 17m 52s. +[2026-03-26 04:03:18,558][__main__][INFO] - Starting iteration 408. +[2026-03-26 04:03:18,975][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 40 and human policies 1. +[2026-03-26 04:03:18,976][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:03:39,926][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls + +Given the values: +- You value hats at 10, books at 1, and balls at 1. +- Bob values hats at 1, books at 10, and balls at 10. + +Since you have a high value for hats and Bob has a high value for books and balls, taking all the hats would maximize your points in this round. Splitting the hats would dilute your total value, so it's best to take all 10 hats and let Bob take the books and balls. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:03:45,559][__main__][INFO] - Number of regex retries in iteration 408: 1 +[2026-03-26 04:03:45,560][__main__][INFO] - agents played in iteration 408 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:03:46,345][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:03:46,369][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:03:46,392][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:03:46,416][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:03:46,416][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:03:46,417][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:03:47,146][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:03:47,602][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:03:48,107][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:03:48,604][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:03:49,102][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:03:49,601][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:03:50,102][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:03:50,601][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:03:51,105][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:03:51,606][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:03:52,104][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:03:52,602][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:03:53,104][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:03:53,609][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:03:54,111][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:03:54,609][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:03:55,108][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:03:55,608][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:03:56,107][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:03:56,604][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:03:57,104][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:03:57,604][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:03:58,103][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:03:58,605][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:03:59,105][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:03:59,603][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:04:00,101][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:04:00,600][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:04:01,114][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:04:01,613][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:04:02,114][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:04:02,612][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:04:03,111][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:04:03,608][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:04:04,106][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:04:04,604][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:04:05,102][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:04:05,601][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:04:06,099][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:04:06,598][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:04:07,096][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:04:07,595][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:04:08,096][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:04:08,598][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:04:09,095][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:04:09,593][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:04:10,092][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:04:10,588][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:04:11,085][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:04:11,582][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:04:12,080][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:04:12,578][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:04:13,078][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:04:13,576][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:04:14,075][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:04:14,573][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:04:15,071][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:04:15,569][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:04:16,067][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:04:16,589][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:04:17,095][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:04:17,593][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:04:18,097][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:04:18,597][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:04:19,101][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9998 tokens. +[2026-03-26 04:04:19,878][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.11%, ΔTime: 00:00:32 +[2026-03-26 04:04:20,629][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:04:20,631][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:04:20,633][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:04:21,393][__main__][INFO] - Iteration 409 took 1m 2s (42.59% Gen, 56.19% Train). Generation: 26s, Training: 35s. Estimated remaining time: 45h 57m 22s. Estimated total time: 52h 0m 55s. Time estimates for 10 more iterations: 10m 24s, 100 more iterations: 1h 44m 1s, 500 more iterations: 8h 40m 9s. +[2026-03-26 04:04:21,395][__main__][INFO] - Starting iteration 409. +[2026-03-26 04:04:21,829][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 40 and human policies 1. +[2026-03-26 04:04:21,830][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:04:33,851][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:04:38,965][__main__][INFO] - Number of regex retries in iteration 409: 1 +[2026-03-26 04:04:38,966][__main__][INFO] - agents played in iteration 409 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:04:39,754][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:04:39,778][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:04:39,802][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:04:39,826][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:04:39,827][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:04:39,827][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:04:40,558][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:04:41,014][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:04:41,520][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:04:42,023][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:04:42,529][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:04:43,038][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:04:43,545][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:04:44,050][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:04:44,553][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:04:45,051][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:04:45,551][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:04:46,051][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:04:46,550][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:04:47,049][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:04:47,548][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:04:48,048][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:04:48,548][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:04:49,046][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:04:49,547][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:04:50,045][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:04:50,544][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:04:51,042][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:04:51,543][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:04:52,043][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:04:52,541][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:04:53,041][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:04:53,540][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:04:54,039][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:04:54,537][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:04:55,036][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:04:55,535][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:04:56,034][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:04:56,543][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:04:57,043][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:04:57,543][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:04:58,045][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:04:58,547][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:04:59,044][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:04:59,542][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:05:00,039][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:05:00,541][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:05:01,038][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:05:01,538][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:05:02,035][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:05:02,532][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:05:03,029][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:05:03,526][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:05:04,023][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:05:04,519][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:05:05,017][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:05:05,516][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:05:06,015][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:05:06,514][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:05:07,010][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:05:07,507][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:05:08,003][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:05:08,501][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:05:08,999][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:05:09,497][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:05:09,996][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:05:10,494][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:05:10,993][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:05:11,490][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:05:11,987][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:05:12,484][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10092 tokens. +[2026-03-26 04:05:13,249][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.29%, ΔTime: 00:00:32 +[2026-03-26 04:05:13,976][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:05:13,978][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:05:13,980][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:05:14,721][__main__][INFO] - Iteration 410 took 52s (32.40% Gen, 66.20% Train). Generation: 17s, Training: 35s. Estimated remaining time: 38h 0m 11s. Estimated total time: 44h 4m 37s. Time estimates for 10 more iterations: 8m 48s, 100 more iterations: 1h 28m 9s, 500 more iterations: 7h 20m 46s. +[2026-03-26 04:05:14,723][__main__][INFO] - Starting iteration 410. +[2026-03-26 04:05:15,165][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 40 and human policies 1. +[2026-03-26 04:05:15,165][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:05:37,141][__main__][INFO] - Number of regex retries in iteration 410: 0 +[2026-03-26 04:05:37,141][__main__][INFO] - agents played in iteration 410 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:05:37,954][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:05:37,978][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:05:38,001][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:05:38,025][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:05:38,025][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:05:38,026][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:05:38,746][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:05:39,199][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:05:39,701][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:05:40,204][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:05:40,701][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:05:41,199][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:05:41,701][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:05:42,199][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:05:42,696][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:05:43,194][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:05:43,691][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:05:44,188][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:05:44,685][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:05:45,181][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:05:45,679][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:05:46,176][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:05:46,674][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:05:47,172][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:05:47,673][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:05:48,173][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:05:48,669][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:05:49,165][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:05:49,661][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:05:50,160][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:05:50,662][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:05:51,166][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:05:51,665][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:05:52,164][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:05:52,663][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:05:53,161][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:05:53,663][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:05:54,161][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:05:54,660][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:05:55,158][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:05:55,654][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:05:56,149][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:05:56,645][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:05:57,140][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:05:57,636][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:05:58,131][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:05:58,626][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:05:59,123][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:05:59,620][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:06:00,117][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:06:00,615][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:06:01,112][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:06:01,607][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:06:02,104][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:06:02,601][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:06:03,099][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:06:03,599][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:06:04,104][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:06:04,626][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:06:05,128][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:06:05,628][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:06:06,129][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:06:06,632][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:06:07,132][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:06:07,630][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:06:08,127][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:06:08,623][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:06:09,118][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:06:09,622][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:06:10,124][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:06:10,629][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10020 tokens. +[2026-03-26 04:06:11,393][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.21%, ΔTime: 00:00:32 +[2026-03-26 04:06:12,124][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:06:12,127][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:06:12,128][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:06:13,611][__main__][INFO] - Iteration 411 took 58s (37.60% Gen, 59.86% Train). Generation: 21s, Training: 34s. Estimated remaining time: 42h 36m 56s. Estimated total time: 48h 42m 21s. Time estimates for 10 more iterations: 9m 44s, 100 more iterations: 1h 37m 24s, 500 more iterations: 8h 7m 3s. +[2026-03-26 04:06:13,613][__main__][INFO] - Starting iteration 411. +[2026-03-26 04:06:14,020][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 41 and human policies 1. +[2026-03-26 04:06:14,021][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:06:18,424][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:06:29,020][__main__][INFO] - Number of regex retries in iteration 411: 1 +[2026-03-26 04:06:29,021][__main__][INFO] - agents played in iteration 411 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:06:29,810][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:06:29,834][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:06:29,858][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:06:29,882][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:06:29,883][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:06:29,884][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:06:30,634][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:06:31,089][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:06:31,590][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:06:32,088][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:06:32,587][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:06:33,085][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:06:33,583][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:06:34,086][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:06:34,584][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:06:35,080][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:06:35,578][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:06:36,076][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:06:36,575][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:06:37,074][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:06:37,573][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:06:38,070][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:06:38,567][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:06:39,067][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:06:39,564][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:06:40,060][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:06:40,558][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:06:41,055][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:06:41,562][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:06:42,063][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:06:42,562][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:06:43,060][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:06:43,558][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:06:44,061][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:06:44,559][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:06:45,058][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:06:45,556][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:06:46,057][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:06:46,556][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:06:47,052][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:06:47,548][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:06:48,048][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:06:48,546][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:06:49,045][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:06:49,545][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:06:50,042][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:06:50,542][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:06:51,038][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:06:51,534][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:06:52,029][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:06:52,526][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:06:53,022][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:06:53,519][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:06:54,014][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:06:54,510][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:06:55,008][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:06:55,504][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:06:56,001][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:06:56,498][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:06:56,996][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:06:57,494][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:06:57,990][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:06:58,487][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:06:58,984][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:06:59,482][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:06:59,979][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:07:00,476][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:07:00,974][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:07:01,472][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:07:01,972][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:07:02,475][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10049 tokens. +[2026-03-26 04:07:03,261][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.19%, Current % of VRAM taken: 60.61%, Block Peak % of device VRAM: 62.12%, ΔTime: 00:00:32 +[2026-03-26 04:07:22,727][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:07:22,729][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:07:22,731][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:07:25,180][__main__][INFO] - Iteration 412 took 1m 11s (21.08% Gen, 75.48% Train). Generation: 14s, Training: 53s. Estimated remaining time: 53h 11m 25s. Estimated total time: 59h 18m 1s. Time estimates for 10 more iterations: 11m 51s, 100 more iterations: 1h 58m 36s, 500 more iterations: 9h 53m 0s. +[2026-03-26 04:07:25,182][__main__][INFO] - Starting iteration 412. +[2026-03-26 04:07:25,586][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 41 and human policies 1. +[2026-03-26 04:07:25,587][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:07:41,985][__main__][INFO] - Number of regex retries in iteration 412: 0 +[2026-03-26 04:07:41,986][__main__][INFO] - agents played in iteration 412 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:07:42,755][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:07:42,779][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:07:42,802][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:07:42,826][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:07:42,826][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:07:42,827][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:07:43,559][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:07:44,011][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:07:44,514][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:07:45,012][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:07:45,518][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:07:46,018][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:07:46,523][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:07:47,021][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:07:47,520][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:07:48,017][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:07:48,522][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:07:49,020][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:07:49,519][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:07:50,017][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:07:50,517][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:07:51,014][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:07:51,513][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:07:52,019][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:07:52,525][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:07:53,025][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:07:53,526][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:07:54,024][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:07:54,551][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:07:55,048][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:07:55,551][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:07:56,048][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:07:56,545][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:07:57,047][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:07:57,545][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:07:58,047][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:07:58,548][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:07:59,048][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:07:59,556][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:08:00,058][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:08:00,558][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:08:01,060][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:08:01,561][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:08:02,062][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:08:02,561][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:08:03,061][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:08:03,561][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:08:04,060][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:08:04,562][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:08:05,060][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:08:05,563][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:08:06,059][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:08:06,557][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:08:07,056][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:08:07,554][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:08:08,052][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:08:08,548][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:08:09,046][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:08:09,544][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:08:10,042][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:08:10,548][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:08:11,046][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:08:11,544][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:08:12,047][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:08:12,552][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:08:13,053][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:08:13,552][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:08:14,050][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:08:14,551][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:08:15,049][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:08:15,550][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10078 tokens. +[2026-03-26 04:08:16,321][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.14%, ΔTime: 00:00:32 +[2026-03-26 04:08:17,077][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:08:17,082][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:08:17,084][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:08:17,792][__main__][INFO] - Iteration 413 took 52s (31.41% Gen, 67.23% Train). Generation: 16s, Training: 35s. Estimated remaining time: 37h 22m 53s. Estimated total time: 43h 30m 22s. Time estimates for 10 more iterations: 8m 42s, 100 more iterations: 1h 27m 0s, 500 more iterations: 7h 15m 3s. +[2026-03-26 04:08:17,795][__main__][INFO] - Starting iteration 413. +[2026-03-26 04:08:18,200][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 41 and human policies 1. +[2026-03-26 04:08:18,201][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:08:33,642][__main__][INFO] - Number of regex retries in iteration 413: 0 +[2026-03-26 04:08:33,643][__main__][INFO] - agents played in iteration 413 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:08:34,405][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:08:34,429][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:08:34,452][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:08:34,475][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:08:34,475][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:08:34,476][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:08:35,202][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:08:35,655][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:08:36,157][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:08:36,655][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:08:37,152][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:08:37,649][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:08:38,149][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:08:38,646][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:08:39,142][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:08:39,640][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:08:40,138][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:08:40,634][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:08:41,136][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:08:41,637][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:08:42,140][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:08:42,640][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:08:43,138][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:08:43,635][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:08:44,132][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:08:44,636][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:08:45,132][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:08:45,629][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:08:46,125][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:08:46,622][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:08:47,119][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:08:47,620][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:08:48,143][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:08:48,644][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:08:49,155][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:08:49,657][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:08:50,158][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:08:50,660][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:08:51,161][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:08:51,659][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:08:52,158][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:08:52,657][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:08:53,157][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:08:53,655][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:08:54,154][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:08:54,652][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:08:55,151][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:08:55,648][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:08:56,145][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:08:56,644][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:08:57,141][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:08:57,639][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:08:58,137][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:08:58,634][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:08:59,132][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:08:59,631][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:09:00,129][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:09:00,630][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:09:01,127][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:09:01,625][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:09:02,125][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:09:02,623][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:09:03,121][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:09:03,620][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:09:04,119][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:09:04,619][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:09:05,123][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:09:05,627][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:09:06,133][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:09:06,632][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:09:07,132][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10014 tokens. +[2026-03-26 04:09:07,899][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.08%, ΔTime: 00:00:32 +[2026-03-26 04:09:08,680][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:09:08,682][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:09:08,684][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:09:09,456][__main__][INFO] - Iteration 414 took 51s (30.13% Gen, 68.36% Train). Generation: 15s, Training: 35s. Estimated remaining time: 36h 34m 28s. Estimated total time: 42h 42m 48s. Time estimates for 10 more iterations: 8m 32s, 100 more iterations: 1h 25m 25s, 500 more iterations: 7h 7m 8s. +[2026-03-26 04:09:09,458][__main__][INFO] - Starting iteration 414. +[2026-03-26 04:09:09,858][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 41 and human policies 1. +[2026-03-26 04:09:09,858][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:09:25,451][__main__][INFO] - Number of regex retries in iteration 414: 0 +[2026-03-26 04:09:25,452][__main__][INFO] - agents played in iteration 414 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:09:26,229][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:09:26,253][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:09:26,276][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:09:26,300][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:09:26,300][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:09:26,301][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:09:27,039][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:09:27,491][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:09:27,995][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:09:28,495][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:09:29,006][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:09:29,510][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:09:30,008][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:09:30,507][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:09:31,006][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:09:31,509][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:09:32,023][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:09:32,524][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:09:33,024][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:09:33,524][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:09:34,024][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:09:34,523][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:09:35,029][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:09:35,528][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:09:36,028][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:09:36,529][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:09:37,027][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:09:37,525][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:09:38,026][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:09:38,525][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:09:39,028][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:09:39,527][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:09:40,023][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:09:40,522][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:09:41,020][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:09:41,520][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:09:42,026][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:09:42,526][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:09:43,024][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:09:43,522][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:09:44,019][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:09:44,517][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:09:45,014][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:09:45,512][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:09:46,017][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:09:46,515][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:09:47,011][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:09:47,509][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:09:48,008][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:09:48,506][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:09:49,003][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:09:49,504][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:09:50,002][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:09:50,503][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:09:51,008][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:09:51,505][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:09:52,003][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:09:52,500][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:09:52,998][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:09:53,496][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:09:53,994][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:09:54,492][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:09:54,992][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:09:55,497][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:09:55,998][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:09:56,500][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:09:57,029][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:09:57,531][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:09:58,037][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:09:58,539][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:09:59,042][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10017 tokens. +[2026-03-26 04:09:59,823][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 04:10:00,567][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:10:00,569][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:10:00,571][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:10:01,288][__main__][INFO] - Iteration 415 took 51s (30.32% Gen, 68.28% Train). Generation: 15s, Training: 35s. Estimated remaining time: 36h 42m 21s. Estimated total time: 42h 51m 33s. Time estimates for 10 more iterations: 8m 34s, 100 more iterations: 1h 25m 43s, 500 more iterations: 7h 8m 35s. +[2026-03-26 04:10:01,290][__main__][INFO] - Starting iteration 415. +[2026-03-26 04:10:01,690][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 41 and human policies 1. +[2026-03-26 04:10:01,690][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:10:16,967][__main__][INFO] - Number of regex retries in iteration 415: 0 +[2026-03-26 04:10:16,968][__main__][INFO] - agents played in iteration 415 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:10:17,744][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:10:17,768][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:10:17,792][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:10:17,816][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:10:17,817][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:10:17,817][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:10:18,556][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:10:19,011][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:10:19,517][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:10:20,016][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:10:20,518][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:10:21,024][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:10:21,535][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:10:22,042][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:10:22,541][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:10:23,040][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:10:23,542][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:10:24,040][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:10:24,537][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:10:25,036][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:10:25,535][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:10:26,033][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:10:26,530][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:10:27,028][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:10:27,526][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:10:28,024][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:10:28,522][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:10:29,030][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:10:29,528][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:10:30,026][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:10:30,528][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:10:31,026][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:10:31,525][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:10:32,024][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:10:32,522][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:10:33,020][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:10:33,518][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:10:34,017][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:10:34,516][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:10:35,014][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:10:35,511][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:10:36,009][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:10:36,507][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:10:37,004][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:10:37,500][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:10:38,001][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:10:38,498][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:10:39,005][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:10:39,506][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:10:40,004][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:10:40,501][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:10:41,002][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:10:41,504][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:10:42,004][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:10:42,502][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:10:43,000][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:10:43,496][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:10:43,993][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:10:44,490][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:10:44,993][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:10:45,491][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:10:45,998][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:10:46,502][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:10:47,003][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:10:47,503][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:10:48,002][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:10:48,503][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:10:49,003][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:10:49,504][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:10:50,005][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:10:50,506][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10042 tokens. +[2026-03-26 04:10:51,296][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.17%, ΔTime: 00:00:32 +[2026-03-26 04:10:52,030][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:10:52,033][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:10:52,034][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:10:52,796][__main__][INFO] - Iteration 416 took 51s (29.89% Gen, 68.61% Train). Generation: 15s, Training: 35s. Estimated remaining time: 36h 25m 15s. Estimated total time: 42h 35m 19s. Time estimates for 10 more iterations: 8m 31s, 100 more iterations: 1h 25m 10s, 500 more iterations: 7h 5m 53s. +[2026-03-26 04:10:52,798][__main__][INFO] - Starting iteration 416. +[2026-03-26 04:10:53,213][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 41 and human policies 1. +[2026-03-26 04:10:53,214][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:11:10,214][__main__][INFO] - Number of regex retries in iteration 416: 0 +[2026-03-26 04:11:10,214][__main__][INFO] - agents played in iteration 416 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:11:10,981][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:11:11,004][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:11:11,027][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:11:11,050][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:11:11,051][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:11:11,052][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:11:11,792][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:11:12,246][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:11:12,747][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:11:13,243][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:11:13,745][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:11:14,245][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:11:14,740][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:11:15,238][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:11:15,736][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:11:16,232][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:11:16,733][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:11:17,229][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:11:17,725][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:11:18,220][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:11:18,718][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:11:19,214][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:11:19,711][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:11:20,209][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:11:20,708][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:11:21,217][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:11:21,717][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:11:22,216][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:11:22,718][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:11:23,216][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:11:23,718][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:11:24,218][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:11:24,719][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:11:25,220][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:11:25,725][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:11:26,223][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:11:26,722][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:11:27,220][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:11:27,718][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:11:28,216][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:11:28,714][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:11:29,212][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:11:29,709][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:11:30,207][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:11:30,704][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:11:31,200][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:11:31,696][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:11:32,197][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:11:32,694][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:11:33,193][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:11:33,690][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:11:34,187][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:11:34,684][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:11:35,181][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:11:35,679][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:11:36,179][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:11:36,685][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:11:37,188][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:11:37,687][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:11:38,184][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:11:38,688][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:11:39,190][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:11:39,688][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:11:40,186][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:11:40,684][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:11:41,182][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:11:41,679][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:11:42,176][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:11:42,674][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:11:43,174][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:11:43,671][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10042 tokens. +[2026-03-26 04:11:44,446][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.04%, Current % of VRAM taken: 60.47%, Block Peak % of device VRAM: 62.08%, ΔTime: 00:00:32 +[2026-03-26 04:12:05,391][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:12:05,394][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:12:05,395][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:12:06,428][__main__][INFO] - Iteration 417 took 1m 13s (23.22% Gen, 75.37% Train). Generation: 17s, Training: 55s. Estimated remaining time: 54h 49m 28s. Estimated total time: 61h 0m 46s. Time estimates for 10 more iterations: 12m 12s, 100 more iterations: 2h 2m 1s, 500 more iterations: 10h 10m 7s. +[2026-03-26 04:12:06,430][__main__][INFO] - Starting iteration 417. +[2026-03-26 04:12:06,841][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 41 and human policies 1. +[2026-03-26 04:12:06,842][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:12:23,817][__main__][INFO] - Number of regex retries in iteration 417: 0 +[2026-03-26 04:12:23,817][__main__][INFO] - agents played in iteration 417 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:12:24,594][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:12:24,617][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:12:24,641][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:12:24,664][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:12:24,664][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:12:24,665][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:12:25,402][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:12:25,854][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:12:26,361][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:12:26,861][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:12:27,360][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:12:27,858][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:12:28,361][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:12:28,859][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:12:29,357][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:12:29,858][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:12:30,356][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:12:30,857][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:12:31,358][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:12:31,858][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:12:32,358][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:12:32,857][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:12:33,357][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:12:33,861][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:12:34,359][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:12:34,859][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:12:35,357][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:12:35,855][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:12:36,356][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:12:36,857][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:12:37,356][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:12:37,854][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:12:38,353][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:12:38,853][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:12:39,349][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:12:39,850][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:12:40,346][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:12:40,845][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:12:41,342][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:12:41,839][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:12:42,336][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:12:42,834][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:12:43,340][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:12:43,838][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:12:44,337][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:12:44,838][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:12:45,337][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:12:45,835][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:12:46,336][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:12:46,836][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:12:47,332][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:12:47,830][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:12:48,330][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:12:48,826][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:12:49,323][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:12:49,824][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:12:50,323][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:12:50,821][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:12:51,320][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:12:51,818][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:12:52,316][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:12:52,813][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:12:53,311][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:12:53,812][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:12:54,316][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:12:54,823][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:12:55,324][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:12:55,826][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:12:56,326][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:12:56,823][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:12:57,328][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10136 tokens. +[2026-03-26 04:12:58,127][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.22%, Current % of VRAM taken: 60.65%, Block Peak % of device VRAM: 62.12%, ΔTime: 00:00:32 +[2026-03-26 04:12:58,900][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:12:58,902][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:12:58,903][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:12:59,637][__main__][INFO] - Iteration 418 took 52s (32.15% Gen, 66.45% Train). Generation: 16s, Training: 35s. Estimated remaining time: 37h 47m 41s. Estimated total time: 43h 59m 52s. Time estimates for 10 more iterations: 8m 47s, 100 more iterations: 1h 27m 59s, 500 more iterations: 7h 19m 58s. +[2026-03-26 04:12:59,639][__main__][INFO] - Starting iteration 418. +[2026-03-26 04:13:00,042][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 41 and human policies 1. +[2026-03-26 04:13:00,043][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:13:16,024][__main__][INFO] - Number of regex retries in iteration 418: 0 +[2026-03-26 04:13:16,025][__main__][INFO] - agents played in iteration 418 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:13:16,799][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:13:16,823][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:13:16,846][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:13:16,870][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:13:16,871][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:13:16,871][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:13:17,608][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:13:18,062][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:13:18,567][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:13:19,065][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:13:19,564][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:13:20,063][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:13:20,561][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:13:21,072][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:13:21,572][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:13:22,071][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:13:22,569][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:13:23,068][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:13:23,565][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:13:24,064][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:13:24,561][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:13:25,060][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:13:25,564][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:13:26,063][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:13:26,563][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:13:27,061][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:13:27,560][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:13:28,057][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:13:28,554][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:13:29,051][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:13:29,548][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:13:30,047][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:13:30,555][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:13:31,057][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:13:31,562][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:13:32,060][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:13:32,561][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:13:33,059][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:13:33,561][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:13:34,090][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:13:34,591][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:13:35,093][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:13:35,605][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:13:36,110][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:13:36,612][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:13:37,113][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:13:37,615][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:13:38,117][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:13:38,617][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:13:39,116][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:13:39,614][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:13:40,111][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:13:40,610][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:13:41,111][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:13:41,610][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:13:42,112][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:13:42,620][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:13:43,122][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:13:43,625][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:13:44,135][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:13:44,641][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:13:45,148][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:13:45,652][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:13:46,149][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:13:46,654][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:13:47,154][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:13:47,656][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:13:48,157][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:13:48,658][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:13:49,156][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:13:49,658][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10147 tokens. +[2026-03-26 04:13:50,426][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.21%, Current % of VRAM taken: 60.64%, Block Peak % of device VRAM: 62.43%, ΔTime: 00:00:32 +[2026-03-26 04:13:51,175][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:13:51,177][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:13:51,179][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:13:51,957][__main__][INFO] - Iteration 419 took 51s (30.78% Gen, 67.71% Train). Generation: 15s, Training: 35s. Estimated remaining time: 37h 2m 45s. Estimated total time: 43h 15m 49s. Time estimates for 10 more iterations: 8m 39s, 100 more iterations: 1h 26m 31s, 500 more iterations: 7h 12m 38s. +[2026-03-26 04:13:51,959][__main__][INFO] - Starting iteration 419. +[2026-03-26 04:13:52,360][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 41 and human policies 1. +[2026-03-26 04:13:52,361][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:14:06,151][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:14:09,111][__main__][INFO] - Number of regex retries in iteration 419: 1 +[2026-03-26 04:14:09,112][__main__][INFO] - agents played in iteration 419 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:14:09,881][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:14:09,905][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:14:09,928][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:14:09,951][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:14:09,952][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:14:09,952][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:14:10,701][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:14:11,154][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:14:11,658][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:14:12,157][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:14:12,655][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:14:13,154][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:14:13,650][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:14:14,146][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:14:14,644][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:14:15,160][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:14:15,664][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:14:16,162][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:14:16,662][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:14:17,159][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:14:17,659][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:14:18,162][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:14:18,661][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:14:19,160][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:14:19,659][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:14:20,160][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:14:20,658][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:14:21,157][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:14:21,656][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:14:22,155][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:14:22,653][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:14:23,151][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:14:23,648][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:14:24,145][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:14:24,642][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:14:25,141][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:14:25,642][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:14:26,143][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:14:26,641][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:14:27,145][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:14:27,646][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:14:28,143][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:14:28,644][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:14:29,145][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:14:29,649][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:14:30,148][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:14:30,651][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:14:31,148][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:14:31,645][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:14:32,148][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:14:32,646][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:14:33,144][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:14:33,642][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:14:34,140][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:14:34,639][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:14:35,142][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:14:35,642][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:14:36,143][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:14:36,645][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:14:37,142][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:14:37,641][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:14:38,138][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:14:38,636][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:14:39,136][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:14:39,638][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:14:40,138][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:14:40,639][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:14:41,141][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:14:41,642][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:14:42,146][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:14:42,655][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10058 tokens. +[2026-03-26 04:14:43,447][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.20%, Current % of VRAM taken: 60.63%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 04:14:44,248][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:14:44,250][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:14:44,252][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:14:45,162][__main__][INFO] - Iteration 420 took 52s (31.72% Gen, 66.55% Train). Generation: 16s, Training: 35s. Estimated remaining time: 37h 46m 11s. Estimated total time: 44h 0m 8s. Time estimates for 10 more iterations: 8m 48s, 100 more iterations: 1h 28m 0s, 500 more iterations: 7h 20m 1s. +[2026-03-26 04:14:45,166][__main__][INFO] - Starting iteration 420. +[2026-03-26 04:14:45,569][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 41 and human policies 1. +[2026-03-26 04:14:45,570][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:15:02,505][__main__][INFO] - Number of regex retries in iteration 420: 0 +[2026-03-26 04:15:02,506][__main__][INFO] - agents played in iteration 420 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:15:03,293][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:15:03,317][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:15:03,341][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:15:03,364][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:15:03,365][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:15:03,365][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:15:04,096][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:15:04,555][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:15:05,059][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:15:05,559][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:15:06,056][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:15:06,557][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:15:07,060][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:15:07,563][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:15:08,064][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:15:08,560][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:15:09,057][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:15:09,555][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:15:10,052][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:15:10,553][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:15:11,050][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:15:11,544][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:15:12,040][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:15:12,537][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:15:13,035][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:15:13,536][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:15:14,033][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:15:14,530][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:15:15,027][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:15:15,523][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:15:16,024][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:15:16,522][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:15:17,020][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:15:17,517][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:15:18,015][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:15:18,513][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:15:19,011][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:15:19,512][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:15:20,011][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:15:20,508][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:15:21,004][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:15:21,500][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:15:21,997][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:15:22,493][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:15:22,990][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:15:23,487][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:15:23,985][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:15:24,482][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:15:24,994][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:15:25,498][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:15:26,003][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:15:26,503][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:15:27,002][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:15:27,501][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:15:27,998][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:15:28,497][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:15:28,998][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:15:29,496][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:15:29,996][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:15:30,495][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:15:30,992][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:15:31,489][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:15:31,985][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:15:32,485][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:15:32,982][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:15:33,480][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:15:33,982][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:15:34,479][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:15:34,978][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:15:35,479][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:15:35,987][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10067 tokens. +[2026-03-26 04:15:36,755][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.20%, Current % of VRAM taken: 60.63%, Block Peak % of device VRAM: 62.44%, ΔTime: 00:00:32 +[2026-03-26 04:15:37,403][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:15:37,405][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:15:37,407][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:15:39,012][__main__][INFO] - Iteration 421 took 53s (31.69% Gen, 65.30% Train). Generation: 16s, Training: 34s. Estimated remaining time: 38h 17m 20s. Estimated total time: 44h 32m 10s. Time estimates for 10 more iterations: 8m 54s, 100 more iterations: 1h 29m 4s, 500 more iterations: 7h 25m 21s. +[2026-03-26 04:15:39,014][__main__][INFO] - Starting iteration 421. +[2026-03-26 04:15:39,420][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 42 and human policies 1. +[2026-03-26 04:15:39,421][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:16:10,099][__main__][INFO] - Number of regex retries in iteration 421: 0 +[2026-03-26 04:16:10,100][__main__][INFO] - agents played in iteration 421 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:16:10,882][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:16:10,907][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:16:10,931][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:16:10,955][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:16:10,956][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:16:10,957][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:16:11,702][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:16:12,154][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:16:12,659][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:16:13,157][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:16:13,667][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:16:14,171][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:16:14,668][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:16:15,166][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:16:15,664][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:16:16,168][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:16:16,666][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:16:17,162][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:16:17,667][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:16:18,169][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:16:18,669][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:16:19,172][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:16:19,671][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:16:20,177][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:16:20,676][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:16:21,173][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:16:21,671][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:16:22,185][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:16:22,687][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:16:23,186][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:16:23,684][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:16:24,183][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:16:24,681][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:16:25,179][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:16:25,684][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:16:26,182][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:16:26,689][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:16:27,190][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:16:27,689][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:16:28,185][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:16:28,681][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:16:29,177][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:16:29,675][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:16:30,176][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:16:30,676][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:16:31,176][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:16:31,673][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:16:32,169][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:16:32,667][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:16:33,164][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:16:33,661][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:16:34,158][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:16:34,668][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:16:35,166][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:16:35,664][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:16:36,161][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:16:36,662][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:16:37,161][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:16:37,659][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:16:38,160][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:16:38,656][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:16:39,153][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:16:39,653][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:16:40,148][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:16:40,656][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:16:41,154][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:16:41,655][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:16:42,151][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:16:42,649][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:16:43,146][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:16:43,643][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10044 tokens. +[2026-03-26 04:16:44,421][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.03%, Current % of VRAM taken: 60.46%, Block Peak % of device VRAM: 62.06%, ΔTime: 00:00:32 +[2026-03-26 04:16:45,139][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:16:45,142][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:16:45,144][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:16:45,852][__main__][INFO] - Iteration 422 took 1m 6s (46.18% Gen, 52.75% Train). Generation: 30s, Training: 35s. Estimated remaining time: 49h 5m 41s. Estimated total time: 55h 21m 38s. Time estimates for 10 more iterations: 11m 4s, 100 more iterations: 1h 50m 43s, 500 more iterations: 9h 13m 36s. +[2026-03-26 04:16:45,855][__main__][INFO] - Starting iteration 422. +[2026-03-26 04:16:46,277][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 42 and human policies 1. +[2026-03-26 04:16:46,278][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:17:01,682][__main__][INFO] - Number of regex retries in iteration 422: 0 +[2026-03-26 04:17:01,683][__main__][INFO] - agents played in iteration 422 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:17:02,464][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:17:02,487][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:17:02,511][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:17:02,534][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:17:02,534][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:17:02,535][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:17:03,275][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:17:03,731][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:17:04,236][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:17:04,738][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:17:05,236][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:17:05,734][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:17:06,236][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:17:06,736][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:17:07,233][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:17:07,731][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:17:08,234][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:17:08,731][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:17:09,234][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:17:09,735][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:17:10,233][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:17:10,730][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:17:11,227][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:17:11,726][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:17:12,225][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:17:12,730][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:17:13,229][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:17:13,728][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:17:14,229][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:17:14,730][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:17:15,228][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:17:15,727][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:17:16,228][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:17:16,727][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:17:17,225][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:17:17,726][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:17:18,224][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:17:18,730][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:17:19,232][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:17:19,735][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:17:20,239][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:17:20,741][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:17:21,239][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:17:21,737][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:17:22,235][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:17:22,733][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:17:23,232][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:17:23,731][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:17:24,227][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:17:24,728][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:17:25,227][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:17:25,725][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:17:26,222][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:17:26,721][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:17:27,220][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:17:27,740][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:17:28,241][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:17:28,739][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:17:29,239][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:17:29,737][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:17:30,236][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:17:30,736][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:17:31,240][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:17:31,737][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:17:32,233][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:17:32,731][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:17:33,227][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:17:33,724][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:17:34,224][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:17:34,721][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:17:35,217][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10094 tokens. +[2026-03-26 04:17:35,977][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.18%, ΔTime: 00:00:32 +[2026-03-26 04:17:36,777][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:17:36,780][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:17:36,781][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:17:37,628][__main__][INFO] - Iteration 423 took 51s (30.00% Gen, 68.35% Train). Generation: 15s, Training: 35s. Estimated remaining time: 36h 30m 47s. Estimated total time: 42h 47m 36s. Time estimates for 10 more iterations: 8m 33s, 100 more iterations: 1h 25m 35s, 500 more iterations: 7h 7m 56s. +[2026-03-26 04:17:37,631][__main__][INFO] - Starting iteration 423. +[2026-03-26 04:17:38,031][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 42 and human policies 1. +[2026-03-26 04:17:38,032][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:17:58,748][__main__][INFO] - Number of regex retries in iteration 423: 0 +[2026-03-26 04:17:58,749][__main__][INFO] - agents played in iteration 423 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:17:59,533][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:17:59,557][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:17:59,580][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:17:59,603][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:17:59,604][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:17:59,605][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:18:00,346][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:18:00,799][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:18:01,302][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:18:01,800][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:18:02,298][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:18:02,796][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:18:03,294][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:18:03,796][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:18:04,293][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:18:04,789][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:18:05,285][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:18:05,780][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:18:06,277][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:18:06,774][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:18:07,270][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:18:07,765][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:18:08,261][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:18:08,757][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:18:09,255][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:18:09,750][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:18:10,249][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:18:10,746][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:18:11,243][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:18:11,758][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:18:12,256][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:18:12,754][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:18:13,252][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:18:13,751][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:18:14,248][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:18:14,745][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:18:15,243][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:18:15,741][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:18:16,238][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:18:16,736][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:18:17,232][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:18:17,732][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:18:18,226][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:18:18,727][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:18:19,233][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:18:19,734][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:18:20,238][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:18:20,736][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:18:21,234][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:18:21,731][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:18:22,228][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:18:22,726][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:18:23,224][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:18:23,722][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:18:24,222][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:18:24,720][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:18:25,218][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:18:25,718][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:18:26,214][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:18:26,711][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:18:27,216][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:18:27,716][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:18:28,214][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:18:28,711][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:18:29,210][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:18:29,705][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:18:30,201][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:18:30,697][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:18:31,193][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:18:31,689][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:18:32,188][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10019 tokens. +[2026-03-26 04:18:32,968][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.99%, Current % of VRAM taken: 60.42%, Block Peak % of device VRAM: 62.09%, ΔTime: 00:00:32 +[2026-03-26 04:18:33,707][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:18:33,709][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:18:33,711][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:18:34,550][__main__][INFO] - Iteration 424 took 56s (36.66% Gen, 61.86% Train). Generation: 20s, Training: 34s. Estimated remaining time: 40h 48m 12s. Estimated total time: 47h 5m 57s. Time estimates for 10 more iterations: 9m 25s, 100 more iterations: 1h 34m 11s, 500 more iterations: 7h 50m 59s. +[2026-03-26 04:18:34,552][__main__][INFO] - Starting iteration 424. +[2026-03-26 04:18:34,957][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 42 and human policies 1. +[2026-03-26 04:18:34,958][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:18:53,547][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:19:00,437][__main__][INFO] - Number of regex retries in iteration 424: 1 +[2026-03-26 04:19:00,437][__main__][INFO] - agents played in iteration 424 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:19:01,204][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:19:01,228][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:19:01,252][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:19:01,275][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:19:01,275][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:19:01,276][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:19:02,003][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:19:02,458][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:19:02,963][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:19:03,464][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:19:03,975][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:19:04,474][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:19:04,974][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:19:05,472][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:19:05,970][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:19:06,467][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:19:06,975][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:19:07,474][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:19:07,973][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:19:08,476][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:19:08,973][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:19:09,469][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:19:09,966][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:19:10,462][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:19:10,960][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:19:11,457][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:19:11,955][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:19:12,453][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:19:12,949][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:19:13,445][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:19:13,952][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:19:14,450][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:19:14,948][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:19:15,446][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:19:15,945][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:19:16,444][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:19:16,942][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:19:17,440][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:19:17,944][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:19:18,456][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:19:18,958][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:19:19,460][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:19:19,959][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:19:20,460][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:19:20,957][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:19:21,457][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:19:21,956][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:19:22,454][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:19:22,952][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:19:23,460][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:19:23,958][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:19:24,455][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:19:24,954][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:19:25,454][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:19:25,951][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:19:26,453][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:19:26,954][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:19:27,454][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:19:27,955][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:19:28,455][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:19:28,957][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:19:29,462][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:19:29,970][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:19:30,476][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:19:30,980][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:19:31,479][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:19:31,981][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:19:32,483][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:19:32,982][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:19:33,482][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:19:33,984][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10054 tokens. +[2026-03-26 04:19:34,750][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.20%, Current % of VRAM taken: 60.62%, Block Peak % of device VRAM: 62.18%, ΔTime: 00:00:32 +[2026-03-26 04:19:35,727][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:19:35,732][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:19:35,734][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:19:36,527][__main__][INFO] - Iteration 425 took 1m 1s (41.38% Gen, 57.33% Train). Generation: 25s, Training: 35s. Estimated remaining time: 44h 59m 42s. Estimated total time: 51h 18m 30s. Time estimates for 10 more iterations: 10m 15s, 100 more iterations: 1h 42m 37s, 500 more iterations: 8h 33m 5s. +[2026-03-26 04:19:36,529][__main__][INFO] - Starting iteration 425. +[2026-03-26 04:19:36,935][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 42 and human policies 1. +[2026-03-26 04:19:36,935][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:20:03,670][__main__][INFO] - Number of regex retries in iteration 425: 0 +[2026-03-26 04:20:03,671][__main__][INFO] - agents played in iteration 425 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:20:04,433][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:20:04,456][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:20:04,480][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:20:04,503][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:20:04,503][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:20:04,504][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:20:05,234][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:20:05,689][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:20:06,210][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:20:06,710][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:20:07,210][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:20:07,707][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:20:08,210][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:20:08,708][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:20:09,224][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:20:09,724][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:20:10,221][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:20:10,719][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:20:11,218][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:20:11,715][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:20:12,215][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:20:12,713][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:20:13,211][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:20:13,710][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:20:14,208][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:20:14,712][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:20:15,210][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:20:15,709][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:20:16,207][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:20:16,706][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:20:17,206][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:20:17,705][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:20:18,205][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:20:18,710][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:20:19,210][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:20:19,710][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:20:20,215][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:20:20,714][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:20:21,215][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:20:21,711][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:20:22,208][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:20:22,705][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:20:23,206][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:20:23,708][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:20:24,207][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:20:24,705][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:20:25,207][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:20:25,710][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:20:26,214][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:20:26,721][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:20:27,226][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:20:27,728][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:20:28,230][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:20:28,732][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:20:29,235][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:20:29,739][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:20:30,240][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:20:30,742][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:20:31,244][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:20:31,744][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:20:32,245][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:20:32,757][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:20:33,256][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:20:33,757][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:20:34,258][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:20:34,759][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:20:35,262][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:20:35,763][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:20:36,265][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:20:36,787][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:20:37,289][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10140 tokens. +[2026-03-26 04:20:38,058][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.19%, Current % of VRAM taken: 60.62%, Block Peak % of device VRAM: 62.32%, ΔTime: 00:00:32 +[2026-03-26 04:20:38,905][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:20:38,907][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:20:38,909][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:20:39,779][__main__][INFO] - Iteration 426 took 1m 2s (42.54% Gen, 56.07% Train). Generation: 26s, Training: 35s. Estimated remaining time: 46h 2m 23s. Estimated total time: 52h 22m 15s. Time estimates for 10 more iterations: 10m 28s, 100 more iterations: 1h 44m 44s, 500 more iterations: 8h 43m 42s. +[2026-03-26 04:20:39,781][__main__][INFO] - Starting iteration 426. +[2026-03-26 04:20:40,195][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 42 and human policies 1. +[2026-03-26 04:20:40,196][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:20:55,652][__main__][INFO] - Number of regex retries in iteration 426: 0 +[2026-03-26 04:20:55,653][__main__][INFO] - agents played in iteration 426 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:20:56,414][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:20:56,437][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:20:56,461][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:20:56,484][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:20:56,485][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:20:56,485][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:20:57,234][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:20:57,690][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:20:58,196][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:20:58,698][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:20:59,208][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:20:59,715][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:21:00,224][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:21:00,725][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:21:01,227][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:21:01,726][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:21:02,227][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:21:02,727][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:21:03,230][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:21:03,730][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:21:04,230][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:21:04,730][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:21:05,232][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:21:05,735][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:21:06,238][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:21:06,741][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:21:07,244][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:21:07,751][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:21:08,255][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:21:08,759][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:21:09,262][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:21:09,761][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:21:10,262][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:21:10,762][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:21:11,262][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:21:11,762][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:21:12,261][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:21:12,761][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:21:13,262][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:21:13,761][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:21:14,261][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:21:14,764][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:21:15,263][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:21:15,761][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:21:16,261][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:21:16,764][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:21:17,271][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:21:17,780][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:21:18,286][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:21:18,801][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:21:19,307][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:21:19,809][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:21:20,311][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:21:20,813][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:21:21,316][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:21:21,820][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:21:22,320][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:21:22,819][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:21:23,321][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:21:23,820][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:21:24,319][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:21:24,817][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:21:25,316][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:21:25,817][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:21:26,320][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:21:26,819][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:21:27,319][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:21:27,819][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:21:28,320][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:21:28,820][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:21:29,321][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10049 tokens. +[2026-03-26 04:21:30,096][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.15%, ΔTime: 00:00:32 +[2026-03-26 04:21:30,838][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:21:30,841][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:21:30,842][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:21:31,805][__main__][INFO] - Iteration 427 took 51s (29.95% Gen, 68.18% Train). Generation: 15s, Training: 35s. Estimated remaining time: 36h 39m 49s. Estimated total time: 43h 0m 32s. Time estimates for 10 more iterations: 8m 36s, 100 more iterations: 1h 26m 1s, 500 more iterations: 7h 10m 5s. +[2026-03-26 04:21:31,807][__main__][INFO] - Starting iteration 427. +[2026-03-26 04:21:32,207][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 42 and human policies 1. +[2026-03-26 04:21:32,208][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:21:43,188][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls + +Given the per-item values: +- You value hats at 10, books at 1, and balls at 1. +- Bob values hats at 1, books at 10, and balls at 10. + +Since you have a much higher value for hats compared to Bob, and your values for books and balls are much lower than Bob's, it makes sense to take all the hats and let Bob take all the books and balls. This ensures you receive the maximum possible points from the items you value highly. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:21:48,839][__main__][INFO] - Number of regex retries in iteration 427: 1 +[2026-03-26 04:21:48,840][__main__][INFO] - agents played in iteration 427 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:21:49,618][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:21:49,641][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:21:49,665][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:21:49,688][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:21:49,689][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:21:49,689][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:21:50,432][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:21:50,891][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:21:51,399][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:21:51,898][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:21:52,398][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:21:52,897][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:21:53,400][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:21:53,898][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:21:54,397][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:21:54,903][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:21:55,402][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:21:55,901][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:21:56,401][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:21:56,899][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:21:57,398][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:21:57,898][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:21:58,397][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:21:58,896][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:21:59,395][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:21:59,892][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:22:00,391][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:22:00,888][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:22:01,387][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:22:01,888][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:22:02,387][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:22:02,887][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:22:03,393][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:22:03,894][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:22:04,394][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:22:04,895][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:22:05,397][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:22:05,898][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:22:06,398][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:22:06,899][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:22:07,399][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:22:07,905][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:22:08,406][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:22:08,910][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:22:09,418][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:22:09,928][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:22:10,433][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:22:10,936][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:22:11,440][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:22:11,944][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:22:12,447][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:22:12,950][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:22:13,480][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:22:13,983][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:22:14,485][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:22:14,984][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:22:15,483][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:22:15,982][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:22:16,480][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:22:16,979][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:22:17,476][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:22:17,978][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:22:18,481][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:22:18,979][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:22:19,479][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:22:19,985][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:22:20,484][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:22:20,983][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:22:21,483][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:22:21,984][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:22:22,487][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10085 tokens. +[2026-03-26 04:22:23,252][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.24%, Current % of VRAM taken: 60.66%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 04:22:24,066][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:22:24,068][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:22:24,070][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:22:24,999][__main__][INFO] - Iteration 428 took 52s (31.51% Gen, 66.73% Train). Generation: 16s, Training: 35s. Estimated remaining time: 37h 38m 1s. Estimated total time: 43h 59m 37s. Time estimates for 10 more iterations: 8m 47s, 100 more iterations: 1h 27m 59s, 500 more iterations: 7h 19m 56s. +[2026-03-26 04:22:25,001][__main__][INFO] - Starting iteration 428. +[2026-03-26 04:22:25,405][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 42 and human policies 1. +[2026-03-26 04:22:25,406][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:22:41,655][__main__][INFO] - Number of regex retries in iteration 428: 0 +[2026-03-26 04:22:41,656][__main__][INFO] - agents played in iteration 428 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:22:42,416][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:22:42,440][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:22:42,463][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:22:42,486][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:22:42,487][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:22:42,488][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:22:43,227][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:22:43,679][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:22:44,181][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:22:44,679][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:22:45,177][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:22:45,680][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:22:46,178][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:22:46,676][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:22:47,175][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:22:47,673][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:22:48,170][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:22:48,667][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:22:49,164][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:22:49,662][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:22:50,160][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:22:50,657][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:22:51,155][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:22:51,654][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:22:52,155][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:22:52,653][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:22:53,152][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:22:53,651][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:22:54,150][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:22:54,649][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:22:55,151][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:22:55,648][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:22:56,146][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:22:56,650][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:22:57,148][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:22:57,645][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:22:58,144][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:22:58,642][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:22:59,140][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:22:59,664][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:23:00,165][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:23:00,671][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:23:01,177][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:23:01,687][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:23:02,198][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:23:02,702][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:23:03,204][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:23:03,709][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:23:04,214][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:23:04,715][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:23:05,217][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:23:05,719][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:23:06,219][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:23:06,720][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:23:07,225][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:23:07,723][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:23:08,226][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:23:08,725][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:23:09,224][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:23:09,724][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:23:10,223][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:23:10,725][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:23:11,224][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:23:11,724][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:23:12,228][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:23:12,728][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:23:13,229][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:23:13,743][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:23:14,246][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:23:14,748][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:23:15,249][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10080 tokens. +[2026-03-26 04:23:16,025][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.60%, Block Peak % of device VRAM: 62.22%, ΔTime: 00:00:32 +[2026-03-26 04:23:16,766][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:23:16,769][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:23:16,771][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:23:17,494][__main__][INFO] - Iteration 429 took 52s (31.19% Gen, 67.41% Train). Generation: 16s, Training: 35s. Estimated remaining time: 37h 2m 0s. Estimated total time: 43h 24m 29s. Time estimates for 10 more iterations: 8m 40s, 100 more iterations: 1h 26m 48s, 500 more iterations: 7h 14m 4s. +[2026-03-26 04:23:17,496][__main__][INFO] - Starting iteration 429. +[2026-03-26 04:23:17,896][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 42 and human policies 1. +[2026-03-26 04:23:17,896][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:23:22,519][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:23:28,358][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls + +Given the distribution of per-item values, I recognize that Bob values books and balls much more highly than I do, while I value hats more highly than Bob does. Since hats are the only item that I value significantly more than Bob, I would focus on asserting control over this item by proposing to take all 10 hats. This maximizes my potential points from hats while minimizing the risk of over-allocation since the other items are not of much value to me. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:23:34,905][__main__][INFO] - Number of regex retries in iteration 429: 2 +[2026-03-26 04:23:34,906][__main__][INFO] - agents played in iteration 429 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:23:35,665][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:23:35,689][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:23:35,712][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:23:35,736][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:23:35,736][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:23:35,737][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:23:36,468][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:23:36,925][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:23:37,438][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:23:37,941][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:23:38,441][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:23:38,941][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:23:39,438][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:23:39,935][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:23:40,433][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:23:40,934][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:23:41,440][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:23:41,941][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:23:42,441][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:23:42,964][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:23:43,464][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:23:43,963][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:23:44,464][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:23:44,962][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:23:45,461][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:23:45,961][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:23:46,461][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:23:46,961][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:23:47,460][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:23:47,961][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:23:48,461][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:23:48,960][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:23:49,460][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:23:49,959][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:23:50,461][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:23:50,960][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:23:51,457][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:23:51,957][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:23:52,459][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:23:52,965][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:23:53,473][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:23:53,984][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:23:54,489][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:23:54,991][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:23:55,493][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:23:55,995][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:23:56,497][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:23:57,002][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:23:57,507][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:23:58,011][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:23:58,515][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:23:59,018][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:23:59,520][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:24:00,022][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:24:00,526][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:24:01,024][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:24:01,524][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:24:02,023][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:24:02,523][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:24:03,022][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:24:03,522][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:24:04,021][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:24:04,521][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:24:05,021][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:24:05,522][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:24:06,023][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:24:06,523][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:24:07,022][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:24:07,523][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:24:08,022][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:24:08,523][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9997 tokens. +[2026-03-26 04:24:09,287][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.03%, Current % of VRAM taken: 60.46%, Block Peak % of device VRAM: 62.25%, ΔTime: 00:00:32 +[2026-03-26 04:24:10,038][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:24:10,040][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:24:10,042][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:24:10,894][__main__][INFO] - Iteration 430 took 52s (32.09% Gen, 66.30% Train). Generation: 17s, Training: 35s. Estimated remaining time: 37h 46m 34s. Estimated total time: 44h 9m 56s. Time estimates for 10 more iterations: 8m 49s, 100 more iterations: 1h 28m 19s, 500 more iterations: 7h 21m 39s. +[2026-03-26 04:24:10,896][__main__][INFO] - Starting iteration 430. +[2026-03-26 04:24:11,311][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 42 and human policies 1. +[2026-03-26 04:24:11,311][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:24:26,805][__main__][INFO] - Number of regex retries in iteration 430: 0 +[2026-03-26 04:24:26,806][__main__][INFO] - agents played in iteration 430 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:24:27,566][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:24:27,590][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:24:27,613][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:24:27,637][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:24:27,637][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:24:27,638][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:24:28,373][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:24:28,826][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:24:29,332][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:24:29,832][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:24:30,334][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:24:30,833][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:24:31,332][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:24:31,831][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:24:32,331][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:24:32,829][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:24:33,329][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:24:33,827][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:24:34,326][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:24:34,825][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:24:35,324][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:24:35,826][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:24:36,325][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:24:36,823][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:24:37,321][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:24:37,819][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:24:38,318][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:24:38,818][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:24:39,316][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:24:39,816][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:24:40,316][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:24:40,812][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:24:41,314][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:24:41,813][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:24:42,312][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:24:42,811][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:24:43,310][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:24:43,813][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:24:44,320][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:24:44,829][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:24:45,337][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:24:45,847][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:24:46,348][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:24:46,848][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:24:47,350][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:24:47,853][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:24:48,353][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:24:48,855][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:24:49,358][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:24:49,862][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:24:50,360][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:24:50,860][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:24:51,359][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:24:51,857][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:24:52,359][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:24:52,858][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:24:53,356][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:24:53,855][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:24:54,353][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:24:54,852][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:24:55,350][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:24:55,848][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:24:56,347][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:24:56,847][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:24:57,347][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:24:57,846][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:24:58,348][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:24:58,848][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:24:59,349][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:24:59,850][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:25:00,351][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10003 tokens. +[2026-03-26 04:25:01,127][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.31%, ΔTime: 00:00:32 +[2026-03-26 04:25:01,896][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:25:01,898][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:25:01,900][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:25:03,226][__main__][INFO] - Iteration 431 took 51s (29.84% Gen, 67.60% Train). Generation: 15s, Training: 35s. Estimated remaining time: 36h 51m 35s. Estimated total time: 43h 15m 49s. Time estimates for 10 more iterations: 8m 39s, 100 more iterations: 1h 26m 31s, 500 more iterations: 7h 12m 38s. +[2026-03-26 04:25:03,229][__main__][INFO] - Starting iteration 431. +[2026-03-26 04:25:03,639][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 43 and human policies 1. +[2026-03-26 04:25:03,640][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:25:19,030][__main__][INFO] - Number of regex retries in iteration 431: 0 +[2026-03-26 04:25:19,031][__main__][INFO] - agents played in iteration 431 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:25:19,799][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:25:19,822][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:25:19,846][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:25:19,869][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:25:19,870][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:25:19,870][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:25:20,603][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:25:21,056][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:25:21,563][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:25:22,061][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:25:22,559][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:25:23,058][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:25:23,557][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:25:24,057][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:25:24,556][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:25:25,061][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:25:25,563][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:25:26,065][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:25:26,566][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:25:27,067][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:25:27,568][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:25:28,072][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:25:28,574][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:25:29,072][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:25:29,572][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:25:30,070][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:25:30,572][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:25:31,074][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:25:31,574][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:25:32,073][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:25:32,569][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:25:33,067][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:25:33,566][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:25:34,072][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:25:34,575][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:25:35,084][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:25:35,589][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:25:36,093][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:25:36,598][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:25:37,100][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:25:37,599][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:25:38,100][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:25:38,603][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:25:39,102][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:25:39,602][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:25:40,103][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:25:40,601][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:25:41,101][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:25:41,604][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:25:42,102][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:25:42,601][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:25:43,101][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:25:43,600][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:25:44,099][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:25:44,599][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:25:45,097][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:25:45,595][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:25:46,092][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:25:46,590][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:25:47,088][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:25:47,587][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:25:48,085][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:25:48,583][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:25:49,083][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:25:49,582][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:25:50,084][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:25:50,583][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:25:51,082][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:25:51,586][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:25:52,087][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:25:52,585][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10032 tokens. +[2026-03-26 04:25:53,350][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.02%, Current % of VRAM taken: 60.45%, Block Peak % of device VRAM: 62.29%, ΔTime: 00:00:32 +[2026-03-26 04:25:54,195][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:25:54,198][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:25:54,200][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:25:55,091][__main__][INFO] - Iteration 432 took 51s (29.91% Gen, 68.35% Train). Generation: 15s, Training: 35s. Estimated remaining time: 36h 27m 31s. Estimated total time: 42h 52m 38s. Time estimates for 10 more iterations: 8m 34s, 100 more iterations: 1h 25m 45s, 500 more iterations: 7h 8m 46s. +[2026-03-26 04:25:55,093][__main__][INFO] - Starting iteration 432. +[2026-03-26 04:25:55,494][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 43 and human policies 1. +[2026-03-26 04:25:55,495][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:26:00,456][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:26:04,054][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls + +Given the values, hats are highly valuable to you and only slightly valuable to Alice. Books and balls are more valuable to Alice. By proposing to take all the hats, I hope to maximize my points in this round since the hats are worth a lot to me and Alice values them very little. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:26:12,964][__main__][INFO] - Number of regex retries in iteration 432: 2 +[2026-03-26 04:26:12,965][__main__][INFO] - agents played in iteration 432 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:26:13,737][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:26:13,761][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:26:13,784][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:26:13,807][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:26:13,808][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:26:13,808][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:26:14,537][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:26:14,989][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:26:15,495][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:26:15,996][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:26:16,493][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:26:16,994][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:26:17,498][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:26:17,997][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:26:18,493][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:26:18,992][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:26:19,491][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:26:19,992][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:26:20,495][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:26:20,993][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:26:21,498][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:26:21,995][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:26:22,494][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:26:22,991][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:26:23,491][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:26:23,993][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:26:24,501][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:26:25,002][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:26:25,505][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:26:26,004][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:26:26,506][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:26:27,006][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:26:27,507][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:26:28,011][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:26:28,510][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:26:29,009][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:26:29,508][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:26:30,007][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:26:30,510][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:26:31,012][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:26:31,531][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:26:32,031][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:26:32,536][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:26:33,036][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:26:33,540][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:26:34,041][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:26:34,539][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:26:35,038][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:26:35,537][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:26:36,037][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:26:36,536][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:26:37,035][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:26:37,535][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:26:38,037][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:26:38,539][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:26:39,039][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:26:39,539][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:26:40,039][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:26:40,540][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:26:41,040][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:26:41,543][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:26:42,042][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:26:42,542][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:26:43,039][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:26:43,537][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:26:44,036][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:26:44,536][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:26:45,037][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:26:45,536][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:26:46,035][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:26:46,532][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10049 tokens. +[2026-03-26 04:26:47,289][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.20%, ΔTime: 00:00:32 +[2026-03-26 04:26:48,030][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:26:48,032][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:26:48,034][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:26:48,979][__main__][INFO] - Iteration 433 took 53s (32.66% Gen, 65.57% Train). Generation: 17s, Training: 35s. Estimated remaining time: 38h 8m 18s. Estimated total time: 44h 34m 18s. Time estimates for 10 more iterations: 8m 54s, 100 more iterations: 1h 29m 8s, 500 more iterations: 7h 25m 43s. +[2026-03-26 04:26:48,982][__main__][INFO] - Starting iteration 433. +[2026-03-26 04:26:49,393][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 43 and human policies 1. +[2026-03-26 04:26:49,394][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:26:58,802][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:27:05,016][__main__][INFO] - Number of regex retries in iteration 433: 1 +[2026-03-26 04:27:05,017][__main__][INFO] - agents played in iteration 433 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:27:05,782][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:27:05,806][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:27:05,829][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:27:05,853][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:27:05,853][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:27:05,854][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:27:06,584][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:27:07,040][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:27:07,544][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:27:08,043][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:27:08,542][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:27:09,045][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:27:09,543][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:27:10,042][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:27:10,541][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:27:11,040][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:27:11,538][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:27:12,036][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:27:12,535][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:27:13,033][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:27:13,531][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:27:14,029][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:27:14,526][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:27:15,026][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:27:15,527][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:27:16,027][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:27:16,530][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:27:17,040][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:27:17,547][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:27:18,053][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:27:18,559][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:27:19,060][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:27:19,563][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:27:20,065][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:27:20,568][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:27:21,071][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:27:21,572][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:27:22,074][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:27:22,578][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:27:23,079][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:27:23,579][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:27:24,079][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:27:24,579][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:27:25,080][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:27:25,580][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:27:26,081][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:27:26,582][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:27:27,082][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:27:27,580][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:27:28,080][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:27:28,579][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:27:29,078][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:27:29,578][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:27:30,077][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:27:30,579][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:27:31,080][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:27:31,580][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:27:32,082][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:27:32,583][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:27:33,086][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:27:33,591][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:27:34,091][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:27:34,594][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:27:35,100][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:27:35,603][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:27:36,106][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:27:36,607][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:27:37,109][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:27:37,610][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:27:38,118][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:27:38,621][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10028 tokens. +[2026-03-26 04:27:39,390][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.61%, Block Peak % of device VRAM: 62.30%, ΔTime: 00:00:32 +[2026-03-26 04:27:40,104][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:27:40,106][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:27:40,108][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:27:40,907][__main__][INFO] - Iteration 434 took 51s (30.33% Gen, 68.12% Train). Generation: 15s, Training: 35s. Estimated remaining time: 36h 28m 51s. Estimated total time: 42h 55m 43s. Time estimates for 10 more iterations: 8m 35s, 100 more iterations: 1h 25m 51s, 500 more iterations: 7h 9m 17s. +[2026-03-26 04:27:40,909][__main__][INFO] - Starting iteration 434. +[2026-03-26 04:27:41,327][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 43 and human policies 1. +[2026-03-26 04:27:41,328][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:27:56,868][__main__][INFO] - Number of regex retries in iteration 434: 0 +[2026-03-26 04:27:56,868][__main__][INFO] - agents played in iteration 434 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:27:57,628][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:27:57,652][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:27:57,675][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:27:57,699][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:27:57,699][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:27:57,700][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:27:58,473][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:27:58,931][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:27:59,439][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:27:59,939][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:28:00,440][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:28:00,939][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:28:01,450][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:28:01,950][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:28:02,452][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:28:02,951][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:28:03,450][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:28:03,961][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:28:04,460][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:28:04,960][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:28:05,459][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:28:05,957][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:28:06,460][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:28:06,972][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:28:07,487][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:28:07,993][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:28:08,501][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:28:09,006][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:28:09,510][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:28:10,012][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:28:10,524][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:28:11,052][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:28:11,560][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:28:12,067][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:28:12,569][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:28:13,074][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:28:13,578][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:28:14,079][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:28:14,579][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:28:15,080][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:28:15,580][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:28:16,080][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:28:16,581][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:28:17,082][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:28:17,582][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:28:18,085][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:28:18,597][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:28:19,097][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:28:19,597][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:28:20,097][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:28:20,597][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:28:21,098][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:28:21,597][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:28:22,098][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:28:22,597][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:28:23,098][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:28:23,599][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:28:24,113][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:28:24,616][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:28:25,121][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:28:25,624][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:28:26,124][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:28:26,624][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:28:27,128][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:28:27,631][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:28:28,134][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:28:28,636][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:28:29,139][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:28:29,640][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:28:30,151][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:28:30,654][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10011 tokens. +[2026-03-26 04:28:31,442][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.17%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.17%, ΔTime: 00:00:32 +[2026-03-26 04:28:32,270][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:28:32,273][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:28:32,275][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:28:33,046][__main__][INFO] - Iteration 435 took 51s (30.05% Gen, 68.46% Train). Generation: 15s, Training: 35s. Estimated remaining time: 36h 38m 12s. Estimated total time: 43h 5m 57s. Time estimates for 10 more iterations: 8m 37s, 100 more iterations: 1h 26m 11s, 500 more iterations: 7h 10m 59s. +[2026-03-26 04:28:33,048][__main__][INFO] - Starting iteration 435. +[2026-03-26 04:28:33,446][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 43 and human policies 1. +[2026-03-26 04:28:33,447][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:28:50,951][__main__][INFO] - Number of regex retries in iteration 435: 0 +[2026-03-26 04:28:50,952][__main__][INFO] - agents played in iteration 435 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:28:51,718][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:28:51,744][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:28:51,769][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:28:51,794][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:28:51,794][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:28:51,795][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:28:52,546][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:28:53,005][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:28:53,510][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:28:54,025][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:28:54,527][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:28:55,028][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:28:55,531][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:28:56,036][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:28:56,546][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:28:57,052][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:28:57,558][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:28:58,074][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:28:58,586][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:28:59,096][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:28:59,600][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:29:00,106][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:29:00,611][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:29:01,121][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:29:01,624][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:29:02,128][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:29:02,631][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:29:03,138][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:29:03,639][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:29:04,142][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:29:04,650][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:29:05,153][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:29:05,651][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:29:06,151][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:29:06,650][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:29:07,149][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:29:07,647][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:29:08,146][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:29:08,645][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:29:09,156][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:29:09,658][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:29:10,159][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:29:10,660][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:29:11,162][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:29:11,663][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:29:12,167][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:29:12,671][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:29:13,174][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:29:13,676][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:29:14,180][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:29:14,684][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:29:15,186][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:29:15,688][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:29:16,190][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:29:16,691][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:29:17,190][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:29:17,690][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:29:18,192][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:29:18,695][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:29:19,199][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:29:19,698][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:29:20,199][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:29:20,698][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:29:21,202][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:29:21,704][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:29:22,204][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:29:22,707][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:29:23,210][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:29:23,709][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:29:24,211][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:29:24,713][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10174 tokens. +[2026-03-26 04:29:25,492][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.23%, Current % of VRAM taken: 60.65%, Block Peak % of device VRAM: 62.23%, ΔTime: 00:00:32 +[2026-03-26 04:29:26,244][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:29:26,246][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:29:26,248][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:29:26,979][__main__][INFO] - Iteration 436 took 53s (32.70% Gen, 65.93% Train). Generation: 17s, Training: 35s. Estimated remaining time: 38h 8m 1s. Estimated total time: 44h 36m 40s. Time estimates for 10 more iterations: 8m 55s, 100 more iterations: 1h 29m 13s, 500 more iterations: 7h 26m 6s. +[2026-03-26 04:29:26,981][__main__][INFO] - Starting iteration 436. +[2026-03-26 04:29:27,385][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 43 and human policies 1. +[2026-03-26 04:29:27,385][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:29:44,381][__main__][INFO] - Number of regex retries in iteration 436: 0 +[2026-03-26 04:29:44,382][__main__][INFO] - agents played in iteration 436 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:29:45,147][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:29:45,171][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:29:45,194][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:29:45,217][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:29:45,218][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:29:45,219][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:29:45,932][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:29:46,410][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:29:46,920][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:29:47,421][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:29:47,923][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:29:48,436][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:29:48,941][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:29:49,442][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:29:49,947][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:29:50,447][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:29:50,959][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:29:51,462][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:29:51,971][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:29:52,477][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:29:52,988][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:29:53,490][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:29:53,992][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:29:54,499][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:29:55,000][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:29:55,499][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:29:56,000][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:29:56,500][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:29:57,001][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:29:57,504][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:29:58,014][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:29:58,515][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:29:59,018][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:29:59,521][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:30:00,021][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:30:00,523][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:30:01,024][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:30:01,527][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:30:02,025][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:30:02,535][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:30:03,036][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:30:03,536][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:30:04,036][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:30:04,536][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:30:05,036][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:30:05,535][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:30:06,034][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:30:06,534][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:30:07,033][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:30:07,533][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:30:08,034][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:30:08,532][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:30:09,029][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:30:09,528][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:30:10,026][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:30:10,525][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:30:11,028][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:30:11,528][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:30:12,031][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:30:12,532][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:30:13,034][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:30:13,532][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:30:14,031][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:30:14,534][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:30:15,037][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:30:15,539][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:30:16,041][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:30:16,543][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:30:17,045][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:30:17,547][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:30:18,048][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9950 tokens. +[2026-03-26 04:30:18,842][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.01%, ΔTime: 00:00:32 +[2026-03-26 04:30:19,591][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:30:19,594][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:30:19,596][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:30:20,351][__main__][INFO] - Iteration 437 took 52s (32.09% Gen, 66.48% Train). Generation: 16s, Training: 35s. Estimated remaining time: 37h 38m 59s. Estimated total time: 44h 8m 30s. Time estimates for 10 more iterations: 8m 49s, 100 more iterations: 1h 28m 17s, 500 more iterations: 7h 21m 25s. +[2026-03-26 04:30:20,353][__main__][INFO] - Starting iteration 437. +[2026-03-26 04:30:20,753][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 43 and human policies 1. +[2026-03-26 04:30:20,754][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:30:42,697][__main__][INFO] - Number of regex retries in iteration 437: 0 +[2026-03-26 04:30:42,698][__main__][INFO] - agents played in iteration 437 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:30:43,484][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:30:43,508][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:30:43,532][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:30:43,557][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:30:43,558][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:30:43,558][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:30:44,371][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:30:44,832][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:30:45,336][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:30:45,838][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:30:46,337][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:30:46,836][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:30:47,340][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:30:47,841][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:30:48,339][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:30:48,839][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:30:49,340][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:30:49,842][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:30:50,343][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:30:50,852][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:30:51,355][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:30:51,859][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:30:52,359][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:30:52,874][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:30:53,379][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:30:53,880][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:30:54,384][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:30:54,888][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:30:55,390][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:30:55,895][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:30:56,395][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:30:56,896][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:30:57,396][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:30:57,897][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:30:58,396][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:30:58,898][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:30:59,398][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:30:59,899][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:31:00,399][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:31:00,899][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:31:01,400][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:31:01,900][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:31:02,399][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:31:02,898][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:31:03,398][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:31:03,900][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:31:04,404][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:31:04,909][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:31:05,417][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:31:05,918][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:31:06,422][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:31:06,927][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:31:07,430][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:31:07,932][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:31:08,432][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:31:08,933][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:31:09,437][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:31:09,936][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:31:10,434][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:31:10,932][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:31:11,431][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:31:11,928][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:31:12,427][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:31:12,928][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:31:13,427][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:31:13,927][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:31:14,426][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:31:14,924][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:31:15,423][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:31:15,922][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:31:16,424][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10026 tokens. +[2026-03-26 04:31:17,242][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.98%, Current % of VRAM taken: 60.41%, Block Peak % of device VRAM: 62.23%, ΔTime: 00:00:32 +[2026-03-26 04:31:18,002][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:31:18,004][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:31:18,006][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:31:18,739][__main__][INFO] - Iteration 438 took 57s (37.84% Gen, 60.89% Train). Generation: 21s, Training: 35s. Estimated remaining time: 41h 48m 48s. Estimated total time: 48h 19m 18s. Time estimates for 10 more iterations: 9m 39s, 100 more iterations: 1h 36m 38s, 500 more iterations: 8h 3m 13s. +[2026-03-26 04:31:18,741][__main__][INFO] - Starting iteration 438. +[2026-03-26 04:31:19,142][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 43 and human policies 1. +[2026-03-26 04:31:19,142][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:31:40,032][__main__][INFO] - Number of regex retries in iteration 438: 0 +[2026-03-26 04:31:40,033][__main__][INFO] - agents played in iteration 438 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:31:40,817][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:31:40,841][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:31:40,865][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:31:40,888][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:31:40,889][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:31:40,889][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:31:41,655][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:31:42,109][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:31:42,617][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:31:43,116][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:31:43,627][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:31:44,133][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:31:44,639][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:31:45,140][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:31:45,641][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:31:46,142][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:31:46,643][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:31:47,150][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:31:47,650][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:31:48,156][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:31:48,657][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:31:49,157][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:31:49,658][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:31:50,159][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:31:50,659][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:31:51,159][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:31:51,659][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:31:52,159][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:31:52,659][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:31:53,159][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:31:53,661][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:31:54,158][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:31:54,659][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:31:55,162][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:31:55,662][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:31:56,162][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:31:56,661][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:31:57,161][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:31:57,661][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:31:58,160][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:31:58,660][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:31:59,158][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:31:59,658][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:32:00,156][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:32:00,654][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:32:01,155][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:32:01,653][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:32:02,151][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:32:02,648][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:32:03,146][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:32:03,645][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:32:04,142][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:32:04,639][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:32:05,136][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:32:05,646][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:32:06,147][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:32:06,646][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:32:07,148][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:32:07,650][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:32:08,154][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:32:08,663][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:32:09,165][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:32:09,665][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:32:10,165][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:32:10,669][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:32:11,168][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:32:11,668][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:32:12,168][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:32:12,667][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:32:13,168][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:32:13,668][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9949 tokens. +[2026-03-26 04:32:14,475][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.09%, ΔTime: 00:00:32 +[2026-03-26 04:32:15,247][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:32:15,249][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:32:15,251][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:32:16,301][__main__][INFO] - Iteration 439 took 57s (36.55% Gen, 61.61% Train). Generation: 20s, Training: 35s. Estimated remaining time: 41h 6m 33s. Estimated total time: 47h 38m 1s. Time estimates for 10 more iterations: 9m 31s, 100 more iterations: 1h 35m 16s, 500 more iterations: 7h 56m 20s. +[2026-03-26 04:32:16,304][__main__][INFO] - Starting iteration 439. +[2026-03-26 04:32:16,713][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 43 and human policies 1. +[2026-03-26 04:32:16,714][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:32:37,626][__main__][INFO] - Number of regex retries in iteration 439: 0 +[2026-03-26 04:32:37,627][__main__][INFO] - agents played in iteration 439 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:32:38,420][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:32:38,444][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:32:38,468][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:32:38,492][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:32:38,492][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:32:38,493][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:32:39,254][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:32:39,710][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:32:40,213][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:32:40,717][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:32:41,215][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:32:41,713][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:32:42,212][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:32:42,716][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:32:43,214][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:32:43,712][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:32:44,211][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:32:44,711][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:32:45,211][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:32:45,713][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:32:46,211][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:32:46,720][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:32:47,221][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:32:47,723][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:32:48,226][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:32:48,725][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:32:49,224][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:32:49,723][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:32:50,222][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:32:50,719][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:32:51,217][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:32:51,716][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:32:52,215][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:32:52,714][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:32:53,215][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:32:53,719][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:32:54,220][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:32:54,729][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:32:55,231][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:32:55,739][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:32:56,237][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:32:56,762][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:32:57,264][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:32:57,764][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:32:58,272][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:32:58,776][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:32:59,284][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:32:59,783][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:33:00,282][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:33:00,781][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:33:01,282][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:33:01,786][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:33:02,286][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:33:02,787][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:33:03,289][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:33:03,794][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:33:04,295][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:33:04,797][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:33:05,298][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:33:05,799][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:33:06,299][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:33:06,799][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:33:07,300][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:33:07,805][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:33:08,309][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:33:08,807][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:33:10,088][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:33:10,589][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:33:11,100][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:33:11,601][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:33:12,101][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9947 tokens. +[2026-03-26 04:33:13,028][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.13%, ΔTime: 00:00:33 +[2026-03-26 04:33:13,805][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:33:13,808][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:33:13,809][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:33:14,777][__main__][INFO] - Iteration 440 took 58s (36.02% Gen, 62.31% Train). Generation: 20s, Training: 36s. Estimated remaining time: 41h 50m 51s. Estimated total time: 48h 23m 17s. Time estimates for 10 more iterations: 9m 40s, 100 more iterations: 1h 36m 46s, 500 more iterations: 8h 3m 52s. +[2026-03-26 04:33:14,782][__main__][INFO] - Starting iteration 440. +[2026-03-26 04:33:15,184][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 43 and human policies 1. +[2026-03-26 04:33:15,185][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:34:12,998][__main__][INFO] - Number of regex retries in iteration 440: 0 +[2026-03-26 04:34:12,999][__main__][INFO] - agents played in iteration 440 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:34:13,786][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:34:13,810][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:34:13,834][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:34:13,859][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:34:13,860][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:34:13,860][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:34:14,626][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:34:15,080][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:34:15,589][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:34:16,097][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:34:16,604][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:34:17,106][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:34:17,608][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:34:18,112][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:34:18,613][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:34:19,124][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:34:19,626][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:34:20,127][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:34:20,626][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:34:21,126][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:34:21,626][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:34:22,127][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:34:22,625][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:34:23,128][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:34:23,634][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:34:24,149][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:34:24,653][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:34:25,157][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:34:25,661][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:34:26,162][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:34:26,665][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:34:27,165][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:34:27,666][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:34:28,173][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:34:28,676][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:34:29,179][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:34:29,686][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:34:30,187][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:34:30,690][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:34:31,188][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:34:31,688][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:34:32,185][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:34:32,685][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:34:33,193][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:34:33,697][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:34:34,199][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:34:34,700][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:34:35,201][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:34:35,704][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:34:36,205][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:34:36,705][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:34:37,210][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:34:37,710][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:34:38,223][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:34:38,725][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:34:39,225][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:34:39,725][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:34:40,229][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:34:40,730][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:34:41,230][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:34:41,731][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:34:42,234][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:34:42,733][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:34:43,235][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:34:43,735][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:34:44,236][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:34:44,736][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:34:45,235][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:34:45,735][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:34:46,236][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:34:46,737][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10012 tokens. +[2026-03-26 04:34:47,544][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.32%, ΔTime: 00:00:32 +[2026-03-26 04:34:48,371][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:34:48,376][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:34:48,378][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:34:49,736][__main__][INFO] - Iteration 441 took 1m 34s (61.14% Gen, 37.42% Train). Generation: 57s, Training: 35s. Estimated remaining time: 72h 13m 37s. Estimated total time: 78h 47m 38s. Time estimates for 10 more iterations: 15m 45s, 100 more iterations: 2h 37m 35s, 500 more iterations: 13h 7m 56s. +[2026-03-26 04:34:49,739][__main__][INFO] - Starting iteration 441. +[2026-03-26 04:34:50,166][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 44 and human policies 1. +[2026-03-26 04:34:50,167][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:37:40,047][__main__][INFO] - Number of regex retries in iteration 441: 0 +[2026-03-26 04:37:40,048][__main__][INFO] - agents played in iteration 441 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:37:40,818][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:37:40,842][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:37:40,865][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:37:40,888][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:37:40,889][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:37:40,889][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:37:41,658][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:37:42,114][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:37:42,620][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:37:43,120][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:37:43,621][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:37:44,118][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:37:44,619][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:37:45,136][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:37:45,635][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:37:46,142][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:37:46,640][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:37:47,141][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:37:47,643][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:37:48,143][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:37:48,646][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:37:49,145][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:37:49,646][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:37:50,146][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:37:50,646][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:37:51,145][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:37:51,644][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:37:52,144][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:37:52,643][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:37:53,143][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:37:53,642][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:37:54,142][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:37:54,642][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:37:55,147][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:37:55,662][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:37:56,163][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:37:56,668][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:37:57,171][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:37:57,671][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:37:58,171][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:37:58,668][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:37:59,170][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:37:59,689][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:38:00,197][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:38:00,697][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:38:01,200][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:38:01,700][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:38:02,201][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:38:02,704][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:38:03,203][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:38:03,703][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:38:04,209][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:38:04,712][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:38:05,212][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:38:05,713][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:38:06,211][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:38:06,712][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:38:07,211][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:38:07,710][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:38:08,211][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:38:08,714][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:38:09,213][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:38:09,712][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:38:10,216][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:38:10,716][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:38:11,216][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:38:11,717][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:38:12,217][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:38:12,714][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:38:13,212][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:38:13,711][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9976 tokens. +[2026-03-26 04:38:14,506][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.32%, ΔTime: 00:00:32 +[2026-03-26 04:38:15,582][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:38:15,584][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:38:15,586][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:38:17,104][__main__][INFO] - Iteration 442 took 3m 26s (82.09% Gen, 17.17% Train). Generation: 2m 49s, Training: 35s. Estimated remaining time: 165h 49m 27s. Estimated total time: 172h 26m 56s. Time estimates for 10 more iterations: 34m 29s, 100 more iterations: 5h 44m 53s, 500 more iterations: 28h 44m 29s. +[2026-03-26 04:38:17,106][__main__][INFO] - Starting iteration 442. +[2026-03-26 04:38:17,562][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 44 and human policies 1. +[2026-03-26 04:38:17,563][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:38:51,670][__main__][INFO] - Number of regex retries in iteration 442: 0 +[2026-03-26 04:38:51,671][__main__][INFO] - agents played in iteration 442 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:38:52,445][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:38:52,468][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:38:52,492][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:38:52,515][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:38:52,516][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:38:52,516][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:38:53,276][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:38:53,729][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:38:54,230][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:38:54,727][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:38:55,225][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:38:55,726][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:38:56,232][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:38:56,740][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:38:57,239][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:38:57,739][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:38:58,238][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:38:58,741][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:38:59,239][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:38:59,738][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:39:00,236][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:39:00,734][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:39:01,230][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:39:01,739][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:39:02,239][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:39:02,740][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:39:03,241][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:39:03,740][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:39:04,241][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:39:04,740][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:39:05,241][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:39:05,741][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:39:06,243][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:39:06,745][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:39:07,246][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:39:07,765][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:39:08,268][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:39:08,767][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:39:09,267][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:39:09,767][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:39:10,266][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:39:10,768][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:39:11,268][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:39:11,768][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:39:12,267][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:39:12,770][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:39:13,275][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:39:13,782][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:39:14,287][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:39:14,790][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:39:15,293][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:39:15,793][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:39:16,289][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:39:16,785][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:39:17,282][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:39:17,779][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:39:18,276][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:39:18,774][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:39:19,270][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:39:19,775][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:39:20,273][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:39:20,771][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:39:21,267][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:39:21,765][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:39:22,264][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:39:22,764][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:39:23,263][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:39:23,762][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:39:24,262][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:39:24,761][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:39:25,262][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9949 tokens. +[2026-03-26 04:39:26,040][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 61.94%, ΔTime: 00:00:32 +[2026-03-26 04:39:26,888][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:39:26,890][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:39:26,892][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:39:27,800][__main__][INFO] - Iteration 443 took 1m 10s (48.56% Gen, 50.15% Train). Generation: 34s, Training: 35s. Estimated remaining time: 51h 53m 14s. Estimated total time: 58h 31m 53s. Time estimates for 10 more iterations: 11m 42s, 100 more iterations: 1h 57m 3s, 500 more iterations: 9h 45m 18s. +[2026-03-26 04:39:27,802][__main__][INFO] - Starting iteration 443. +[2026-03-26 04:39:28,202][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 44 and human policies 1. +[2026-03-26 04:39:28,203][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:39:45,657][__main__][INFO] - Number of regex retries in iteration 443: 0 +[2026-03-26 04:39:45,657][__main__][INFO] - agents played in iteration 443 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:39:46,435][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:39:46,459][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:39:46,482][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:39:46,506][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:39:46,506][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:39:46,507][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:39:47,260][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:39:47,714][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:39:48,219][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:39:48,718][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:39:49,217][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:39:49,715][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:39:50,214][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:39:50,711][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:39:51,208][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:39:51,707][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:39:52,207][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:39:52,719][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:39:53,221][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:39:53,723][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:39:54,250][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:39:54,753][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:39:55,258][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:39:55,760][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:39:56,257][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:39:56,765][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:39:57,264][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:39:57,764][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:39:58,264][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:39:58,760][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:39:59,256][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:39:59,759][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:40:00,263][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:40:00,764][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:40:01,266][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:40:01,768][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:40:02,270][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:40:02,770][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:40:03,272][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:40:03,776][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:40:04,275][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:40:04,778][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:40:05,276][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:40:05,777][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:40:06,279][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:40:06,777][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:40:07,281][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:40:07,793][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:40:08,291][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:40:08,789][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:40:09,288][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:40:09,786][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:40:10,283][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:40:10,781][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:40:11,279][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:40:11,777][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:40:12,277][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:40:12,778][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:40:13,279][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:40:13,779][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:40:14,281][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:40:14,779][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:40:15,277][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:40:15,778][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:40:16,276][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:40:16,779][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:40:17,282][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:40:17,782][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:40:18,281][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:40:18,782][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:40:19,280][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10023 tokens. +[2026-03-26 04:40:20,070][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.25%, ΔTime: 00:00:32 +[2026-03-26 04:40:20,817][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:40:20,819][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:40:20,820][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:40:21,606][__main__][INFO] - Iteration 444 took 53s (32.68% Gen, 65.84% Train). Generation: 17s, Training: 35s. Estimated remaining time: 37h 50m 41s. Estimated total time: 44h 30m 13s. Time estimates for 10 more iterations: 8m 54s, 100 more iterations: 1h 29m 0s, 500 more iterations: 7h 25m 2s. +[2026-03-26 04:40:21,608][__main__][INFO] - Starting iteration 444. +[2026-03-26 04:40:22,066][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 44 and human policies 1. +[2026-03-26 04:40:22,067][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:40:59,119][__main__][INFO] - Number of regex retries in iteration 444: 0 +[2026-03-26 04:40:59,120][__main__][INFO] - agents played in iteration 444 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:40:59,881][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:40:59,905][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:40:59,928][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:40:59,951][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:40:59,952][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:40:59,952][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:41:00,730][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:41:01,184][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:41:01,692][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:41:02,197][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:41:02,699][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:41:03,199][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:41:03,701][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:41:04,202][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:41:04,718][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:41:05,220][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:41:05,723][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:41:06,225][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:41:06,729][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:41:07,240][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:41:07,742][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:41:08,244][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:41:08,745][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:41:09,246][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:41:09,750][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:41:10,268][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:41:10,770][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:41:11,271][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:41:11,772][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:41:12,277][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:41:12,779][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:41:13,279][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:41:13,791][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:41:14,291][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:41:14,789][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:41:15,287][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:41:15,786][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:41:16,285][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:41:16,784][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:41:17,287][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:41:17,790][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:41:18,292][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:41:18,798][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:41:19,299][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:41:19,815][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:41:20,319][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:41:20,820][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:41:21,323][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:41:21,824][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:41:22,326][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:41:22,827][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:41:23,327][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:41:23,828][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:41:24,329][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:41:24,843][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:41:25,343][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:41:25,841][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:41:26,341][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:41:26,840][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:41:27,338][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:41:27,837][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:41:28,335][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:41:28,835][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:41:29,339][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:41:29,842][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:41:30,348][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:41:30,861][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:41:31,372][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:41:31,881][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:41:32,386][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:41:32,892][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10068 tokens. +[2026-03-26 04:41:33,716][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.20%, Current % of VRAM taken: 60.63%, Block Peak % of device VRAM: 62.19%, ΔTime: 00:00:32 +[2026-03-26 04:41:34,553][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:41:34,556][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:41:34,557][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:41:35,480][__main__][INFO] - Iteration 445 took 1m 13s (50.47% Gen, 48.27% Train). Generation: 37s, Training: 35s. Estimated remaining time: 54h 29m 56s. Estimated total time: 61h 10m 43s. Time estimates for 10 more iterations: 12m 14s, 100 more iterations: 2h 2m 21s, 500 more iterations: 10h 11m 47s. +[2026-03-26 04:41:35,482][__main__][INFO] - Starting iteration 445. +[2026-03-26 04:41:35,887][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 44 and human policies 1. +[2026-03-26 04:41:35,888][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:41:50,794][__main__][INFO] - Number of regex retries in iteration 445: 0 +[2026-03-26 04:41:50,795][__main__][INFO] - agents played in iteration 445 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:41:51,574][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:41:51,598][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:41:51,621][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:41:51,645][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:41:51,645][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:41:51,646][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:41:52,415][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:41:52,869][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:41:53,380][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:41:53,880][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:41:54,380][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:41:54,879][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:41:55,381][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:41:55,894][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:41:56,396][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:41:56,896][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:41:57,398][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:41:57,899][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:41:58,403][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:41:58,907][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:41:59,407][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:41:59,908][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:42:00,409][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:42:00,908][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:42:01,407][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:42:01,906][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:42:02,406][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:42:02,909][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:42:03,408][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:42:03,908][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:42:04,408][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:42:04,919][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:42:05,420][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:42:05,918][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:42:06,423][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:42:06,922][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:42:07,426][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:42:07,926][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:42:08,426][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:42:08,926][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:42:09,425][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:42:09,926][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:42:10,425][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:42:10,922][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:42:11,421][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:42:11,918][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:42:12,415][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:42:12,914][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:42:13,414][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:42:13,918][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:42:14,418][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:42:14,917][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:42:15,417][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:42:15,919][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:42:16,419][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:42:16,920][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:42:17,421][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:42:17,920][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:42:18,423][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:42:18,928][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:42:19,429][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:42:19,930][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:42:20,442][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:42:20,943][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:42:21,443][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:42:21,943][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:42:22,446][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:42:22,949][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:42:23,458][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:42:23,966][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:42:24,469][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9996 tokens. +[2026-03-26 04:42:25,286][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.21%, ΔTime: 00:00:32 +[2026-03-26 04:42:26,017][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:42:26,019][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:42:26,021][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:42:26,836][__main__][INFO] - Iteration 446 took 50s (29.26% Gen, 69.14% Train). Generation: 14s, Training: 35s. Estimated remaining time: 35h 45m 48s. Estimated total time: 42h 27m 26s. Time estimates for 10 more iterations: 8m 29s, 100 more iterations: 1h 24m 54s, 500 more iterations: 7h 4m 34s. +[2026-03-26 04:42:26,838][__main__][INFO] - Starting iteration 446. +[2026-03-26 04:42:27,249][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 44 and human policies 1. +[2026-03-26 04:42:27,249][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:44:15,369][mllm.models.large_language_model_local][WARNING] - Response Propagation: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:44:26,414][__main__][INFO] - Number of regex retries in iteration 446: 1 +[2026-03-26 04:44:26,415][__main__][INFO] - agents played in iteration 446 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:44:27,194][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:44:27,217][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:44:27,241][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:44:27,264][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:44:27,264][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:44:27,265][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:44:28,017][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:44:28,476][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:44:28,988][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:44:29,491][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:44:29,989][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:44:30,487][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:44:30,988][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:44:31,487][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:44:31,985][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:44:32,487][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:44:32,987][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:44:33,487][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:44:33,985][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:44:34,486][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:44:34,990][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:44:35,490][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:44:35,986][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:44:36,488][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:44:36,991][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:44:37,488][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:44:37,988][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:44:38,487][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:44:38,986][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:44:39,485][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:44:39,984][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:44:40,480][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:44:40,981][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:44:41,492][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:44:41,991][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:44:42,494][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:44:43,015][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:44:43,514][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:44:44,013][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:44:44,511][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:44:45,010][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:44:45,507][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:44:46,007][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:44:46,503][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:44:47,002][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:44:47,500][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:44:48,002][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:44:48,501][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:44:48,998][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:44:49,496][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:44:49,993][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:44:50,491][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:44:50,988][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:44:51,484][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:44:51,981][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:44:52,479][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:44:52,978][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:44:53,476][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:44:53,975][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:44:54,474][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:44:54,975][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:44:55,472][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:44:55,969][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:44:56,467][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:44:56,965][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:44:57,467][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:44:57,961][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:44:58,459][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:44:58,963][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:44:59,461][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:44:59,961][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10040 tokens. +[2026-03-26 04:45:00,746][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.15%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 04:45:01,501][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:45:01,504][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:45:01,506][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:45:02,599][__main__][INFO] - Iteration 447 took 2m 35s (76.71% Gen, 22.59% Train). Generation: 1m 59s, Training: 35s. Estimated remaining time: 122h 43m 21s. Estimated total time: 129h 27m 34s. Time estimates for 10 more iterations: 25m 53s, 100 more iterations: 4h 18m 55s, 500 more iterations: 21h 34m 35s. +[2026-03-26 04:45:02,601][__main__][INFO] - Starting iteration 447. +[2026-03-26 04:45:03,023][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 44 and human policies 1. +[2026-03-26 04:45:03,023][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:45:20,435][__main__][INFO] - Number of regex retries in iteration 447: 0 +[2026-03-26 04:45:20,436][__main__][INFO] - agents played in iteration 447 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:45:21,213][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:45:21,237][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:45:21,261][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:45:21,284][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:45:21,285][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:45:21,285][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:45:22,047][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:45:22,501][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:45:23,010][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:45:23,516][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:45:24,017][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:45:24,516][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:45:25,021][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:45:25,520][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:45:26,020][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:45:26,524][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:45:27,037][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:45:27,551][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:45:28,056][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:45:28,560][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:45:29,085][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:45:29,591][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:45:30,098][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:45:30,601][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:45:31,105][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:45:31,606][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:45:32,107][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:45:32,608][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:45:33,109][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:45:33,623][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:45:34,128][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:45:34,631][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:45:35,129][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:45:35,630][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:45:36,133][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:45:36,635][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:45:37,137][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:45:37,648][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:45:38,151][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:45:38,652][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:45:39,154][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:45:39,658][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:45:40,157][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:45:40,656][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:45:41,155][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:45:41,658][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:45:42,158][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:45:42,661][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:45:43,165][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:45:43,668][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:45:44,167][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:45:44,667][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:45:45,173][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:45:45,677][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:45:46,184][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:45:46,684][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:45:47,186][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:45:47,686][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:45:48,190][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:45:48,689][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:45:49,188][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:45:49,688][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:45:50,187][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:45:50,688][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:45:51,196][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:45:51,701][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:45:52,202][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:45:52,705][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:45:53,206][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:45:53,707][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:45:54,208][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10109 tokens. +[2026-03-26 04:45:55,006][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.28%, ΔTime: 00:00:32 +[2026-03-26 04:45:55,741][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:45:55,743][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:45:55,745][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:45:56,428][__main__][INFO] - Iteration 448 took 53s (32.60% Gen, 66.11% Train). Generation: 17s, Training: 35s. Estimated remaining time: 37h 45m 9s. Estimated total time: 44h 30m 16s. Time estimates for 10 more iterations: 8m 54s, 100 more iterations: 1h 29m 0s, 500 more iterations: 7h 25m 2s. +[2026-03-26 04:45:56,430][__main__][INFO] - Starting iteration 448. +[2026-03-26 04:45:56,846][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 44 and human policies 1. +[2026-03-26 04:45:56,847][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:46:06,339][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 20 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:46:07,411][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 20 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 2/3 +[2026-03-26 04:46:11,785][__main__][INFO] - Number of regex retries in iteration 448: 2 +[2026-03-26 04:46:11,786][__main__][INFO] - agents played in iteration 448 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:46:12,569][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:46:12,594][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:46:12,618][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:46:12,643][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:46:12,644][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:46:12,644][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:46:13,390][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:46:13,846][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:46:14,348][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:46:14,846][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:46:15,344][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:46:15,842][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:46:16,340][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:46:16,840][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:46:17,339][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:46:17,842][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:46:18,346][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:46:18,857][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:46:19,357][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:46:19,857][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:46:20,359][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:46:20,857][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:46:21,357][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:46:21,862][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:46:22,392][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:46:22,892][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:46:23,396][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:46:23,897][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:46:24,399][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:46:24,901][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:46:25,404][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:46:25,909][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:46:26,409][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:46:26,908][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:46:27,407][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:46:27,904][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:46:28,402][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:46:28,901][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:46:29,403][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:46:29,900][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:46:30,402][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:46:30,899][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:46:31,400][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:46:31,899][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:46:32,405][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:46:32,911][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:46:33,414][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:46:33,912][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:46:34,410][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:46:34,911][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:46:35,409][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:46:35,906][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:46:36,408][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:46:36,906][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:46:37,416][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:46:37,915][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:46:38,414][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:46:38,913][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:46:39,412][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:46:39,911][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:46:40,412][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:46:40,910][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:46:41,407][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:46:41,903][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:46:42,399][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:46:42,895][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:46:43,393][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:46:43,890][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:46:44,386][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:46:44,881][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:46:45,377][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9922 tokens. +[2026-03-26 04:46:46,176][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.99%, Current % of VRAM taken: 60.41%, Block Peak % of device VRAM: 62.19%, ΔTime: 00:00:32 +[2026-03-26 04:46:46,888][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:46:46,890][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:46:46,892][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:46:48,059][__main__][INFO] - Iteration 449 took 51s (29.17% Gen, 68.55% Train). Generation: 14s, Training: 35s. Estimated remaining time: 35h 54m 42s. Estimated total time: 42h 40m 41s. Time estimates for 10 more iterations: 8m 32s, 100 more iterations: 1h 25m 21s, 500 more iterations: 7h 6m 46s. +[2026-03-26 04:46:48,061][__main__][INFO] - Starting iteration 449. +[2026-03-26 04:46:48,462][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 44 and human policies 1. +[2026-03-26 04:46:48,463][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:46:58,185][mllm.models.large_language_model_local][WARNING] - Response Proposal: x hats, 10 books, 10 balls + +In this round, I've proposed keeping all 10 balls and all 10 books, and none of the hats. Since my value for books is significantly lower than Bob's, and my value for hats is the same as Bob's, it makes sense to take the items where I have the higher value. Balls have a much higher value for me, so I'll claim all of those as well. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:47:03,931][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 0 balls + +Given Alice's high value for hats and books, it's strategic for you to maximize your value for balls, which you and Alice both value relatively highly. By proposing to take all 10 balls, you can secure the highest possible points for this round. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:47:04,560][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 0 balls + +This proposal might seem counterintuitive given the values, but it's a strategic decision to avoid the risk of a proportional split in case Bob also opts for a similar strategy, which might lead to a lower combined outcome compared to focusing on items with higher individual worth. However, if both opt for this, the game could be more unpredictable and might warrant adjusting the strategy based on Bob's previous behavior. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:47:05,630][__main__][INFO] - Number of regex retries in iteration 449: 3 +[2026-03-26 04:47:05,631][__main__][INFO] - agents played in iteration 449 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:47:06,417][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:47:06,442][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:47:06,467][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:47:06,492][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:47:06,493][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:47:06,493][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:47:07,241][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:47:07,698][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:47:08,206][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:47:08,709][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:47:09,211][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:47:09,712][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:47:10,211][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:47:10,712][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:47:11,212][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:47:11,712][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:47:12,207][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:47:12,706][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:47:13,210][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:47:13,709][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:47:14,213][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:47:14,713][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:47:15,213][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:47:15,710][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:47:16,208][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:47:16,705][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:47:17,201][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:47:17,699][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:47:18,197][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:47:18,697][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:47:19,196][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:47:19,692][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:47:20,190][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:47:20,686][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:47:21,182][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:47:21,680][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:47:22,177][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:47:22,675][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:47:23,175][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:47:23,674][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:47:24,173][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:47:24,671][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:47:25,170][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:47:25,669][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:47:26,168][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:47:26,680][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:47:27,181][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:47:27,684][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:47:28,186][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:47:28,688][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:47:29,190][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:47:29,692][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:47:30,193][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:47:30,694][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:47:31,196][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:47:31,704][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:47:32,209][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:47:32,711][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:47:33,215][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:47:33,720][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:47:34,224][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:47:34,723][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:47:35,222][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:47:35,722][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:47:36,224][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:47:36,728][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:47:37,241][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:47:37,742][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:47:38,244][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:47:38,744][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:47:39,243][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9965 tokens. +[2026-03-26 04:47:40,076][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.00%, Current % of VRAM taken: 60.43%, Block Peak % of device VRAM: 62.21%, ΔTime: 00:00:32 +[2026-03-26 04:47:40,813][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:47:40,816][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:47:40,817][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:47:41,686][__main__][INFO] - Iteration 450 took 53s (32.26% Gen, 66.11% Train). Generation: 17s, Training: 35s. Estimated remaining time: 37h 34m 21s. Estimated total time: 44h 21m 14s. Time estimates for 10 more iterations: 8m 52s, 100 more iterations: 1h 28m 42s, 500 more iterations: 7h 23m 32s. +[2026-03-26 04:47:41,688][__main__][INFO] - Starting iteration 450. +[2026-03-26 04:47:42,097][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 44 and human policies 1. +[2026-03-26 04:47:42,098][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:47:47,028][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:47:58,334][__main__][INFO] - Number of regex retries in iteration 450: 1 +[2026-03-26 04:47:58,335][__main__][INFO] - agents played in iteration 450 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:47:59,110][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:47:59,134][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:47:59,157][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:47:59,180][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:47:59,181][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:47:59,181][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:47:59,932][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:48:00,384][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:48:00,892][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:48:01,391][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:48:01,896][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:48:02,397][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:48:02,894][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:48:03,392][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:48:03,891][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:48:04,395][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:48:04,896][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:48:05,394][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:48:05,893][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:48:06,392][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:48:06,891][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:48:07,390][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:48:07,889][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:48:08,404][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:48:08,910][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:48:09,413][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:48:09,921][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:48:10,431][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:48:10,939][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:48:11,450][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:48:11,956][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:48:12,459][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:48:12,966][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:48:13,467][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:48:13,970][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:48:14,469][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:48:14,969][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:48:15,469][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:48:15,969][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:48:16,469][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:48:16,969][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:48:17,464][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:48:17,965][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:48:18,466][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:48:18,966][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:48:19,467][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:48:19,963][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:48:20,461][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:48:20,960][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:48:21,459][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:48:21,958][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:48:22,456][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:48:22,955][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:48:23,451][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:48:23,962][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:48:24,465][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:48:24,967][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:48:25,472][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:48:25,977][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:48:26,479][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:48:26,981][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:48:27,486][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:48:27,988][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:48:28,488][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:48:28,987][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:48:29,486][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:48:29,986][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:48:30,485][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:48:30,985][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:48:31,483][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:48:31,984][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10000 tokens. +[2026-03-26 04:48:32,790][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.26%, ΔTime: 00:00:32 +[2026-03-26 04:48:33,583][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:48:33,586][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:48:33,588][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:48:35,162][__main__][INFO] - Iteration 451 took 53s (30.60% Gen, 66.43% Train). Generation: 16s, Training: 35s. Estimated remaining time: 37h 25m 29s. Estimated total time: 44h 13m 16s. Time estimates for 10 more iterations: 8m 50s, 100 more iterations: 1h 28m 26s, 500 more iterations: 7h 22m 12s. +[2026-03-26 04:48:35,164][__main__][INFO] - Starting iteration 451. +[2026-03-26 04:48:35,579][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 45 and human policies 1. +[2026-03-26 04:48:35,580][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:48:51,065][__main__][INFO] - Number of regex retries in iteration 451: 0 +[2026-03-26 04:48:51,066][__main__][INFO] - agents played in iteration 451 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:48:51,834][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:48:51,859][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:48:51,883][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:48:51,907][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:48:51,908][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:48:51,908][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:48:52,610][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:48:53,066][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:48:53,567][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:48:54,074][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:48:54,574][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:48:55,073][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:48:55,571][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:48:56,086][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:48:56,595][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:48:57,097][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:48:57,599][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:48:58,115][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:48:58,616][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:48:59,126][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:48:59,630][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:49:00,134][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:49:00,644][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:49:01,150][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:49:01,654][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:49:02,159][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:49:02,662][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:49:03,164][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:49:03,668][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:49:04,186][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:49:04,690][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:49:05,193][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:49:05,701][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:49:06,210][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:49:06,717][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:49:07,218][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:49:07,719][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:49:08,218][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:49:08,719][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:49:09,221][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:49:09,722][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:49:10,228][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:49:10,725][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:49:11,226][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:49:11,728][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:49:12,229][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:49:12,747][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:49:13,250][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:49:13,751][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:49:14,256][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:49:14,759][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:49:15,261][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:49:15,762][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:49:16,265][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:49:16,767][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:49:17,271][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:49:17,771][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:49:18,273][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:49:18,773][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:49:19,276][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:49:19,780][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:49:20,283][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:49:20,782][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:49:21,282][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:49:21,781][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:49:22,278][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:49:22,778][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:49:23,291][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:49:23,797][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:49:24,303][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:49:24,811][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9978 tokens. +[2026-03-26 04:49:25,593][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.22%, Current % of VRAM taken: 60.65%, Block Peak % of device VRAM: 62.19%, ΔTime: 00:00:32 +[2026-03-26 04:49:26,330][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:49:26,332][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:49:26,334][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:49:27,192][__main__][INFO] - Iteration 452 took 51s (30.01% Gen, 68.33% Train). Generation: 15s, Training: 35s. Estimated remaining time: 36h 12m 2s. Estimated total time: 43h 0m 40s. Time estimates for 10 more iterations: 8m 36s, 100 more iterations: 1h 26m 1s, 500 more iterations: 7h 10m 6s. +[2026-03-26 04:49:27,194][__main__][INFO] - Starting iteration 452. +[2026-03-26 04:49:27,623][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 45 and human policies 1. +[2026-03-26 04:49:27,623][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:49:31,850][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:49:38,076][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 1 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:49:38,428][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:49:43,169][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 0 balls + +This round seems不利的策略。考虑到你的物品价值相对较低（1 和 10），而鲍勃的物品价值相对较高（10 和 1），你应该争取尽可能多的价值更高的物品。在这种情况下，你和鲍勃都选择了不获取任何物品，这导致了两人的收益都非常低。你会失去获取更多高价值物品的机会。 + +建议调整策略，尝试获取一些低价值但稳定的物品，以增加总得分。 did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:49:44,106][__main__][INFO] - Number of regex retries in iteration 452: 4 +[2026-03-26 04:49:44,106][__main__][INFO] - agents played in iteration 452 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:49:44,884][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:49:44,910][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:49:44,937][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:49:44,963][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:49:44,964][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:49:44,964][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:49:45,749][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:49:46,204][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:49:46,708][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:49:47,209][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:49:47,719][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:49:48,220][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:49:48,720][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:49:49,221][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:49:49,719][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:49:50,219][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:49:50,718][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:49:51,218][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:49:51,717][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:49:52,230][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:49:52,742][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:49:53,242][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:49:53,745][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:49:54,274][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:49:54,784][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:49:55,289][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:49:55,797][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:49:56,303][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:49:56,809][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:49:57,323][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:49:57,831][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:49:58,334][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:49:58,839][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:49:59,340][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:49:59,841][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:50:00,341][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:50:00,842][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:50:01,343][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:50:01,842][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:50:02,342][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:50:02,842][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:50:03,355][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:50:03,857][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:50:04,360][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:50:04,861][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:50:05,364][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:50:05,865][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:50:06,367][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:50:06,867][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:50:07,371][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:50:07,871][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:50:08,377][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:50:08,881][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:50:09,384][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:50:09,885][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:50:10,388][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:50:10,896][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:50:11,400][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:50:11,902][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:50:12,403][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:50:12,902][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:50:13,402][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:50:13,901][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:50:14,398][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:50:14,894][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:50:15,390][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:50:15,885][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:50:16,381][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:50:16,882][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:50:17,380][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:50:17,876][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9999 tokens. +[2026-03-26 04:50:18,651][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.01%, Current % of VRAM taken: 60.44%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 04:50:19,409][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:50:19,411][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:50:19,413][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:50:20,363][__main__][INFO] - Iteration 453 took 52s (31.25% Gen, 66.94% Train). Generation: 16s, Training: 35s. Estimated remaining time: 37h 7m 31s. Estimated total time: 43h 57m 2s. Time estimates for 10 more iterations: 8m 47s, 100 more iterations: 1h 27m 54s, 500 more iterations: 7h 19m 30s. +[2026-03-26 04:50:20,365][__main__][INFO] - Starting iteration 453. +[2026-03-26 04:50:20,766][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 45 and human policies 1. +[2026-03-26 04:50:20,767][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:53:08,158][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:53:20,720][__main__][INFO] - Number of regex retries in iteration 453: 1 +[2026-03-26 04:53:20,721][__main__][INFO] - agents played in iteration 453 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:53:21,495][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:53:21,519][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:53:21,543][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:53:21,566][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:53:21,567][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:53:21,567][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:53:22,305][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:53:22,761][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:53:23,268][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:53:23,774][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:53:24,278][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:53:24,780][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:53:25,281][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:53:25,787][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:53:26,291][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:53:26,790][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:53:27,288][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:53:27,789][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:53:28,287][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:53:28,787][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:53:29,284][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:53:29,783][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:53:30,295][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:53:30,794][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:53:31,295][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:53:31,795][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:53:32,297][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:53:32,795][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:53:33,295][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:53:33,794][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:53:34,292][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:53:34,786][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:53:35,281][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:53:35,779][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:53:36,274][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:53:36,774][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:53:37,271][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:53:37,770][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:53:38,266][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:53:38,761][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:53:39,254][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:53:39,752][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:53:40,253][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:53:40,751][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:53:41,251][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:53:41,746][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:53:42,244][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:53:42,752][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:53:43,250][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:53:43,747][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:53:44,245][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:53:44,742][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:53:45,241][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:53:45,740][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:53:46,235][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:53:46,732][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:53:47,229][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:53:47,726][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:53:48,231][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:53:48,730][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:53:49,228][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:53:49,725][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:53:50,225][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:53:50,724][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:53:51,222][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:53:51,721][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:53:52,222][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:53:52,721][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:53:53,219][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:53:53,719][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:53:54,221][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10045 tokens. +[2026-03-26 04:53:54,919][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 62.29%, ΔTime: 00:00:32 +[2026-03-26 04:53:55,686][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:53:55,688][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:53:55,690][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:53:56,477][__main__][INFO] - Iteration 454 took 3m 35s (83.42% Gen, 16.21% Train). Generation: 2m 59s, Training: 34s. Estimated remaining time: 172h 52m 26s. Estimated total time: 179h 45m 34s. Time estimates for 10 more iterations: 35m 57s, 100 more iterations: 5h 59m 31s, 500 more iterations: 29h 57m 35s. +[2026-03-26 04:53:56,479][__main__][INFO] - Starting iteration 454. +[2026-03-26 04:53:56,882][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 45 and human policies 1. +[2026-03-26 04:53:56,883][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:54:35,972][__main__][INFO] - Number of regex retries in iteration 454: 0 +[2026-03-26 04:54:35,973][__main__][INFO] - agents played in iteration 454 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:54:36,744][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:54:36,767][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:54:36,791][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:54:36,814][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:54:36,815][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:54:36,815][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:54:37,522][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:54:37,975][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:54:38,476][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:54:38,975][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:54:39,473][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:54:39,974][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:54:40,471][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:54:40,968][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:54:41,465][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:54:41,966][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:54:42,465][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:54:42,965][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:54:43,466][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:54:43,965][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:54:44,465][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:54:44,964][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:54:45,463][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:54:45,983][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:54:46,483][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:54:46,983][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:54:47,485][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:54:47,983][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:54:48,488][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:54:48,990][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:54:49,524][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:54:50,020][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:54:50,522][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:54:51,020][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:54:51,519][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:54:52,019][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:54:52,516][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:54:53,018][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:54:53,518][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:54:54,017][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:54:54,513][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:54:55,008][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:54:55,512][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:54:56,015][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:54:56,511][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:54:57,010][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:54:57,515][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:54:58,023][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:54:58,523][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:54:59,027][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:54:59,531][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:55:00,037][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:55:00,542][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:55:01,047][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:55:01,548][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:55:02,050][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:55:02,564][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:55:03,069][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:55:03,573][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:55:04,075][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:55:04,576][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:55:05,081][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:55:05,582][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:55:06,082][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:55:06,582][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:55:07,083][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:55:07,585][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:55:08,086][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:55:08,588][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:55:09,090][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:55:09,589][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10007 tokens. +[2026-03-26 04:55:10,370][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.49%, ΔTime: 00:00:32 +[2026-03-26 04:55:11,111][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:55:11,114][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:55:11,115][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:55:11,837][__main__][INFO] - Iteration 455 took 1m 14s (52.15% Gen, 46.88% Train). Generation: 39s, Training: 35s. Estimated remaining time: 55h 33m 23s. Estimated total time: 62h 27m 46s. Time estimates for 10 more iterations: 12m 29s, 100 more iterations: 2h 4m 55s, 500 more iterations: 10h 24m 37s. +[2026-03-26 04:55:11,839][__main__][INFO] - Starting iteration 455. +[2026-03-26 04:55:12,254][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 45 and human policies 1. +[2026-03-26 04:55:12,255][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:55:35,169][__main__][INFO] - Number of regex retries in iteration 455: 0 +[2026-03-26 04:55:35,169][__main__][INFO] - agents played in iteration 455 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:55:35,930][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:55:35,954][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:55:35,977][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:55:36,001][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:55:36,001][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:55:36,002][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:55:36,738][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:55:37,190][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:55:37,693][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:55:38,190][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:55:38,687][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:55:39,185][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:55:39,679][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:55:40,176][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:55:40,672][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:55:41,169][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:55:41,665][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:55:42,162][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:55:42,660][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:55:43,158][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:55:43,656][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:55:44,153][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:55:44,650][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:55:45,147][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:55:45,642][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:55:46,139][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:55:46,637][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:55:47,135][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:55:47,633][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:55:48,137][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:55:48,633][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:55:49,130][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:55:49,627][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:55:50,127][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:55:50,626][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:55:51,129][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:55:51,632][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:55:52,131][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:55:52,631][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:55:53,131][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:55:53,633][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:55:54,132][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:55:54,627][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:55:55,124][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:55:55,622][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:55:56,117][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:55:56,613][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:55:57,111][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:55:57,608][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:55:58,104][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:55:58,601][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:55:59,099][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:55:59,599][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:56:00,101][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:56:00,598][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:56:01,098][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:56:01,598][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:56:02,097][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:56:02,594][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:56:03,091][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:56:03,611][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:56:04,109][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:56:04,606][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:56:05,103][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:56:05,599][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:56:06,093][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:56:06,590][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:56:07,083][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:56:07,577][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:56:08,073][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:56:08,567][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9903 tokens. +[2026-03-26 04:56:09,329][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.00%, Current % of VRAM taken: 60.43%, Block Peak % of device VRAM: 62.05%, ΔTime: 00:00:32 +[2026-03-26 04:56:10,021][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:56:10,023][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:56:10,025][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:56:10,696][__main__][INFO] - Iteration 456 took 58s (39.21% Gen, 59.64% Train). Generation: 22s, Training: 34s. Estimated remaining time: 41h 46m 46s. Estimated total time: 48h 42m 8s. Time estimates for 10 more iterations: 9m 44s, 100 more iterations: 1h 37m 24s, 500 more iterations: 8h 7m 1s. +[2026-03-26 04:56:10,698][__main__][INFO] - Starting iteration 456. +[2026-03-26 04:56:11,120][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 45 and human policies 1. +[2026-03-26 04:56:11,121][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:56:47,799][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:56:50,242][__main__][INFO] - Number of regex retries in iteration 456: 1 +[2026-03-26 04:56:50,242][__main__][INFO] - agents played in iteration 456 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:56:51,009][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:56:51,033][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:56:51,056][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:56:51,080][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:56:51,080][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:56:51,081][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:56:51,815][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:56:52,269][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:56:52,771][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:56:53,270][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:56:53,768][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:56:54,270][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:56:54,767][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:56:55,265][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:56:55,764][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:56:56,260][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:56:56,758][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:56:57,254][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:56:57,751][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:56:58,248][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:56:58,746][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:56:59,241][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:56:59,743][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:57:00,241][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:57:00,740][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:57:01,239][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:57:01,744][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:57:02,243][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:57:02,740][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:57:03,262][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:57:03,760][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:57:04,257][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:57:04,756][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:57:05,254][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:57:05,759][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:57:06,263][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:57:06,762][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:57:07,263][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:57:07,763][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:57:08,264][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:57:08,764][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:57:09,265][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:57:09,766][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:57:10,265][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:57:10,768][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:57:11,267][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:57:11,766][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:57:12,267][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:57:12,766][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:57:13,267][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:57:13,764][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:57:14,263][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:57:14,762][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:57:15,261][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:57:15,761][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:57:16,262][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:57:16,763][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:57:17,263][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:57:17,765][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:57:18,265][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:57:18,763][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:57:19,265][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:57:19,768][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:57:20,267][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:57:20,764][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:57:21,263][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:57:21,761][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:57:22,259][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:57:22,760][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:57:23,260][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:57:23,759][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9966 tokens. +[2026-03-26 04:57:24,527][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.32%, ΔTime: 00:00:32 +[2026-03-26 04:57:25,320][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:57:25,322][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:57:25,323][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:57:26,091][__main__][INFO] - Iteration 457 took 1m 14s (52.18% Gen, 46.79% Train). Generation: 39s, Training: 35s. Estimated remaining time: 55h 31m 59s. Estimated total time: 62h 28m 36s. Time estimates for 10 more iterations: 12m 29s, 100 more iterations: 2h 4m 57s, 500 more iterations: 10h 24m 46s. +[2026-03-26 04:57:26,093][__main__][INFO] - Starting iteration 457. +[2026-03-26 04:57:26,514][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 45 and human policies 1. +[2026-03-26 04:57:26,514][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:57:44,631][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:57:44,711][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 04:57:55,605][__main__][INFO] - Number of regex retries in iteration 457: 2 +[2026-03-26 04:57:55,606][__main__][INFO] - agents played in iteration 457 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:57:56,375][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:57:56,398][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:57:56,422][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:57:56,445][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:57:56,446][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:57:56,446][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:57:57,193][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:57:57,655][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:57:58,162][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:57:58,665][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:57:59,191][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:57:59,707][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:58:00,211][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:58:00,715][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:58:01,223][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:58:01,723][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:58:02,229][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:58:02,731][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:58:03,236][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:58:03,753][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:58:04,259][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:58:04,762][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:58:05,262][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:58:05,767][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:58:06,269][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:58:06,782][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:58:07,292][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:58:07,797][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:58:08,299][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:58:08,803][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:58:09,306][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:58:09,807][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:58:10,312][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:58:10,820][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:58:11,323][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:58:11,825][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:58:12,332][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:58:12,829][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:58:13,332][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:58:13,831][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:58:14,331][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:58:14,828][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:58:15,325][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:58:15,832][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:58:16,334][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:58:16,832][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:58:17,339][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:58:17,843][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:58:18,346][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:58:18,847][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:58:19,347][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:58:19,847][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:58:20,349][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:58:20,847][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:58:21,346][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:58:21,846][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:58:22,351][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:58:22,862][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:58:23,368][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:58:23,868][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:58:24,369][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:58:24,870][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:58:25,371][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:58:25,869][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:58:26,368][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:58:26,866][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:58:27,364][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:58:27,863][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:58:28,361][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:58:28,860][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:58:29,375][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9971 tokens. +[2026-03-26 04:58:30,151][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.24%, Current % of VRAM taken: 60.67%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 04:58:30,884][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:58:30,887][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:58:30,888][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:58:31,613][__main__][INFO] - Iteration 458 took 1m 5s (44.69% Gen, 54.20% Train). Generation: 29s, Training: 35s. Estimated remaining time: 47h 17m 18s. Estimated total time: 54h 15m 1s. Time estimates for 10 more iterations: 10m 51s, 100 more iterations: 1h 48m 30s, 500 more iterations: 9h 2m 30s. +[2026-03-26 04:58:31,616][__main__][INFO] - Starting iteration 458. +[2026-03-26 04:58:32,053][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 45 and human policies 1. +[2026-03-26 04:58:32,054][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:58:46,437][__main__][INFO] - Number of regex retries in iteration 458: 0 +[2026-03-26 04:58:46,438][__main__][INFO] - agents played in iteration 458 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:58:47,202][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:58:47,226][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:58:47,249][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:58:47,272][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:58:47,273][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:58:47,274][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:58:48,020][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:58:48,473][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:58:48,980][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:58:49,490][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:58:49,995][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:58:50,505][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:58:51,010][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:58:51,513][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:58:52,017][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:58:52,521][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:58:53,023][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:58:53,526][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:58:54,027][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:58:54,529][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:58:55,031][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:58:55,529][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:58:56,027][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:58:56,525][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:58:57,023][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:58:57,524][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:58:58,024][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:58:58,522][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:58:59,020][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:58:59,519][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:59:00,016][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:59:00,520][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:59:01,020][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:59:01,521][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:59:02,021][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:59:02,522][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:59:03,022][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:59:03,521][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:59:04,021][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:59:04,515][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:59:05,010][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:59:05,505][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:59:06,000][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:59:06,497][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:59:06,991][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 04:59:07,484][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 04:59:07,980][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 04:59:08,478][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 04:59:08,977][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 04:59:09,479][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 04:59:09,978][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 04:59:10,484][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 04:59:10,985][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 04:59:11,488][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 04:59:11,987][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 04:59:12,487][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 04:59:12,986][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 04:59:13,488][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 04:59:13,986][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 04:59:14,486][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 04:59:14,986][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 04:59:15,486][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 04:59:15,986][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 04:59:16,508][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 04:59:17,008][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 04:59:17,515][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 04:59:18,020][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 04:59:18,524][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 04:59:19,031][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 04:59:19,534][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 04:59:20,036][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9951 tokens. +[2026-03-26 04:59:20,827][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.61%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 04:59:21,550][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 04:59:21,552][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 04:59:21,554][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 04:59:22,452][__main__][INFO] - Iteration 459 took 50s (28.54% Gen, 69.67% Train). Generation: 14s, Training: 35s. Estimated remaining time: 35h 1m 24s. Estimated total time: 41h 59m 58s. Time estimates for 10 more iterations: 8m 23s, 100 more iterations: 1h 23m 59s, 500 more iterations: 6h 59m 59s. +[2026-03-26 04:59:22,454][__main__][INFO] - Starting iteration 459. +[2026-03-26 04:59:22,920][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 45 and human policies 1. +[2026-03-26 04:59:22,921][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 04:59:39,291][__main__][INFO] - Number of regex retries in iteration 459: 0 +[2026-03-26 04:59:39,292][__main__][INFO] - agents played in iteration 459 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 04:59:40,070][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:59:40,094][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:59:40,117][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:59:40,140][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 04:59:40,141][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 04:59:40,141][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 04:59:40,869][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 04:59:41,325][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 04:59:41,826][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 04:59:42,321][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 04:59:42,816][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 04:59:43,321][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 04:59:43,819][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 04:59:44,316][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 04:59:44,814][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 04:59:45,309][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 04:59:45,805][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 04:59:46,302][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 04:59:46,798][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 04:59:47,300][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 04:59:47,802][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 04:59:48,298][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 04:59:48,796][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 04:59:49,297][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 04:59:49,796][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 04:59:50,296][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 04:59:50,803][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 04:59:51,299][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 04:59:51,799][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 04:59:52,299][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 04:59:52,799][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 04:59:53,297][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 04:59:53,792][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 04:59:54,289][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 04:59:54,785][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 04:59:55,282][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 04:59:55,776][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 04:59:56,272][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 04:59:56,768][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 04:59:57,267][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 04:59:57,769][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 04:59:58,271][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 04:59:58,773][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 04:59:59,272][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 04:59:59,773][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:00:00,277][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:00:00,779][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:00:01,280][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:00:01,777][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:00:02,281][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:00:02,782][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:00:03,281][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:00:03,779][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:00:04,274][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:00:04,771][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:00:05,266][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:00:05,767][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:00:06,263][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:00:06,765][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:00:07,283][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:00:07,784][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:00:08,283][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:00:08,780][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:00:09,283][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:00:09,778][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:00:10,276][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:00:10,778][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:00:11,281][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:00:11,782][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:00:12,284][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:00:12,786][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9850 tokens. +[2026-03-26 05:00:13,547][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.17%, Current % of VRAM taken: 60.60%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 05:00:14,348][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:00:14,350][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:00:14,352][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:00:15,089][__main__][INFO] - Iteration 460 took 52s (31.38% Gen, 67.21% Train). Generation: 16s, Training: 35s. Estimated remaining time: 36h 29m 1s. Estimated total time: 43h 28m 27s. Time estimates for 10 more iterations: 8m 41s, 100 more iterations: 1h 26m 56s, 500 more iterations: 7h 14m 44s. +[2026-03-26 05:00:15,091][__main__][INFO] - Starting iteration 460. +[2026-03-26 05:00:15,503][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 45 and human policies 1. +[2026-03-26 05:00:15,504][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:01:09,187][__main__][INFO] - Number of regex retries in iteration 460: 0 +[2026-03-26 05:01:09,188][__main__][INFO] - agents played in iteration 460 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:01:09,981][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:01:10,004][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:01:10,028][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:01:10,051][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:01:10,052][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:01:10,052][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:01:10,781][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:01:11,240][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:01:11,750][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:01:12,248][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:01:12,742][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:01:13,238][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:01:13,733][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:01:14,236][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:01:14,730][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:01:15,225][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:01:15,722][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:01:16,219][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:01:16,715][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:01:17,217][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:01:17,717][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:01:18,223][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:01:18,728][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:01:19,227][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:01:19,725][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:01:20,222][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:01:20,719][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:01:21,218][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:01:21,715][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:01:22,213][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:01:22,709][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:01:23,206][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:01:23,705][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:01:24,212][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:01:24,710][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:01:25,209][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:01:25,707][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:01:26,206][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:01:26,704][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:01:27,202][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:01:27,699][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:01:28,196][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:01:28,693][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:01:29,191][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:01:29,695][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:01:30,195][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:01:30,692][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:01:31,189][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:01:31,686][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:01:32,184][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:01:32,681][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:01:33,178][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:01:33,674][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:01:34,171][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:01:34,668][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:01:35,164][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:01:35,661][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:01:36,168][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:01:36,668][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:01:37,164][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:01:37,660][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:01:38,157][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:01:38,656][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:01:39,155][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:01:39,653][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:01:40,152][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:01:40,656][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:01:41,160][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:01:41,661][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:01:42,161][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:01:42,665][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9849 tokens. +[2026-03-26 05:01:43,473][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.09%, ΔTime: 00:00:32 +[2026-03-26 05:01:44,290][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:01:44,292][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:01:44,293][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:01:45,948][__main__][INFO] - Iteration 461 took 1m 30s (59.36% Gen, 38.81% Train). Generation: 53s, Training: 35s. Estimated remaining time: 68h 21m 21s. Estimated total time: 75h 22m 18s. Time estimates for 10 more iterations: 15m 4s, 100 more iterations: 2h 30m 44s, 500 more iterations: 12h 33m 43s. +[2026-03-26 05:01:45,950][__main__][INFO] - Starting iteration 461. +[2026-03-26 05:01:46,372][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 46 and human policies 1. +[2026-03-26 05:01:46,373][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:01:55,419][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 20 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:02:03,113][__main__][INFO] - Number of regex retries in iteration 461: 1 +[2026-03-26 05:02:03,114][__main__][INFO] - agents played in iteration 461 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:02:03,889][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:02:03,912][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:02:03,936][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:02:03,959][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:02:03,960][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:02:03,960][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:02:04,708][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:02:05,163][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:02:05,663][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:02:06,158][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:02:06,657][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:02:07,154][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:02:07,648][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:02:08,144][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:02:08,642][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:02:10,594][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:02:11,098][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:02:11,613][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:02:12,139][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:02:12,645][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:02:13,147][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:02:13,650][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:02:14,154][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:02:14,647][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:02:15,147][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:02:15,646][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:02:16,148][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:02:16,646][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:02:17,145][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:02:17,639][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:02:18,138][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:02:18,632][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:02:19,137][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:02:19,632][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:02:20,130][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:02:20,626][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:02:21,119][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:02:21,613][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:02:22,111][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:02:22,608][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:02:23,112][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:02:23,611][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:02:24,110][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:02:24,609][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:02:25,108][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:02:25,607][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:02:26,107][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:02:26,605][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:02:27,104][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:02:27,604][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:02:28,102][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:02:28,602][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:02:29,100][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:02:29,599][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:02:30,098][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:02:30,599][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:02:31,101][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:02:31,603][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:02:32,105][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:02:32,612][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:02:33,118][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:02:33,616][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:02:34,119][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:02:34,622][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:02:35,130][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:02:35,634][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:02:36,140][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:02:36,637][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:02:37,133][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:02:37,631][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:02:38,131][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9953 tokens. +[2026-03-26 05:02:38,921][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.19%, Current % of VRAM taken: 60.61%, Block Peak % of device VRAM: 62.26%, ΔTime: 00:00:34 +[2026-03-26 05:02:39,658][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:02:39,661][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:02:39,662][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:02:40,506][__main__][INFO] - Iteration 462 took 54s (30.92% Gen, 67.51% Train). Generation: 16s, Training: 36s. Estimated remaining time: 38h 4m 53s. Estimated total time: 45h 6m 45s. Time estimates for 10 more iterations: 9m 1s, 100 more iterations: 1h 30m 13s, 500 more iterations: 7h 31m 7s. +[2026-03-26 05:02:40,508][__main__][INFO] - Starting iteration 462. +[2026-03-26 05:02:40,912][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 46 and human policies 1. +[2026-03-26 05:02:40,913][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:03:07,432][__main__][INFO] - Number of regex retries in iteration 462: 0 +[2026-03-26 05:03:07,433][__main__][INFO] - agents played in iteration 462 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:03:08,200][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:03:08,224][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:03:08,248][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:03:08,271][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:03:08,272][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:03:08,272][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:03:09,005][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:03:09,457][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:03:09,963][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:03:10,460][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:03:10,959][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:03:11,458][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:03:11,955][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:03:12,452][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:03:12,951][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:03:13,449][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:03:13,951][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:03:14,449][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:03:14,947][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:03:15,446][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:03:15,945][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:03:16,444][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:03:16,942][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:03:17,440][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:03:17,937][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:03:18,434][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:03:18,931][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:03:19,428][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:03:19,929][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:03:20,427][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:03:20,925][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:03:21,425][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:03:21,924][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:03:22,422][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:03:22,921][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:03:23,417][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:03:23,917][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:03:24,416][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:03:24,919][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:03:25,416][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:03:25,911][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:03:26,405][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:03:26,901][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:03:27,397][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:03:27,889][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:03:28,388][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:03:28,883][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:03:29,380][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:03:29,877][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:03:30,373][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:03:30,867][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:03:31,364][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:03:31,861][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:03:32,357][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:03:32,855][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:03:33,351][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:03:33,847][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:03:34,348][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:03:34,842][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:03:35,337][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:03:35,837][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:03:36,331][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:03:36,829][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:03:37,327][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:03:37,826][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:03:38,327][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:03:38,828][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:03:39,342][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:03:39,844][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:03:40,351][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:03:40,855][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9882 tokens. +[2026-03-26 05:03:41,663][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.07%, ΔTime: 00:00:32 +[2026-03-26 05:03:42,415][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:03:42,418][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:03:42,420][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:03:43,083][__main__][INFO] - Iteration 463 took 1m 2s (42.66% Gen, 56.27% Train). Generation: 26s, Training: 34s. Estimated remaining time: 44h 45m 39s. Estimated total time: 51h 48m 34s. Time estimates for 10 more iterations: 10m 21s, 100 more iterations: 1h 43m 37s, 500 more iterations: 8h 38m 5s. +[2026-03-26 05:03:43,085][__main__][INFO] - Starting iteration 463. +[2026-03-26 05:03:43,496][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 46 and human policies 1. +[2026-03-26 05:03:43,496][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:04:00,287][__main__][INFO] - Number of regex retries in iteration 463: 0 +[2026-03-26 05:04:00,288][__main__][INFO] - agents played in iteration 463 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:04:01,056][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:04:01,080][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:04:01,103][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:04:01,127][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:04:01,127][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:04:01,128][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:04:01,862][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:04:02,315][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:04:02,831][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:04:03,337][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:04:03,837][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:04:04,338][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:04:04,839][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:04:05,340][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:04:05,848][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:04:06,349][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:04:06,846][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:04:07,343][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:04:07,843][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:04:08,341][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:04:08,838][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:04:09,338][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:04:09,838][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:04:10,336][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:04:10,834][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:04:11,332][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:04:11,828][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:04:12,329][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:04:12,825][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:04:13,324][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:04:13,820][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:04:14,320][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:04:14,818][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:04:15,315][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:04:15,812][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:04:16,306][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:04:16,807][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:04:17,301][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:04:17,800][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:04:18,301][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:04:18,800][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:04:19,302][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:04:19,798][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:04:20,303][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:04:20,798][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:04:21,305][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:04:21,807][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:04:22,307][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:04:22,829][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:04:23,337][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:04:23,837][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:04:24,335][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:04:24,834][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:04:25,341][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:04:25,844][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:04:26,346][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:04:26,849][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:04:27,353][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:04:27,855][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:04:28,352][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:04:28,848][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:04:29,344][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:04:29,843][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:04:30,346][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:04:30,845][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:04:31,344][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:04:31,844][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:04:32,348][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:04:32,855][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:04:33,363][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:04:33,867][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10036 tokens. +[2026-03-26 05:04:34,671][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.14%, ΔTime: 00:00:32 +[2026-03-26 05:04:35,437][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:04:35,439][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:04:35,441][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:04:36,322][__main__][INFO] - Iteration 464 took 52s (31.79% Gen, 66.54% Train). Generation: 16s, Training: 35s. Estimated remaining time: 36h 57m 32s. Estimated total time: 44h 1m 20s. Time estimates for 10 more iterations: 8m 48s, 100 more iterations: 1h 28m 2s, 500 more iterations: 7h 20m 13s. +[2026-03-26 05:04:36,324][__main__][INFO] - Starting iteration 464. +[2026-03-26 05:04:36,732][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 46 and human policies 1. +[2026-03-26 05:04:36,732][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:04:44,372][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:04:52,842][__main__][INFO] - Number of regex retries in iteration 464: 1 +[2026-03-26 05:04:52,843][__main__][INFO] - agents played in iteration 464 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:04:53,622][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:04:53,646][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:04:53,670][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:04:53,694][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:04:53,694][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:04:53,695][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:04:54,441][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:04:54,897][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:04:55,399][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:04:55,896][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:04:56,397][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:04:56,900][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:04:57,399][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:04:57,894][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:04:58,400][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:04:58,899][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:04:59,398][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:04:59,902][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:05:00,403][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:05:00,901][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:05:01,403][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:05:01,900][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:05:02,399][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:05:02,900][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:05:03,402][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:05:03,902][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:05:04,399][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:05:04,898][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:05:05,392][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:05:05,886][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:05:06,378][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:05:06,877][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:05:07,375][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:05:07,871][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:05:08,366][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:05:08,860][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:05:09,358][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:05:09,853][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:05:10,348][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:05:10,853][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:05:11,351][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:05:11,850][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:05:12,350][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:05:12,846][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:05:13,345][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:05:13,844][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:05:14,345][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:05:14,844][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:05:15,344][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:05:15,844][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:05:16,344][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:05:16,844][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:05:17,353][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:05:17,857][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:05:18,353][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:05:18,851][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:05:19,353][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:05:19,855][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:05:20,356][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:05:20,859][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:05:21,362][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:05:21,871][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:05:22,376][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:05:22,885][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:05:23,391][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:05:23,897][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:05:24,400][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:05:24,908][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:05:25,423][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:05:25,930][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:05:26,458][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9909 tokens. +[2026-03-26 05:05:27,266][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.29%, Current % of VRAM taken: 60.72%, Block Peak % of device VRAM: 62.26%, ΔTime: 00:00:32 +[2026-03-26 05:05:28,034][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:05:28,036][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:05:28,038][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:05:28,787][__main__][INFO] - Iteration 465 took 52s (30.95% Gen, 67.61% Train). Generation: 16s, Training: 35s. Estimated remaining time: 36h 18m 9s. Estimated total time: 43h 22m 49s. Time estimates for 10 more iterations: 8m 40s, 100 more iterations: 1h 26m 45s, 500 more iterations: 7h 13m 48s. +[2026-03-26 05:05:28,789][__main__][INFO] - Starting iteration 465. +[2026-03-26 05:05:29,217][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 46 and human policies 1. +[2026-03-26 05:05:29,218][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:05:33,389][mllm.models.large_language_model_local][WARNING] - Response Proposed: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:05:44,671][__main__][INFO] - Number of regex retries in iteration 465: 1 +[2026-03-26 05:05:44,672][__main__][INFO] - agents played in iteration 465 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:05:45,432][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.59%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:05:45,456][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.59%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:05:45,479][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.59%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:05:45,502][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.59%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:05:45,503][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:05:45,504][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:05:46,244][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:05:46,699][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:05:47,202][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:05:47,703][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:05:48,201][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:05:48,704][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:05:49,208][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:05:49,709][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:05:50,211][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:05:50,712][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:05:51,212][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:05:51,711][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:05:52,210][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:05:52,709][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:05:53,208][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:05:53,707][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:05:54,206][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:05:54,705][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:05:55,204][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:05:55,707][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:05:56,204][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:05:56,703][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:05:57,205][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:05:57,703][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:05:58,215][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:05:58,714][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:05:59,215][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:05:59,713][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:06:00,213][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:06:00,712][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:06:01,212][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:06:01,711][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:06:02,210][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:06:02,708][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:06:03,205][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:06:03,700][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:06:04,200][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:06:04,699][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:06:05,196][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:06:05,696][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:06:06,193][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:06:06,691][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:06:07,191][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:06:07,688][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:06:08,188][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:06:08,687][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:06:09,185][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:06:09,684][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:06:10,182][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:06:10,680][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:06:11,179][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:06:11,677][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:06:12,175][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:06:12,672][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:06:13,169][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:06:13,667][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:06:14,164][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:06:14,659][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:06:15,156][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:06:15,653][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:06:16,150][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:06:16,645][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:06:17,145][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:06:17,645][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:06:18,143][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9941 tokens. +[2026-03-26 05:06:18,954][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.02%, Current % of VRAM taken: 60.45%, Block Peak % of device VRAM: 62.06%, ΔTime: 00:00:32 +[2026-03-26 05:06:19,689][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:06:19,692][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:06:19,693][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:06:20,412][__main__][INFO] - Iteration 466 took 51s (30.19% Gen, 68.41% Train). Generation: 15s, Training: 35s. Estimated remaining time: 35h 34m 16s. Estimated total time: 42h 39m 48s. Time estimates for 10 more iterations: 8m 31s, 100 more iterations: 1h 25m 19s, 500 more iterations: 7h 6m 38s. +[2026-03-26 05:06:20,414][__main__][INFO] - Starting iteration 466. +[2026-03-26 05:06:20,818][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 46 and human policies 1. +[2026-03-26 05:06:20,818][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:06:34,867][__main__][INFO] - Number of regex retries in iteration 466: 0 +[2026-03-26 05:06:34,868][__main__][INFO] - agents played in iteration 466 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:06:35,637][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:06:35,661][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:06:35,684][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:06:35,708][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:06:35,708][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:06:35,709][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:06:36,452][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:06:36,905][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:06:37,406][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:06:37,902][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:06:38,406][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:06:38,903][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:06:39,399][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:06:39,894][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:06:40,390][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:06:40,885][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:06:41,382][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:06:41,874][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:06:42,370][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:06:42,866][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:06:43,358][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:06:43,850][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:06:44,343][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:06:44,839][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:06:45,336][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:06:45,833][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:06:46,330][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:06:46,827][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:06:47,324][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:06:47,822][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:06:48,320][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:06:48,819][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:06:49,317][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:06:49,820][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:06:50,325][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:06:50,828][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:06:51,327][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:06:51,827][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:06:52,325][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:06:52,824][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:06:53,323][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:06:53,819][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:06:54,315][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:06:54,813][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:06:55,311][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:06:55,807][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:06:56,305][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:06:56,803][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:06:57,301][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:06:57,798][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:06:58,294][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:06:58,791][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:06:59,289][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:06:59,786][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:07:00,284][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:07:00,782][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:07:01,284][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:07:01,782][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:07:02,281][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:07:02,778][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:07:03,277][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:07:03,776][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:07:04,274][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:07:04,771][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:07:05,274][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:07:05,773][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:07:06,275][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:07:06,774][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:07:07,275][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:07:07,777][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:07:08,280][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9828 tokens. +[2026-03-26 05:07:09,087][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.10%, ΔTime: 00:00:32 +[2026-03-26 05:07:09,854][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:07:09,856][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:07:09,858][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:07:10,548][__main__][INFO] - Iteration 467 took 49s (28.25% Gen, 70.36% Train). Generation: 14s, Training: 34s. Estimated remaining time: 34h 20m 12s. Estimated total time: 41h 26m 34s. Time estimates for 10 more iterations: 8m 17s, 100 more iterations: 1h 22m 53s, 500 more iterations: 6h 54m 25s. +[2026-03-26 05:07:10,551][__main__][INFO] - Starting iteration 467. +[2026-03-26 05:07:10,960][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 46 and human policies 1. +[2026-03-26 05:07:10,961][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:07:26,929][__main__][INFO] - Number of regex retries in iteration 467: 0 +[2026-03-26 05:07:26,930][__main__][INFO] - agents played in iteration 467 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:07:27,730][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:07:27,753][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:07:27,777][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:07:27,800][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:07:27,801][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:07:27,801][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:07:28,541][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:07:28,994][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:07:29,496][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:07:29,990][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:07:30,487][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:07:30,984][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:07:31,486][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:07:31,983][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:07:32,480][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:07:32,977][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:07:33,474][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:07:33,970][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:07:34,470][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:07:34,966][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:07:35,462][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:07:35,958][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:07:36,455][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:07:36,949][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:07:37,447][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:07:37,942][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:07:38,437][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:07:38,942][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:07:39,439][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:07:39,940][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:07:40,438][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:07:40,934][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:07:41,429][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:07:41,927][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:07:42,422][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:07:42,920][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:07:43,415][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:07:43,912][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:07:44,408][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:07:44,903][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:07:45,402][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:07:45,901][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:07:46,401][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:07:46,900][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:07:47,399][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:07:47,897][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:07:48,392][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:07:48,884][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:07:49,383][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:07:49,885][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:07:50,385][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:07:50,880][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:07:51,384][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:07:51,886][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:07:52,384][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:07:52,881][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:07:53,378][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:07:53,875][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:07:54,373][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:07:54,870][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:07:55,368][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:07:55,866][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:07:56,364][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:07:56,860][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:07:57,354][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:07:57,849][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:07:58,345][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:07:58,840][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:07:59,335][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:07:59,831][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:08:00,331][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9772 tokens. +[2026-03-26 05:08:01,118][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.94%, Current % of VRAM taken: 60.37%, Block Peak % of device VRAM: 62.11%, ΔTime: 00:00:32 +[2026-03-26 05:08:01,857][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:08:01,859][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:08:01,861][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:08:02,636][__main__][INFO] - Iteration 468 took 51s (30.90% Gen, 67.60% Train). Generation: 15s, Training: 34s. Estimated remaining time: 35h 56m 34s. Estimated total time: 43h 3m 48s. Time estimates for 10 more iterations: 8m 36s, 100 more iterations: 1h 26m 7s, 500 more iterations: 7h 10m 38s. +[2026-03-26 05:08:02,638][__main__][INFO] - Starting iteration 468. +[2026-03-26 05:08:03,053][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 46 and human policies 1. +[2026-03-26 05:08:03,054][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:08:19,202][__main__][INFO] - Number of regex retries in iteration 468: 0 +[2026-03-26 05:08:19,203][__main__][INFO] - agents played in iteration 468 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:08:19,989][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:08:20,013][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:08:20,037][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:08:20,061][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:08:20,061][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:08:20,062][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:08:20,803][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:08:21,259][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:08:21,766][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:08:22,263][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:08:22,763][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:08:23,258][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:08:23,753][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:08:24,251][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:08:24,755][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:08:25,254][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:08:25,752][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:08:26,251][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:08:26,747][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:08:27,244][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:08:27,742][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:08:28,243][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:08:28,741][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:08:29,240][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:08:29,739][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:08:30,238][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:08:30,738][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:08:31,239][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:08:31,737][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:08:32,234][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:08:32,731][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:08:33,227][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:08:33,724][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:08:34,225][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:08:34,721][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:08:35,221][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:08:35,718][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:08:36,215][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:08:36,706][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:08:37,202][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:08:37,698][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:08:38,193][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:08:38,686][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:08:39,180][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:08:39,676][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:08:40,170][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:08:40,661][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:08:41,155][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:08:41,652][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:08:42,148][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:08:42,643][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:08:43,138][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:08:43,633][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:08:44,131][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:08:44,628][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:08:45,139][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:08:45,635][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:08:46,140][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:08:46,640][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:08:47,142][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:08:47,638][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:08:48,142][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:08:48,643][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:08:49,142][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:08:49,641][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:08:50,141][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:08:50,642][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:08:51,141][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:08:51,641][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:08:52,140][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:08:52,640][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9901 tokens. +[2026-03-26 05:08:53,430][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.22%, ΔTime: 00:00:32 +[2026-03-26 05:08:54,186][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:08:54,189][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:08:54,191][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:08:54,933][__main__][INFO] - Iteration 469 took 51s (31.13% Gen, 67.44% Train). Generation: 16s, Training: 34s. Estimated remaining time: 36h 5m 55s. Estimated total time: 43h 14m 2s. Time estimates for 10 more iterations: 8m 38s, 100 more iterations: 1h 26m 28s, 500 more iterations: 7h 12m 20s. +[2026-03-26 05:08:54,935][__main__][INFO] - Starting iteration 469. +[2026-03-26 05:08:55,338][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 46 and human policies 1. +[2026-03-26 05:08:55,338][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:09:13,794][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:09:25,191][__main__][INFO] - Number of regex retries in iteration 469: 1 +[2026-03-26 05:09:25,192][__main__][INFO] - agents played in iteration 469 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:09:25,983][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:09:26,007][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:09:26,031][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:09:26,054][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:09:26,054][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:09:26,055][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:09:26,801][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:09:27,252][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:09:27,759][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:09:28,252][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:09:28,760][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:09:29,254][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:09:29,752][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:09:30,244][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:09:30,742][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:09:31,246][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:09:31,744][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:09:32,248][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:09:32,751][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:09:33,249][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:09:33,746][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:09:34,244][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:09:34,743][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:09:35,241][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:09:35,739][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:09:36,236][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:09:36,731][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:09:37,229][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:09:37,725][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:09:38,222][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:09:38,718][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:09:39,214][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:09:39,708][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:09:40,205][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:09:40,702][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:09:41,199][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:09:41,696][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:09:42,191][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:09:42,688][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:09:43,180][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:09:43,672][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:09:44,167][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:09:44,672][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:09:45,171][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:09:45,667][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:09:46,163][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:09:46,663][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:09:47,160][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:09:47,658][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:09:48,162][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:09:48,657][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:09:49,148][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:09:49,644][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:09:50,142][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:09:50,639][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:09:51,148][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:09:51,649][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:09:52,150][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:09:52,651][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:09:53,155][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:09:53,658][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:09:54,159][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:09:54,659][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:09:55,154][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:09:55,651][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:09:56,145][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:09:56,639][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:09:57,135][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:09:57,627][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:09:58,125][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:09:58,620][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9758 tokens. +[2026-03-26 05:09:59,402][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 12.99%, Current % of VRAM taken: 59.41%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 05:10:00,125][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:10:00,128][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:10:00,129][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:10:00,877][__main__][INFO] - Iteration 470 took 1m 5s (45.55% Gen, 53.31% Train). Generation: 29s, Training: 34s. Estimated remaining time: 47h 27m 46s. Estimated total time: 54h 36m 58s. Time estimates for 10 more iterations: 10m 55s, 100 more iterations: 1h 49m 13s, 500 more iterations: 9h 6m 9s. +[2026-03-26 05:10:00,878][__main__][INFO] - Starting iteration 470. +[2026-03-26 05:10:01,286][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 46 and human policies 1. +[2026-03-26 05:10:01,287][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:11:13,347][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:11:13,420][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:11:21,213][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:11:24,150][__main__][INFO] - Number of regex retries in iteration 470: 3 +[2026-03-26 05:11:24,151][__main__][INFO] - agents played in iteration 470 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:11:24,931][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:11:24,954][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:11:24,978][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:11:25,002][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:11:25,002][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:11:25,003][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:11:25,751][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:11:26,215][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:11:26,726][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:11:27,226][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:11:27,725][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:11:28,224][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:11:28,725][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:11:29,224][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:11:29,724][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:11:30,225][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:11:30,728][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:11:31,229][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:11:31,724][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:11:32,233][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:11:32,736][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:11:33,237][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:11:33,738][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:11:34,232][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:11:34,726][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:11:35,228][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:11:35,722][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:11:36,225][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:11:36,726][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:11:37,226][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:11:37,737][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:11:38,241][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:11:38,747][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:11:39,251][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:11:39,757][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:11:40,261][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:11:40,765][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:11:41,269][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:11:41,770][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:11:42,274][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:11:42,777][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:11:43,280][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:11:43,780][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:11:44,283][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:11:44,793][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:11:45,292][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:11:45,800][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:11:46,305][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:11:46,804][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:11:47,302][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:11:47,814][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:11:48,316][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:11:48,821][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:11:49,320][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:11:49,825][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:11:50,327][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:11:50,829][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:11:51,333][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:11:51,840][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:11:52,342][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:11:52,847][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:11:53,348][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:11:53,847][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:11:54,350][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:11:54,854][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:11:55,356][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:11:55,856][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:11:56,355][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:11:56,854][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:11:57,353][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:11:57,851][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9877 tokens. +[2026-03-26 05:11:58,633][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.03%, ΔTime: 00:00:32 +[2026-03-26 05:11:59,498][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:11:59,500][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:11:59,502][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:12:00,906][__main__][INFO] - Iteration 471 took 1m 59s (69.27% Gen, 29.55% Train). Generation: 1m 22s, Training: 35s. Estimated remaining time: 92h 29m 48s. Estimated total time: 99h 41m 0s. Time estimates for 10 more iterations: 19m 56s, 100 more iterations: 3h 19m 22s, 500 more iterations: 16h 36m 50s. +[2026-03-26 05:12:00,908][__main__][INFO] - Starting iteration 471. +[2026-03-26 05:12:01,318][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 47 and human policies 1. +[2026-03-26 05:12:01,319][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:12:17,382][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 0 balls + +This seems to be a strategic decision to avoid the risk of splitting items with Alice, as both sets of per-item values are high for balls and low for hats and books. Since Alice is likely to propose keeping all items, proposing to take none minimizes the risk of sharing and maximizing the certainty of the high value balls. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:12:18,507][__main__][INFO] - Number of regex retries in iteration 471: 1 +[2026-03-26 05:12:18,508][__main__][INFO] - agents played in iteration 471 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:12:19,283][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:12:19,307][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:12:19,331][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:12:19,355][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:12:19,355][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:12:19,356][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:12:20,116][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:12:20,575][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:12:21,075][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:12:21,574][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:12:22,073][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:12:22,568][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:12:23,064][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:12:23,561][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:12:24,058][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:12:24,554][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:12:25,048][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:12:25,550][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:12:26,048][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:12:26,548][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:12:27,048][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:12:27,542][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:12:28,045][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:12:28,547][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:12:29,044][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:12:29,540][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:12:30,040][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:12:30,535][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:12:31,027][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:12:31,521][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:12:32,013][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:12:32,510][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:12:33,008][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:12:33,507][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:12:34,004][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:12:34,503][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:12:35,010][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:12:35,513][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:12:36,017][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:12:36,513][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:12:37,009][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:12:37,507][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:12:38,005][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:12:38,499][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:12:38,994][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:12:39,486][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:12:39,979][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:12:40,477][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:12:40,976][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:12:41,471][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:12:41,969][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:12:42,467][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:12:42,965][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:12:43,461][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:12:43,989][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:12:44,481][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:12:44,972][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:12:45,467][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:12:45,967][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:12:46,465][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:12:46,962][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:12:47,457][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:12:47,952][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:12:48,452][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:12:48,951][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:12:49,447][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:12:49,946][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:12:50,446][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:12:50,955][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:12:51,459][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:12:51,958][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9602 tokens. +[2026-03-26 05:12:52,752][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.97%, Current % of VRAM taken: 60.40%, Block Peak % of device VRAM: 61.93%, ΔTime: 00:00:32 +[2026-03-26 05:12:53,469][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:12:53,472][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:12:53,473][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:12:54,178][__main__][INFO] - Iteration 472 took 52s (32.52% Gen, 66.15% Train). Generation: 17s, Training: 34s. Estimated remaining time: 36h 50m 57s. Estimated total time: 44h 3m 2s. Time estimates for 10 more iterations: 8m 48s, 100 more iterations: 1h 28m 6s, 500 more iterations: 7h 20m 30s. +[2026-03-26 05:12:54,180][__main__][INFO] - Starting iteration 472. +[2026-03-26 05:12:54,586][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 47 and human policies 1. +[2026-03-26 05:12:54,587][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:13:19,709][__main__][INFO] - Number of regex retries in iteration 472: 0 +[2026-03-26 05:13:19,710][__main__][INFO] - agents played in iteration 472 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:13:20,504][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:13:20,528][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:13:20,552][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:13:20,576][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:13:20,577][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:13:20,577][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:13:21,324][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:13:21,777][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:13:22,286][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:13:22,789][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:13:23,288][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:13:23,784][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:13:24,281][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:13:24,788][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:13:25,289][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:13:25,789][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:13:26,290][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:13:26,789][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:13:27,286][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:13:27,784][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:13:28,283][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:13:28,782][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:13:29,290][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:13:29,791][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:13:30,292][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:13:30,791][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:13:31,288][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:13:31,789][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:13:32,298][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:13:32,815][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:13:33,320][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:13:33,819][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:13:34,316][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:13:34,815][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:13:35,323][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:13:35,830][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:13:36,334][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:13:36,837][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:13:37,341][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:13:37,844][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:13:38,348][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:13:38,855][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:13:39,373][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:13:39,877][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:13:40,382][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:13:40,881][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:13:41,383][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:13:41,881][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:13:42,375][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:13:42,875][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:13:43,388][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:13:43,884][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:13:44,378][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:13:44,870][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:13:45,367][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:13:45,865][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:13:46,363][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:13:46,861][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:13:47,369][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:13:47,872][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:13:48,371][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:13:48,870][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:13:49,368][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:13:49,865][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:13:50,364][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:13:50,860][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:13:51,357][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:13:51,854][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:13:52,360][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:13:52,864][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:13:53,362][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9821 tokens. +[2026-03-26 05:13:54,169][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.05%, Current % of VRAM taken: 60.48%, Block Peak % of device VRAM: 62.21%, ΔTime: 00:00:32 +[2026-03-26 05:13:54,928][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:13:54,931][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:13:54,932][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:13:55,689][__main__][INFO] - Iteration 473 took 1m 1s (41.12% Gen, 57.64% Train). Generation: 25s, Training: 35s. Estimated remaining time: 43h 42m 4s. Estimated total time: 50h 55m 11s. Time estimates for 10 more iterations: 10m 11s, 100 more iterations: 1h 41m 50s, 500 more iterations: 8h 29m 11s. +[2026-03-26 05:13:55,691][__main__][INFO] - Starting iteration 473. +[2026-03-26 05:13:56,149][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 47 and human policies 1. +[2026-03-26 05:13:56,150][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:14:11,045][__main__][INFO] - Number of regex retries in iteration 473: 0 +[2026-03-26 05:14:11,045][__main__][INFO] - agents played in iteration 473 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:14:11,814][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:14:11,838][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:14:11,861][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:14:11,885][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:14:11,885][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:14:11,886][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:14:12,628][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:14:13,087][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:14:13,598][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:14:14,105][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:14:14,606][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:14:15,107][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:14:15,609][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:14:16,110][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:14:16,609][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:14:17,106][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:14:17,604][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:14:18,102][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:14:18,600][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:14:19,099][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:14:19,596][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:14:20,095][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:14:20,601][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:14:21,106][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:14:21,606][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:14:22,105][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:14:22,604][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:14:23,103][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:14:23,602][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:14:24,101][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:14:24,599][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:14:25,099][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:14:25,599][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:14:26,098][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:14:26,593][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:14:27,096][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:14:27,594][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:14:28,092][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:14:28,587][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:14:29,080][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:14:29,576][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:14:30,071][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:14:30,562][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:14:31,056][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:14:31,551][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:14:32,045][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:14:32,541][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:14:33,038][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:14:33,532][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:14:34,028][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:14:34,527][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:14:35,031][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:14:35,530][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:14:36,025][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:14:36,524][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:14:37,018][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:14:37,511][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:14:38,008][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:14:38,504][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:14:38,998][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:14:39,501][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:14:40,002][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:14:40,499][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:14:40,994][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:14:41,492][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:14:41,991][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:14:42,490][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:14:42,989][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:14:43,487][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:14:43,985][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:14:44,484][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9740 tokens. +[2026-03-26 05:14:45,283][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.04%, Current % of VRAM taken: 60.46%, Block Peak % of device VRAM: 62.12%, ΔTime: 00:00:32 +[2026-03-26 05:14:46,071][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:14:46,074][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:14:46,075][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:14:47,108][__main__][INFO] - Iteration 474 took 50s (29.23% Gen, 68.74% Train). Generation: 14s, Training: 35s. Estimated remaining time: 35h 14m 0s. Estimated total time: 42h 27m 59s. Time estimates for 10 more iterations: 8m 29s, 100 more iterations: 1h 24m 55s, 500 more iterations: 7h 4m 39s. +[2026-03-26 05:14:47,110][__main__][INFO] - Starting iteration 474. +[2026-03-26 05:14:47,515][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 47 and human policies 1. +[2026-03-26 05:14:47,516][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:14:52,024][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:15:03,729][__main__][INFO] - Number of regex retries in iteration 474: 1 +[2026-03-26 05:15:03,730][__main__][INFO] - agents played in iteration 474 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:15:04,501][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:15:04,525][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:15:04,549][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:15:04,572][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:15:04,572][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:15:04,573][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:15:05,329][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:15:05,791][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:15:06,297][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:15:06,793][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:15:07,294][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:15:07,795][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:15:08,298][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:15:08,799][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:15:09,303][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:15:09,796][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:15:10,288][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:15:10,784][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:15:11,275][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:15:11,767][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:15:12,258][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:15:12,751][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:15:13,250][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:15:13,747][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:15:14,249][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:15:14,743][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:15:15,244][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:15:15,743][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:15:16,243][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:15:16,742][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:15:17,237][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:15:17,739][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:15:18,240][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:15:18,734][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:15:19,226][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:15:19,723][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:15:20,219][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:15:20,710][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:15:21,205][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:15:21,702][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:15:22,198][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:15:22,695][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:15:23,192][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:15:23,692][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:15:24,189][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:15:24,686][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:15:25,183][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:15:25,678][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:15:26,179][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:15:26,673][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:15:27,174][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:15:27,667][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:15:28,163][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:15:28,655][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:15:29,150][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:15:29,651][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:15:30,158][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:15:30,658][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:15:31,168][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:15:31,670][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:15:32,172][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:15:32,674][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:15:33,176][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:15:33,679][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:15:34,185][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:15:34,689][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:15:35,192][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:15:35,695][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:15:36,199][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:15:36,708][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:15:37,215][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9794 tokens. +[2026-03-26 05:15:38,018][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.35%, ΔTime: 00:00:32 +[2026-03-26 05:15:38,768][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:15:38,771][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:15:38,772][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:15:39,544][__main__][INFO] - Iteration 475 took 52s (31.16% Gen, 67.35% Train). Generation: 16s, Training: 35s. Estimated remaining time: 36h 6m 38s. Estimated total time: 43h 21m 29s. Time estimates for 10 more iterations: 8m 40s, 100 more iterations: 1h 26m 42s, 500 more iterations: 7h 13m 34s. +[2026-03-26 05:15:39,546][__main__][INFO] - Starting iteration 475. +[2026-03-26 05:15:39,945][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 47 and human policies 1. +[2026-03-26 05:15:39,945][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:15:56,662][__main__][INFO] - Number of regex retries in iteration 475: 0 +[2026-03-26 05:15:56,663][__main__][INFO] - agents played in iteration 475 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:15:57,453][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:15:57,477][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:15:57,500][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:15:57,523][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:15:57,524][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:15:57,525][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:15:58,296][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:15:58,748][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:15:59,253][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:15:59,764][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:16:00,271][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:16:00,768][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:16:01,267][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:16:01,767][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:16:02,274][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:16:02,777][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:16:03,278][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:16:03,776][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:16:04,275][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:16:04,774][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:16:05,276][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:16:05,782][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:16:06,278][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:16:06,777][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:16:07,279][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:16:07,771][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:16:08,269][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:16:08,766][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:16:09,274][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:16:09,773][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:16:10,273][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:16:10,773][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:16:11,266][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:16:11,759][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:16:12,261][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:16:12,757][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:16:13,251][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:16:13,744][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:16:14,240][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:16:14,735][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:16:15,229][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:16:15,734][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:16:16,236][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:16:16,736][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:16:17,230][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:16:17,728][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:16:18,226][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:16:18,726][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:16:19,219][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:16:19,719][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:16:20,215][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:16:20,708][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:16:21,200][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:16:21,699][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:16:22,199][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:16:22,702][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:16:23,207][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:16:23,709][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:16:24,212][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:16:24,717][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:16:25,217][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:16:25,724][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:16:26,229][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:16:26,729][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:16:27,227][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:16:27,730][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:16:28,227][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:16:28,728][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:16:29,233][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:16:29,729][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:16:30,227][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9678 tokens. +[2026-03-26 05:16:31,030][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.97%, Current % of VRAM taken: 60.40%, Block Peak % of device VRAM: 62.00%, ΔTime: 00:00:32 +[2026-03-26 05:16:31,740][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:16:31,742][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:16:31,744][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:16:32,497][__main__][INFO] - Iteration 476 took 52s (31.81% Gen, 66.75% Train). Generation: 16s, Training: 35s. Estimated remaining time: 36h 31m 56s. Estimated total time: 43h 47m 39s. Time estimates for 10 more iterations: 8m 45s, 100 more iterations: 1h 27m 35s, 500 more iterations: 7h 17m 56s. +[2026-03-26 05:16:32,499][__main__][INFO] - Starting iteration 476. +[2026-03-26 05:16:32,919][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 47 and human policies 1. +[2026-03-26 05:16:32,920][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:16:37,707][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:16:48,737][__main__][INFO] - Number of regex retries in iteration 476: 1 +[2026-03-26 05:16:48,738][__main__][INFO] - agents played in iteration 476 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:16:49,508][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:16:49,532][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:16:49,556][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:16:49,580][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:16:49,581][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:16:49,581][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:16:50,359][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:16:50,812][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:16:51,316][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:16:51,816][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:16:52,323][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:16:52,826][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:16:53,327][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:16:53,827][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:16:54,327][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:16:54,829][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:16:55,339][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:16:55,837][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:16:56,332][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:16:56,827][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:16:57,322][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:16:57,818][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:16:58,327][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:16:58,830][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:16:59,328][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:16:59,822][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:17:00,316][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:17:00,809][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:17:01,304][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:17:01,798][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:17:02,297][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:17:02,796][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:17:03,295][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:17:03,790][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:17:04,313][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:17:04,825][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:17:05,329][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:17:05,829][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:17:06,329][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:17:06,827][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:17:07,325][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:17:07,824][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:17:08,321][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:17:08,826][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:17:09,327][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:17:09,827][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:17:10,330][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:17:10,831][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:17:11,332][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:17:11,829][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:17:12,331][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:17:12,831][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:17:13,332][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:17:13,833][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:17:14,336][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:17:14,850][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:17:15,356][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:17:15,860][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:17:16,363][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:17:16,871][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:17:17,377][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:17:17,882][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:17:18,382][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:17:18,884][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:17:19,393][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:17:19,899][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:17:20,404][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:17:20,908][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:17:21,404][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:17:21,902][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:17:22,400][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9817 tokens. +[2026-03-26 05:17:23,189][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.20%, ΔTime: 00:00:32 +[2026-03-26 05:17:23,942][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:17:23,946][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:17:23,948][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:17:24,696][__main__][INFO] - Iteration 477 took 51s (30.55% Gen, 68.00% Train). Generation: 15s, Training: 35s. Estimated remaining time: 35h 52m 17s. Estimated total time: 43h 8m 53s. Time estimates for 10 more iterations: 8m 37s, 100 more iterations: 1h 26m 17s, 500 more iterations: 7h 11m 28s. +[2026-03-26 05:17:24,699][__main__][INFO] - Starting iteration 477. +[2026-03-26 05:17:25,110][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 47 and human policies 1. +[2026-03-26 05:17:25,111][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:17:30,585][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:17:41,514][__main__][INFO] - Number of regex retries in iteration 477: 1 +[2026-03-26 05:17:41,514][__main__][INFO] - agents played in iteration 477 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:17:42,292][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:17:42,317][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:17:42,345][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:17:42,372][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:17:42,373][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:17:42,373][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:17:43,155][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:17:43,611][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:17:44,111][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:17:44,623][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:17:45,128][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:17:45,626][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:17:46,128][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:17:46,629][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:17:47,126][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:17:47,634][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:17:48,138][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:17:48,640][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:17:49,140][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:17:49,641][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:17:50,142][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:17:50,643][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:17:51,144][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:17:51,643][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:17:52,149][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:17:52,648][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:17:53,147][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:17:53,645][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:17:54,139][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:17:54,631][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:17:55,128][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:17:55,623][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:17:56,121][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:17:56,617][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:17:57,115][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:17:57,612][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:17:58,110][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:17:58,616][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:17:59,118][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:17:59,616][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:18:00,116][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:18:00,616][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:18:01,114][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:18:01,613][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:18:02,113][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:18:02,621][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:18:03,122][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:18:03,622][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:18:04,118][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:18:04,622][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:18:05,124][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:18:05,621][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:18:06,124][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:18:06,637][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:18:07,137][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:18:07,642][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:18:08,148][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:18:08,652][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:18:09,156][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:18:09,666][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:18:10,173][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:18:10,677][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:18:11,182][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:18:11,682][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:18:12,182][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:18:12,685][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:18:13,186][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:18:13,692][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:18:14,194][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:18:14,694][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:18:15,190][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9800 tokens. +[2026-03-26 05:18:16,003][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.92%, Current % of VRAM taken: 60.34%, Block Peak % of device VRAM: 62.25%, ΔTime: 00:00:32 +[2026-03-26 05:18:16,753][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:18:16,756][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:18:16,758][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:18:17,456][__main__][INFO] - Iteration 478 took 52s (31.34% Gen, 67.33% Train). Generation: 16s, Training: 35s. Estimated remaining time: 36h 19m 50s. Estimated total time: 43h 37m 19s. Time estimates for 10 more iterations: 8m 43s, 100 more iterations: 1h 27m 14s, 500 more iterations: 7h 16m 13s. +[2026-03-26 05:18:17,458][__main__][INFO] - Starting iteration 478. +[2026-03-26 05:18:17,864][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 47 and human policies 1. +[2026-03-26 05:18:17,865][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:18:25,520][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 20 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:18:33,896][__main__][INFO] - Number of regex retries in iteration 478: 1 +[2026-03-26 05:18:33,897][__main__][INFO] - agents played in iteration 478 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:18:34,665][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:18:34,689][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:18:34,713][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:18:34,737][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:18:34,737][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:18:34,738][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:18:35,501][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:18:35,962][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:18:36,472][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:18:36,978][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:18:37,480][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:18:37,981][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:18:38,481][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:18:38,977][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:18:39,475][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:18:39,979][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:18:40,480][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:18:40,983][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:18:41,484][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:18:41,984][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:18:42,485][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:18:42,985][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:18:43,484][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:18:43,977][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:18:44,475][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:18:44,984][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:18:45,484][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:18:45,983][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:18:46,477][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:18:46,971][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:18:47,464][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:18:47,962][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:18:48,461][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:18:48,960][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:18:49,458][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:18:49,954][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:18:50,452][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:18:50,953][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:18:51,452][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:18:51,950][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:18:52,448][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:18:52,947][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:18:53,446][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:18:53,945][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:18:54,454][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:18:54,959][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:18:55,459][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:18:55,957][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:18:56,454][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:18:56,951][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:18:57,450][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:18:57,953][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:18:58,456][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:18:58,959][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:18:59,462][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:18:59,967][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:19:00,468][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:19:00,966][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:19:01,474][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:19:01,981][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:19:02,484][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:19:02,987][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:19:03,490][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:19:03,994][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:19:04,497][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:19:05,001][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:19:05,503][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:19:06,005][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:19:06,504][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:19:07,003][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:19:07,504][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9925 tokens. +[2026-03-26 05:19:08,307][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.23%, ΔTime: 00:00:32 +[2026-03-26 05:19:09,058][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:19:09,061][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:19:09,063][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:19:09,799][__main__][INFO] - Iteration 479 took 51s (30.87% Gen, 67.71% Train). Generation: 16s, Training: 35s. Estimated remaining time: 35h 58m 29s. Estimated total time: 43h 16m 50s. Time estimates for 10 more iterations: 8m 39s, 100 more iterations: 1h 26m 33s, 500 more iterations: 7h 12m 48s. +[2026-03-26 05:19:09,802][__main__][INFO] - Starting iteration 479. +[2026-03-26 05:19:10,203][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 47 and human policies 1. +[2026-03-26 05:19:10,204][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:19:25,526][__main__][INFO] - Number of regex retries in iteration 479: 0 +[2026-03-26 05:19:25,527][__main__][INFO] - agents played in iteration 479 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:19:26,285][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:19:26,309][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:19:26,333][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:19:26,356][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:19:26,357][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:19:26,357][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:19:27,105][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:19:27,566][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:19:28,072][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:19:28,572][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:19:29,071][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:19:29,569][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:19:30,067][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:19:30,574][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:19:31,073][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:19:31,569][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:19:32,067][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:19:32,568][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:19:33,073][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:19:33,570][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:19:34,068][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:19:34,564][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:19:35,061][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:19:35,560][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:19:36,071][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:19:36,573][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:19:37,073][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:19:37,571][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:19:38,069][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:19:38,566][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:19:39,071][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:19:39,567][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:19:40,065][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:19:40,563][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:19:41,060][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:19:41,557][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:19:42,052][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:19:42,559][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:19:43,064][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:19:43,566][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:19:44,067][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:19:44,589][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:19:45,088][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:19:45,592][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:19:46,091][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:19:46,597][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:19:47,099][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:19:47,603][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:19:48,105][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:19:48,606][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:19:49,107][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:19:49,608][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:19:50,112][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:19:50,627][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:19:51,149][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:19:51,661][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:19:52,164][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:19:52,671][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:19:53,176][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:19:53,683][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:19:54,188][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:19:54,690][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:19:55,196][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:19:55,700][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:19:56,202][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:19:56,699][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:19:57,202][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:19:57,700][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:19:58,194][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:19:58,690][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:19:59,187][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9881 tokens. +[2026-03-26 05:19:59,977][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.00%, Current % of VRAM taken: 60.43%, Block Peak % of device VRAM: 62.19%, ΔTime: 00:00:32 +[2026-03-26 05:20:00,723][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:20:00,726][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:20:00,727][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:20:01,501][__main__][INFO] - Iteration 480 took 51s (29.87% Gen, 68.62% Train). Generation: 15s, Training: 35s. Estimated remaining time: 35h 25m 41s. Estimated total time: 42h 44m 54s. Time estimates for 10 more iterations: 8m 32s, 100 more iterations: 1h 25m 29s, 500 more iterations: 7h 7m 29s. +[2026-03-26 05:20:01,503][__main__][INFO] - Starting iteration 480. +[2026-03-26 05:20:01,904][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 47 and human policies 1. +[2026-03-26 05:20:01,905][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:20:06,943][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:20:18,415][__main__][INFO] - Number of regex retries in iteration 480: 1 +[2026-03-26 05:20:18,416][__main__][INFO] - agents played in iteration 480 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:20:19,180][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:20:19,203][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:20:19,227][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:20:19,251][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:20:19,251][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:20:19,252][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:20:20,007][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:20:20,461][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:20:20,960][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:20:21,456][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:20:21,953][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:20:22,448][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:20:22,954][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:20:23,458][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:20:23,956][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:20:24,453][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:20:24,952][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:20:25,450][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:20:25,949][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:20:26,447][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:20:26,944][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:20:27,443][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:20:27,946][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:20:28,442][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:20:28,939][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:20:29,442][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:20:29,944][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:20:30,442][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:20:30,943][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:20:31,444][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:20:31,941][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:20:32,437][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:20:32,934][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:20:33,430][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:20:33,926][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:20:34,423][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:20:34,919][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:20:35,416][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:20:35,912][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:20:36,410][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:20:36,916][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:20:37,415][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:20:37,914][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:20:38,412][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:20:38,905][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:20:39,405][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:20:39,913][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:20:40,424][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:20:40,929][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:20:41,430][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:20:41,931][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:20:42,444][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:20:42,949][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:20:43,452][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:20:43,956][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:20:44,457][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:20:44,961][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:20:45,465][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:20:45,969][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:20:46,468][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:20:46,971][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:20:47,476][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:20:47,983][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:20:48,485][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:20:48,985][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:20:49,487][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:20:49,985][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:20:50,486][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:20:50,989][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:20:51,489][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:20:51,988][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9872 tokens. +[2026-03-26 05:20:52,781][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.13%, ΔTime: 00:00:32 +[2026-03-26 05:20:53,533][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:20:53,536][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:20:53,537][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:20:54,993][__main__][INFO] - Iteration 481 took 53s (31.10% Gen, 66.15% Train). Generation: 16s, Training: 35s. Estimated remaining time: 36h 54m 22s. Estimated total time: 44h 14m 28s. Time estimates for 10 more iterations: 8m 50s, 100 more iterations: 1h 28m 28s, 500 more iterations: 7h 22m 24s. +[2026-03-26 05:20:54,995][__main__][INFO] - Starting iteration 481. +[2026-03-26 05:20:55,403][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 48 and human policies 1. +[2026-03-26 05:20:55,404][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:21:00,403][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:21:09,477][__main__][INFO] - Number of regex retries in iteration 481: 1 +[2026-03-26 05:21:09,478][__main__][INFO] - agents played in iteration 481 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:21:10,255][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:21:10,281][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:21:10,307][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:21:10,331][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:21:10,332][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:21:10,333][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:21:11,088][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:21:11,549][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:21:12,057][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:21:12,560][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:21:13,062][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:21:13,571][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:21:14,080][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:21:14,582][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:21:15,085][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:21:15,584][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:21:16,087][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:21:16,596][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:21:17,097][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:21:17,596][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:21:18,096][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:21:18,592][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:21:19,099][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:21:19,607][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:21:20,113][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:21:20,617][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:21:21,116][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:21:21,614][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:21:22,128][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:21:22,636][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:21:23,140][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:21:23,640][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:21:24,139][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:21:24,637][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:21:25,134][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:21:25,631][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:21:26,128][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:21:26,627][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:21:27,124][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:21:27,624][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:21:28,123][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:21:28,621][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:21:29,122][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:21:29,630][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:21:30,130][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:21:30,631][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:21:31,131][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:21:31,635][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:21:32,142][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:21:32,647][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:21:33,161][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:21:33,667][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:21:34,181][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:21:34,691][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:21:35,202][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:21:35,705][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:21:36,208][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:21:36,712][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:21:37,214][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:21:37,717][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:21:38,218][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:21:38,717][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:21:39,216][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:21:39,728][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:21:40,229][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:21:40,729][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:21:41,229][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:21:41,728][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:21:42,227][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:21:42,727][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:21:43,226][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9984 tokens. +[2026-03-26 05:21:44,031][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.21%, ΔTime: 00:00:32 +[2026-03-26 05:21:44,862][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:21:44,865][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:21:44,867][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:21:45,880][__main__][INFO] - Iteration 482 took 50s (27.88% Gen, 70.11% Train). Generation: 14s, Training: 35s. Estimated remaining time: 34h 42m 56s. Estimated total time: 42h 3m 53s. Time estimates for 10 more iterations: 8m 24s, 100 more iterations: 1h 24m 7s, 500 more iterations: 7h 0m 38s. +[2026-03-26 05:21:45,883][__main__][INFO] - Starting iteration 482. +[2026-03-26 05:21:46,283][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 48 and human policies 1. +[2026-03-26 05:21:46,283][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:21:56,196][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 20 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:22:03,836][__main__][INFO] - Number of regex retries in iteration 482: 1 +[2026-03-26 05:22:03,837][__main__][INFO] - agents played in iteration 482 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:22:04,609][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:22:04,633][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:22:04,657][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:22:04,681][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:22:04,681][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:22:04,682][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:22:05,447][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:22:05,895][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:22:06,398][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:22:06,890][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:22:07,383][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:22:07,879][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:22:08,370][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:22:08,876][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:22:09,376][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:22:09,879][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:22:10,375][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:22:10,875][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:22:11,392][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:22:11,887][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:22:12,384][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:22:12,883][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:22:13,387][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:22:13,883][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:22:14,379][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:22:14,873][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:22:15,368][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:22:15,861][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:22:16,354][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:22:16,852][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:22:17,346][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:22:17,845][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:22:18,345][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:22:18,842][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:22:19,339][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:22:19,837][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:22:20,335][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:22:20,837][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:22:21,336][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:22:21,833][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:22:22,332][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:22:22,832][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:22:23,335][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:22:23,835][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:22:24,333][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:22:24,832][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:22:25,334][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:22:25,836][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:22:26,338][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:22:26,839][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:22:27,343][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:22:27,850][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:22:28,351][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:22:28,852][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:22:29,353][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:22:29,849][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:22:30,345][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:22:30,846][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:22:31,350][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:22:31,851][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:22:32,347][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:22:32,846][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:22:33,347][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:22:33,845][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:22:34,341][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:22:34,840][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:22:35,339][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:22:35,839][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:22:36,340][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:22:36,838][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:22:37,335][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9802 tokens. +[2026-03-26 05:22:38,132][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 12.93%, Current % of VRAM taken: 59.36%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 05:22:39,006][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:22:39,008][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:22:39,010][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:22:40,153][__main__][INFO] - Iteration 483 took 53s (32.58% Gen, 65.29% Train). Generation: 17s, Training: 35s. Estimated remaining time: 37h 31m 40s. Estimated total time: 44h 53m 31s. Time estimates for 10 more iterations: 8m 58s, 100 more iterations: 1h 29m 47s, 500 more iterations: 7h 28m 55s. +[2026-03-26 05:22:40,155][__main__][INFO] - Starting iteration 483. +[2026-03-26 05:22:40,562][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 48 and human policies 1. +[2026-03-26 05:22:40,563][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:22:55,034][__main__][INFO] - Number of regex retries in iteration 483: 0 +[2026-03-26 05:22:55,034][__main__][INFO] - agents played in iteration 483 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:22:55,793][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:22:55,817][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:22:55,841][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:22:55,864][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:22:55,865][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:22:55,865][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:22:56,616][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:22:57,077][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:22:57,586][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:22:58,087][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:22:58,589][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:22:59,094][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:22:59,593][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:23:00,093][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:23:00,593][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:23:01,087][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:23:01,584][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:23:02,084][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:23:02,582][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:23:03,074][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:23:03,569][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:23:04,068][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:23:04,571][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:23:05,073][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:23:05,573][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:23:06,070][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:23:06,569][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:23:07,067][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:23:07,562][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:23:08,062][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:23:08,561][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:23:09,062][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:23:09,560][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:23:10,056][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:23:10,551][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:23:11,048][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:23:11,551][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:23:12,053][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:23:12,552][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:23:13,058][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:23:13,563][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:23:14,065][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:23:14,566][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:23:15,080][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:23:15,585][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:23:16,091][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:23:16,595][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:23:17,098][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:23:17,602][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:23:18,107][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:23:18,613][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:23:19,118][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:23:19,618][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:23:20,119][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:23:20,618][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:23:21,118][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:23:21,617][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:23:22,130][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:23:22,629][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:23:23,128][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:23:23,621][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:23:24,120][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:23:24,622][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:23:25,119][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:23:25,615][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:23:26,130][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:23:26,638][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:23:27,136][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:23:27,632][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:23:28,131][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:23:28,627][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9873 tokens. +[2026-03-26 05:23:29,440][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.97%, Current % of VRAM taken: 60.39%, Block Peak % of device VRAM: 62.33%, ΔTime: 00:00:32 +[2026-03-26 05:23:30,234][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:23:30,236][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:23:30,238][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:23:31,540][__main__][INFO] - Iteration 484 took 50s (28.39% Gen, 69.06% Train). Generation: 14s, Training: 35s. Estimated remaining time: 35h 6m 12s. Estimated total time: 42h 28m 55s. Time estimates for 10 more iterations: 8m 29s, 100 more iterations: 1h 24m 57s, 500 more iterations: 7h 4m 49s. +[2026-03-26 05:23:31,542][__main__][INFO] - Starting iteration 484. +[2026-03-26 05:23:31,952][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 48 and human policies 1. +[2026-03-26 05:23:31,953][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:23:48,102][__main__][INFO] - Number of regex retries in iteration 484: 0 +[2026-03-26 05:23:48,103][__main__][INFO] - agents played in iteration 484 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:23:48,875][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:23:48,900][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:23:48,925][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:23:48,950][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:23:48,950][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:23:48,951][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:23:49,709][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:23:50,164][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:23:50,671][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:23:51,177][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:23:51,682][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:23:52,182][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:23:52,682][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:23:53,182][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:23:53,683][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:23:54,177][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:23:54,670][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:23:55,171][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:23:55,668][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:23:56,162][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:23:56,660][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:23:57,154][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:23:57,646][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:23:58,144][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:23:58,643][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:23:59,150][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:23:59,649][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:24:00,147][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:24:00,646][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:24:01,149][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:24:01,647][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:24:02,146][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:24:02,645][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:24:03,150][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:24:03,651][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:24:04,155][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:24:04,654][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:24:05,161][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:24:05,666][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:24:06,167][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:24:06,665][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:24:07,167][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:24:07,678][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:24:08,187][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:24:08,692][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:24:09,202][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:24:09,708][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:24:10,218][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:24:10,724][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:24:11,232][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:24:11,732][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:24:12,235][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:24:12,743][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:24:13,250][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:24:13,754][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:24:14,250][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:24:14,747][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:24:15,245][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:24:15,742][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:24:16,242][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:24:16,738][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:24:17,232][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:24:17,734][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:24:18,239][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:24:18,745][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:24:19,249][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:24:19,753][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:24:20,257][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:24:20,758][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:24:21,260][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:24:21,765][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9850 tokens. +[2026-03-26 05:24:22,561][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.23%, Current % of VRAM taken: 60.65%, Block Peak % of device VRAM: 62.15%, ΔTime: 00:00:32 +[2026-03-26 05:24:23,668][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:24:23,671][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:24:23,672][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:24:24,997][__main__][INFO] - Iteration 485 took 53s (30.45% Gen, 67.05% Train). Generation: 16s, Training: 35s. Estimated remaining time: 36h 48m 45s. Estimated total time: 44h 12m 21s. Time estimates for 10 more iterations: 8m 50s, 100 more iterations: 1h 28m 24s, 500 more iterations: 7h 22m 3s. +[2026-03-26 05:24:24,999][__main__][INFO] - Starting iteration 485. +[2026-03-26 05:24:25,403][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 48 and human policies 1. +[2026-03-26 05:24:25,403][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:24:29,568][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:24:36,705][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:24:40,084][__main__][INFO] - Number of regex retries in iteration 485: 2 +[2026-03-26 05:24:40,084][__main__][INFO] - agents played in iteration 485 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:24:40,848][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:24:40,872][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:24:40,895][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:24:40,919][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:24:40,920][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:24:40,921][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:24:41,674][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:24:42,134][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:24:42,645][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:24:43,147][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:24:43,648][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:24:44,147][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:24:44,652][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:24:45,154][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:24:45,663][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:24:46,163][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:24:46,664][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:24:47,163][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:24:47,663][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:24:48,163][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:24:48,665][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:24:49,164][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:24:49,663][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:24:50,163][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:24:50,662][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:24:51,160][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:24:51,657][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:24:52,155][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:24:52,653][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:24:53,163][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:24:53,660][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:24:54,161][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:24:54,661][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:24:55,165][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:24:55,669][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:24:56,171][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:24:56,677][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:24:57,186][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:24:57,690][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:24:58,190][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:24:58,688][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:24:59,191][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:24:59,694][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:25:00,194][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:25:00,693][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:25:01,189][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:25:01,692][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:25:02,195][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:25:02,699][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:25:03,210][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:25:03,721][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:25:04,224][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:25:04,730][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:25:05,229][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:25:05,732][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:25:06,231][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:25:06,731][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:25:07,231][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:25:07,739][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:25:08,245][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:25:08,749][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:25:09,249][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:25:09,747][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:25:10,249][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:25:10,751][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:25:11,253][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:25:11,755][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:25:12,262][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:25:12,766][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:25:13,278][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:25:13,781][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9960 tokens. +[2026-03-26 05:25:14,686][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.17%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.36%, ΔTime: 00:00:33 +[2026-03-26 05:25:15,497][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:25:15,499][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:25:15,501][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:25:17,175][__main__][INFO] - Iteration 486 took 51s (28.36% Gen, 68.41% Train). Generation: 14s, Training: 35s. Estimated remaining time: 35h 44m 10s. Estimated total time: 43h 8m 39s. Time estimates for 10 more iterations: 8m 37s, 100 more iterations: 1h 26m 17s, 500 more iterations: 7h 11m 26s. +[2026-03-26 05:25:17,177][__main__][INFO] - Starting iteration 486. +[2026-03-26 05:25:17,582][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 48 and human policies 1. +[2026-03-26 05:25:17,583][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:25:32,842][__main__][INFO] - Number of regex retries in iteration 486: 0 +[2026-03-26 05:25:32,843][__main__][INFO] - agents played in iteration 486 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:25:33,606][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:25:33,630][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:25:33,653][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:25:33,677][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:25:33,678][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:25:33,678][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:25:34,429][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:25:34,884][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:25:35,388][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:25:35,888][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:25:36,389][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:25:36,892][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:25:37,401][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:25:37,906][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:25:38,408][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:25:38,911][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:25:39,405][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:25:39,906][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:25:40,399][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:25:40,891][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:25:41,384][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:25:41,877][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:25:42,374][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:25:42,877][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:25:43,373][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:25:43,875][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:25:44,376][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:25:44,874][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:25:45,368][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:25:45,862][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:25:46,359][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:25:46,856][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:25:47,353][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:25:47,851][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:25:48,356][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:25:48,866][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:25:49,375][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:25:49,878][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:25:50,382][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:25:50,886][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:25:51,389][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:25:51,893][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:25:52,396][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:25:52,899][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:25:53,402][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:25:53,906][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:25:54,406][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:25:54,906][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:25:55,407][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:25:55,913][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:25:56,412][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:25:56,912][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:25:57,412][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:25:57,911][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:25:58,408][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:25:58,905][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:25:59,402][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:25:59,901][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:26:00,394][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:26:00,888][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:26:01,386][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:26:01,888][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:26:02,384][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:26:02,884][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:26:03,385][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:26:03,891][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:26:04,393][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:26:04,894][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:26:05,398][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:26:05,902][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:26:06,406][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9803 tokens. +[2026-03-26 05:26:07,204][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.39%, ΔTime: 00:00:32 +[2026-03-26 05:26:07,960][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:26:07,962][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:26:07,964][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:26:08,741][__main__][INFO] - Iteration 487 took 51s (29.83% Gen, 68.65% Train). Generation: 15s, Training: 35s. Estimated remaining time: 35h 12m 37s. Estimated total time: 42h 37m 57s. Time estimates for 10 more iterations: 8m 31s, 100 more iterations: 1h 25m 15s, 500 more iterations: 7h 6m 19s. +[2026-03-26 05:26:08,743][__main__][INFO] - Starting iteration 487. +[2026-03-26 05:26:09,146][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 48 and human policies 1. +[2026-03-26 05:26:09,147][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:26:26,631][__main__][INFO] - Number of regex retries in iteration 487: 0 +[2026-03-26 05:26:26,632][__main__][INFO] - agents played in iteration 487 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:26:27,404][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:26:27,428][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:26:27,451][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:26:27,475][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:26:27,475][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:26:27,476][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:26:28,240][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:26:28,721][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:26:29,226][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:26:29,721][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:26:30,214][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:26:30,708][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:26:31,200][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:26:31,703][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:26:32,202][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:26:32,702][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:26:33,204][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:26:33,705][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:26:34,208][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:26:34,705][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:26:35,205][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:26:35,705][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:26:36,206][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:26:36,705][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:26:37,203][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:26:37,696][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:26:38,194][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:26:38,694][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:26:39,188][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:26:39,690][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:26:40,192][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:26:40,697][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:26:41,204][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:26:41,705][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:26:42,215][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:26:42,721][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:26:43,232][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:26:43,737][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:26:44,240][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:26:44,744][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:26:45,243][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:26:45,746][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:26:46,246][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:26:46,746][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:26:47,246][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:26:47,747][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:26:48,253][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:26:48,751][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:26:49,248][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:26:49,747][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:26:50,248][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:26:50,746][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:26:51,242][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:26:51,740][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:26:52,240][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:26:52,734][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:26:53,230][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:26:53,723][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:26:54,218][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:26:54,713][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:26:55,212][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:26:55,707][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:26:56,203][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:26:56,706][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:26:57,209][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:26:57,712][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:26:58,213][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:26:58,720][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:26:59,221][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:26:59,728][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:27:00,229][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9778 tokens. +[2026-03-26 05:27:01,037][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.10%, ΔTime: 00:00:32 +[2026-03-26 05:27:01,953][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:27:01,955][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:27:01,957][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:27:02,919][__main__][INFO] - Iteration 488 took 53s (32.52% Gen, 65.69% Train). Generation: 17s, Training: 35s. Estimated remaining time: 37h 22m 27s. Estimated total time: 44h 48m 41s. Time estimates for 10 more iterations: 8m 57s, 100 more iterations: 1h 29m 37s, 500 more iterations: 7h 28m 6s. +[2026-03-26 05:27:02,922][__main__][INFO] - Starting iteration 488. +[2026-03-26 05:27:03,324][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 48 and human policies 1. +[2026-03-26 05:27:03,325][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:27:19,173][__main__][INFO] - Number of regex retries in iteration 488: 0 +[2026-03-26 05:27:19,174][__main__][INFO] - agents played in iteration 488 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:27:19,959][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:27:19,983][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:27:20,006][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:27:20,030][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:27:20,031][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:27:20,031][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:27:20,795][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:27:21,255][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:27:21,764][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:27:22,267][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:27:22,769][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:27:23,273][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:27:23,768][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:27:24,269][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:27:24,774][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:27:25,280][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:27:25,779][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:27:26,279][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:27:26,779][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:27:27,277][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:27:27,787][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:27:28,288][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:27:28,795][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:27:29,289][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:27:29,789][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:27:30,292][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:27:30,800][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:27:31,311][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:27:31,818][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:27:32,327][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:27:32,832][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:27:33,337][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:27:33,843][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:27:34,350][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:27:34,853][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:27:35,362][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:27:35,864][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:27:36,367][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:27:36,868][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:27:37,366][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:27:37,868][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:27:38,367][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:27:38,873][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:27:39,376][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:27:39,875][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:27:40,374][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:27:40,872][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:27:41,371][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:27:41,869][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:27:42,368][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:27:42,872][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:27:43,373][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:27:43,872][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:27:44,372][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:27:44,870][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:27:45,364][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:27:45,857][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:27:46,355][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:27:46,848][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:27:47,354][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:27:47,846][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:27:48,342][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:27:48,834][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:27:49,332][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:27:49,831][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:27:50,335][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:27:50,837][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:27:51,336][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:27:51,839][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:27:52,345][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:27:52,845][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9932 tokens. +[2026-03-26 05:27:53,658][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.19%, ΔTime: 00:00:32 +[2026-03-26 05:27:54,609][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:27:54,611][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:27:54,613][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:27:55,767][__main__][INFO] - Iteration 489 took 52s (30.22% Gen, 67.58% Train). Generation: 15s, Training: 35s. Estimated remaining time: 36h 15m 1s. Estimated total time: 43h 42m 9s. Time estimates for 10 more iterations: 8m 44s, 100 more iterations: 1h 27m 24s, 500 more iterations: 7h 17m 1s. +[2026-03-26 05:27:55,769][__main__][INFO] - Starting iteration 489. +[2026-03-26 05:27:56,170][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 48 and human policies 1. +[2026-03-26 05:27:56,171][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:28:10,518][__main__][INFO] - Number of regex retries in iteration 489: 0 +[2026-03-26 05:28:10,519][__main__][INFO] - agents played in iteration 489 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:28:11,282][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:28:11,307][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:28:11,332][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:28:11,356][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:28:11,357][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:28:11,357][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:28:12,114][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:28:12,572][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:28:13,074][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:28:13,571][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:28:14,069][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:28:14,577][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:28:15,081][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:28:15,582][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:28:16,081][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:28:16,579][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:28:17,083][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:28:17,585][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:28:18,085][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:28:18,583][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:28:19,081][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:28:19,581][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:28:20,087][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:28:20,592][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:28:21,088][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:28:21,581][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:28:22,078][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:28:22,583][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:28:23,077][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:28:23,571][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:28:24,069][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:28:24,570][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:28:25,070][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:28:25,571][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:28:26,072][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:28:26,571][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:28:27,076][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:28:27,588][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:28:28,090][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:28:28,590][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:28:29,089][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:28:29,589][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:28:30,088][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:28:30,594][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:28:31,098][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:28:31,598][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:28:32,103][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:28:32,609][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:28:33,116][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:28:33,621][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:28:34,131][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:28:34,636][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:28:35,133][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:28:35,633][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:28:36,130][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:28:36,630][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:28:37,129][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:28:37,629][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:28:38,137][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:28:38,641][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:28:39,145][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:28:39,643][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:28:40,140][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:28:40,634][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:28:41,134][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:28:41,629][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:28:42,123][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:28:42,625][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:28:43,125][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:28:43,624][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:28:44,129][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9846 tokens. +[2026-03-26 05:28:44,934][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.03%, Current % of VRAM taken: 60.46%, Block Peak % of device VRAM: 62.28%, ΔTime: 00:00:32 +[2026-03-26 05:28:45,743][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:28:45,745][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:28:45,747][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:28:46,474][__main__][INFO] - Iteration 490 took 50s (28.52% Gen, 70.03% Train). Generation: 14s, Training: 35s. Estimated remaining time: 34h 27m 17s. Estimated total time: 41h 55m 15s. Time estimates for 10 more iterations: 8m 23s, 100 more iterations: 1h 23m 50s, 500 more iterations: 6h 59m 12s. +[2026-03-26 05:28:46,477][__main__][INFO] - Starting iteration 490. +[2026-03-26 05:28:46,884][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 48 and human policies 1. +[2026-03-26 05:28:46,885][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:28:50,654][mllm.models.large_language_model_local][WARNING] - Response Proposal: 5 hats, 5 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:28:53,433][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 20 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:29:01,061][__main__][INFO] - Number of regex retries in iteration 490: 2 +[2026-03-26 05:29:01,062][__main__][INFO] - agents played in iteration 490 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:29:01,825][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:29:01,851][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:29:01,876][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:29:01,901][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:29:01,902][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:29:01,902][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:29:02,667][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:29:03,123][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:29:03,631][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:29:04,130][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:29:04,630][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:29:05,129][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:29:05,628][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:29:06,127][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:29:06,626][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:29:07,126][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:29:07,629][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:29:08,135][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:29:08,635][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:29:09,135][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:29:09,633][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:29:10,132][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:29:10,631][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:29:11,125][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:29:11,623][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:29:12,121][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:29:12,613][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:29:13,110][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:29:13,602][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:29:14,104][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:29:14,603][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:29:15,103][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:29:15,599][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:29:16,097][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:29:16,595][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:29:17,090][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:29:17,584][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:29:18,076][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:29:18,578][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:29:19,076][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:29:19,576][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:29:20,077][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:29:20,589][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:29:21,099][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:29:21,608][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:29:22,110][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:29:22,615][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:29:23,119][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:29:23,624][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:29:24,130][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:29:24,634][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:29:25,138][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:29:25,644][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:29:26,147][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:29:26,648][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:29:27,142][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:29:27,639][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:29:28,139][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:29:28,633][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:29:29,129][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:29:29,622][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:29:30,115][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:29:30,608][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:29:31,108][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:29:31,608][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:29:32,109][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:29:32,612][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:29:33,111][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:29:33,611][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:29:34,110][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:29:34,610][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9830 tokens. +[2026-03-26 05:29:35,407][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 62.14%, ΔTime: 00:00:32 +[2026-03-26 05:29:36,162][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:29:36,164][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:29:36,166][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:29:37,609][__main__][INFO] - Iteration 491 took 50s (27.95% Gen, 69.20% Train). Generation: 14s, Training: 35s. Estimated remaining time: 34h 47m 27s. Estimated total time: 42h 16m 16s. Time estimates for 10 more iterations: 8m 27s, 100 more iterations: 1h 24m 32s, 500 more iterations: 7h 2m 42s. +[2026-03-26 05:29:37,611][__main__][INFO] - Starting iteration 491. +[2026-03-26 05:29:38,029][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 49 and human policies 1. +[2026-03-26 05:29:38,030][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:29:55,077][__main__][INFO] - Number of regex retries in iteration 491: 0 +[2026-03-26 05:29:55,078][__main__][INFO] - agents played in iteration 491 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:29:55,856][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:29:55,881][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:29:55,905][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:29:55,930][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:29:55,930][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:29:55,931][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:29:56,666][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:29:57,122][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:29:57,621][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:29:58,129][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:29:58,627][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:29:59,126][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:29:59,635][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:30:00,134][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:30:00,628][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:30:01,132][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:30:01,634][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:30:02,131][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:30:02,629][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:30:03,126][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:30:03,623][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:30:04,120][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:30:04,622][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:30:05,125][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:30:05,622][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:30:06,120][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:30:06,622][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:30:07,119][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:30:07,622][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:30:08,119][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:30:08,626][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:30:09,128][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:30:09,629][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:30:10,134][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:30:10,635][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:30:11,141][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:30:11,646][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:30:12,152][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:30:12,658][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:30:13,167][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:30:13,672][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:30:14,176][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:30:14,679][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:30:15,183][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:30:15,687][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:30:16,191][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:30:16,694][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:30:17,200][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:30:17,700][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:30:18,199][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:30:18,699][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:30:19,198][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:30:19,696][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:30:20,203][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:30:20,697][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:30:21,196][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:30:21,694][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:30:22,190][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:30:22,694][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:30:23,201][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:30:23,702][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:30:24,203][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:30:24,707][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:30:25,207][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:30:25,708][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:30:26,208][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:30:26,714][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:30:27,216][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:30:27,715][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:30:28,221][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:30:28,729][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9905 tokens. +[2026-03-26 05:30:29,553][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.14%, ΔTime: 00:00:32 +[2026-03-26 05:30:30,297][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:30:30,299][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:30:30,301][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:30:31,014][__main__][INFO] - Iteration 492 took 52s (32.18% Gen, 66.48% Train). Generation: 17s, Training: 35s. Estimated remaining time: 36h 39m 33s. Estimated total time: 44h 9m 16s. Time estimates for 10 more iterations: 8m 49s, 100 more iterations: 1h 28m 18s, 500 more iterations: 7h 21m 32s. +[2026-03-26 05:30:31,016][__main__][INFO] - Starting iteration 492. +[2026-03-26 05:30:31,416][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 49 and human policies 1. +[2026-03-26 05:30:31,417][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:30:47,850][__main__][INFO] - Number of regex retries in iteration 492: 0 +[2026-03-26 05:30:47,851][__main__][INFO] - agents played in iteration 492 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:30:48,618][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:30:48,642][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:30:48,665][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:30:48,689][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:30:48,690][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:30:48,691][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:30:49,454][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:30:49,911][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:30:50,417][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:30:50,919][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:30:51,434][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:30:51,937][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:30:52,439][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:30:52,941][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:30:53,444][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:30:53,939][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:30:54,432][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:30:54,926][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:30:55,426][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:30:55,928][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:30:56,422][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:30:56,915][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:30:57,421][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:30:57,925][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:30:58,424][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:30:58,923][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:30:59,419][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:30:59,917][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:31:00,416][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:31:00,915][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:31:01,414][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:31:01,913][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:31:02,429][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:31:02,936][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:31:03,439][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:31:03,939][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:31:04,445][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:31:04,946][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:31:05,448][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:31:05,956][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:31:06,452][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:31:06,950][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:31:07,452][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:31:07,951][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:31:08,446][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:31:08,942][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:31:09,447][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:31:09,960][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:31:10,466][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:31:10,966][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:31:11,469][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:31:11,967][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:31:12,465][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:31:12,967][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:31:13,466][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:31:13,971][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:31:14,476][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:31:14,975][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:31:15,475][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:31:15,976][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:31:16,479][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:31:16,986][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:31:17,487][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:31:17,992][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:31:18,493][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:31:18,998][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:31:19,494][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:31:19,993][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:31:20,496][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:31:20,995][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:31:21,500][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9971 tokens. +[2026-03-26 05:31:22,318][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.29%, Current % of VRAM taken: 60.71%, Block Peak % of device VRAM: 62.43%, ΔTime: 00:00:32 +[2026-03-26 05:31:23,043][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:31:23,045][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:31:23,047][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:31:23,772][__main__][INFO] - Iteration 493 took 52s (31.39% Gen, 67.22% Train). Generation: 16s, Training: 35s. Estimated remaining time: 36h 7m 14s. Estimated total time: 43h 37m 49s. Time estimates for 10 more iterations: 8m 43s, 100 more iterations: 1h 27m 15s, 500 more iterations: 7h 16m 18s. +[2026-03-26 05:31:23,775][__main__][INFO] - Starting iteration 493. +[2026-03-26 05:31:24,187][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 49 and human policies 1. +[2026-03-26 05:31:24,188][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:31:40,377][__main__][INFO] - Number of regex retries in iteration 493: 0 +[2026-03-26 05:31:40,378][__main__][INFO] - agents played in iteration 493 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:31:41,158][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.60%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:31:41,182][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.60%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:31:41,206][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.60%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:31:41,230][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.60%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:31:41,231][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:31:41,231][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:31:41,996][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:31:42,449][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:31:42,954][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:31:43,458][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:31:43,960][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:31:44,460][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:31:44,962][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:31:45,463][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:31:45,970][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:31:46,471][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:31:46,971][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:31:47,470][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:31:47,966][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:31:48,467][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:31:48,962][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:31:49,462][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:31:49,959][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:31:50,471][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:31:50,968][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:31:51,471][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:31:51,967][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:31:52,461][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:31:52,959][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:31:53,462][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:31:53,967][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:31:54,468][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:31:54,970][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:31:55,471][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:31:55,971][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:31:56,472][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:31:56,999][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:31:57,508][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:31:58,003][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:31:58,500][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:31:58,998][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:31:59,497][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:31:59,996][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:32:00,495][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:32:00,998][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:32:01,502][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:32:02,000][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:32:02,502][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:32:03,003][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:32:03,507][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:32:04,011][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:32:04,524][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:32:05,037][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:32:05,560][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:32:06,067][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:32:06,572][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:32:07,077][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:32:07,583][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:32:08,087][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:32:08,591][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:32:09,097][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:32:09,600][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:32:10,110][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:32:10,613][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:32:11,113][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:32:11,609][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:32:12,109][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:32:12,612][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:32:13,114][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:32:13,616][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:32:14,120][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9879 tokens. +[2026-03-26 05:32:14,932][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.22%, Current % of VRAM taken: 60.65%, Block Peak % of device VRAM: 62.22%, ΔTime: 00:00:32 +[2026-03-26 05:32:15,696][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:32:15,698][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:32:15,700][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:32:16,413][__main__][INFO] - Iteration 494 took 52s (31.00% Gen, 67.63% Train). Generation: 16s, Training: 35s. Estimated remaining time: 35h 59m 51s. Estimated total time: 43h 31m 19s. Time estimates for 10 more iterations: 8m 42s, 100 more iterations: 1h 27m 2s, 500 more iterations: 7h 15m 13s. +[2026-03-26 05:32:16,415][__main__][INFO] - Starting iteration 494. +[2026-03-26 05:32:16,815][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 49 and human policies 1. +[2026-03-26 05:32:16,816][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:32:25,560][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:32:33,768][__main__][INFO] - Number of regex retries in iteration 494: 1 +[2026-03-26 05:32:33,769][__main__][INFO] - agents played in iteration 494 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:32:34,560][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:32:34,584][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:32:34,609][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:32:34,633][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:32:34,633][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:32:34,634][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:32:35,400][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:32:35,856][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:32:36,361][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:32:36,863][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:32:37,361][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:32:37,865][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:32:38,362][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:32:38,860][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:32:39,362][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:32:39,871][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:32:40,374][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:32:40,882][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:32:41,380][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:32:41,882][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:32:42,383][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:32:42,880][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:32:43,383][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:32:43,902][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:32:44,404][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:32:44,913][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:32:45,421][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:32:45,925][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:32:46,428][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:32:46,929][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:32:47,434][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:32:47,933][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:32:48,434][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:32:48,931][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:32:49,430][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:32:49,928][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:32:50,427][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:32:50,924][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:32:51,422][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:32:51,922][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:32:52,422][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:32:52,931][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:32:53,437][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:32:53,940][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:32:54,444][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:32:54,945][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:32:55,446][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:32:55,950][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:32:56,458][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:32:56,964][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:32:57,479][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:32:57,982][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:32:58,482][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:32:58,987][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:32:59,493][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:32:59,999][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:33:00,498][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:33:01,004][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:33:01,507][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:33:02,014][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:33:02,517][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:33:03,020][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:33:03,523][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:33:04,022][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:33:04,530][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:33:05,033][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:33:05,536][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:33:06,037][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:33:06,541][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:33:07,042][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:33:07,541][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10143 tokens. +[2026-03-26 05:33:08,349][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.16%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.41%, ΔTime: 00:00:32 +[2026-03-26 05:33:09,154][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:33:09,156][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:33:09,159][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:33:09,881][__main__][INFO] - Iteration 495 took 53s (31.95% Gen, 66.69% Train). Generation: 16s, Training: 35s. Estimated remaining time: 36h 40m 59s. Estimated total time: 44h 13m 20s. Time estimates for 10 more iterations: 8m 50s, 100 more iterations: 1h 28m 26s, 500 more iterations: 7h 22m 13s. +[2026-03-26 05:33:09,884][__main__][INFO] - Starting iteration 495. +[2026-03-26 05:33:10,309][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 49 and human policies 1. +[2026-03-26 05:33:10,310][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:33:14,358][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:33:25,408][__main__][INFO] - Number of regex retries in iteration 495: 1 +[2026-03-26 05:33:25,409][__main__][INFO] - agents played in iteration 495 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:33:26,198][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:33:26,221][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:33:26,245][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:33:26,268][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:33:26,269][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:33:26,269][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:33:27,039][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:33:27,501][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:33:28,007][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:33:28,514][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:33:29,014][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:33:29,513][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:33:30,024][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:33:30,524][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:33:31,021][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:33:31,520][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:33:32,018][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:33:32,528][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:33:33,029][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:33:33,529][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:33:34,028][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:33:34,528][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:33:35,026][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:33:35,531][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:33:36,036][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:33:36,535][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:33:37,033][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:33:37,531][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:33:38,029][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:33:38,528][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:33:39,038][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:33:39,534][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:33:40,029][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:33:40,523][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:33:41,022][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:33:41,522][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:33:42,033][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:33:42,529][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:33:43,029][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:33:43,528][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:33:44,026][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:33:44,526][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:33:45,025][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:33:45,524][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:33:46,022][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:33:46,521][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:33:47,017][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:33:47,512][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:33:48,005][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:33:48,503][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:33:49,005][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:33:49,511][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:33:50,015][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:33:50,510][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:33:51,023][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:33:51,528][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:33:52,034][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:33:52,539][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:33:53,044][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:33:53,550][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:33:54,054][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:33:54,564][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:33:55,072][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:33:55,574][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:33:56,073][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:33:56,572][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:33:57,069][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:33:57,568][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:33:58,068][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:33:58,572][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:33:59,082][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9896 tokens. +[2026-03-26 05:33:59,925][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.22%, Current % of VRAM taken: 60.64%, Block Peak % of device VRAM: 62.14%, ΔTime: 00:00:32 +[2026-03-26 05:34:00,689][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:34:00,692][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:34:00,694][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:34:01,578][__main__][INFO] - Iteration 496 took 51s (29.45% Gen, 68.82% Train). Generation: 15s, Training: 35s. Estimated remaining time: 35h 10m 15s. Estimated total time: 42h 43m 28s. Time estimates for 10 more iterations: 8m 32s, 100 more iterations: 1h 25m 26s, 500 more iterations: 7h 7m 14s. +[2026-03-26 05:34:01,581][__main__][INFO] - Starting iteration 496. +[2026-03-26 05:34:01,982][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 49 and human policies 1. +[2026-03-26 05:34:01,983][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:34:18,110][__main__][INFO] - Number of regex retries in iteration 496: 0 +[2026-03-26 05:34:18,111][__main__][INFO] - agents played in iteration 496 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:34:18,906][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:34:18,932][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:34:18,957][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:34:18,984][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:34:18,984][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:34:18,985][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:34:19,762][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:34:20,219][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:34:20,725][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:34:21,231][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:34:21,730][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:34:22,230][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:34:22,727][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:34:23,225][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:34:23,722][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:34:24,222][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:34:24,729][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:34:25,237][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:34:25,739][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:34:26,239][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:34:26,741][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:34:27,241][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:34:27,741][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:34:28,246][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:34:28,747][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:34:29,248][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:34:29,748][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:34:30,245][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:34:30,744][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:34:31,242][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:34:31,742][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:34:32,242][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:34:32,740][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:34:33,237][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:34:33,734][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:34:34,240][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:34:34,747][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:34:35,246][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:34:35,745][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:34:36,245][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:34:36,746][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:34:37,246][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:34:37,745][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:34:38,244][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:34:38,745][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:34:39,243][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:34:39,743][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:34:40,243][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:34:40,742][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:34:41,247][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:34:41,756][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:34:42,262][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:34:42,797][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:34:43,301][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:34:43,805][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:34:44,309][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:34:44,811][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:34:45,314][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:34:45,817][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:34:46,319][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:34:46,821][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:34:47,324][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:34:47,825][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:34:48,325][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:34:48,832][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:34:49,341][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:34:49,845][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:34:50,348][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:34:50,849][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:34:51,350][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:34:51,852][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10018 tokens. +[2026-03-26 05:34:52,661][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.15%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.19%, ΔTime: 00:00:32 +[2026-03-26 05:34:53,414][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:34:53,416][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:34:53,418][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:34:54,153][__main__][INFO] - Iteration 497 took 52s (30.91% Gen, 67.67% Train). Generation: 16s, Training: 35s. Estimated remaining time: 35h 54m 30s. Estimated total time: 43h 28m 35s. Time estimates for 10 more iterations: 8m 41s, 100 more iterations: 1h 26m 57s, 500 more iterations: 7h 14m 45s. +[2026-03-26 05:34:54,156][__main__][INFO] - Starting iteration 497. +[2026-03-26 05:34:54,556][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 49 and human policies 1. +[2026-03-26 05:34:54,557][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:35:12,628][mllm.models.large_language_model_local][WARNING] - Response Proposal: x hats, y books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:35:23,493][__main__][INFO] - Number of regex retries in iteration 497: 1 +[2026-03-26 05:35:23,494][__main__][INFO] - agents played in iteration 497 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:35:24,268][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:35:24,291][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:35:24,315][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:35:24,339][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:35:24,339][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:35:24,340][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:35:25,108][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:35:25,561][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:35:26,063][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:35:26,561][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:35:27,064][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:35:27,562][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:35:28,063][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:35:28,569][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:35:29,069][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:35:29,572][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:35:30,069][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:35:30,568][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:35:31,065][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:35:31,566][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:35:32,068][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:35:32,575][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:35:33,074][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:35:33,574][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:35:34,077][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:35:34,583][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:35:35,088][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:35:35,596][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:35:36,100][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:35:36,604][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:35:37,106][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:35:37,608][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:35:38,111][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:35:38,614][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:35:39,115][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:35:39,619][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:35:40,127][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:35:40,629][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:35:41,136][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:35:41,642][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:35:42,147][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:35:42,653][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:35:43,158][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:35:43,655][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:35:44,154][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:35:44,655][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:35:45,153][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:35:45,653][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:35:46,154][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:35:46,655][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:35:47,154][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:35:47,660][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:35:48,164][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:35:48,667][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:35:49,171][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:35:49,670][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:35:50,173][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:35:50,675][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:35:51,175][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:35:51,679][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:35:52,183][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:35:52,692][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:35:53,197][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:35:53,696][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:35:54,195][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:35:54,696][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:35:55,195][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:35:55,697][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:35:56,210][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:35:56,714][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:35:57,213][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9933 tokens. +[2026-03-26 05:35:58,024][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.87%, Current % of VRAM taken: 60.30%, Block Peak % of device VRAM: 62.10%, ΔTime: 00:00:32 +[2026-03-26 05:35:58,778][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:35:58,780][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:35:58,781][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:35:59,515][__main__][INFO] - Iteration 498 took 1m 4s (44.55% Gen, 54.32% Train). Generation: 28s, Training: 35s. Estimated remaining time: 46h 32m 47s. Estimated total time: 54h 7m 58s. Time estimates for 10 more iterations: 10m 49s, 100 more iterations: 1h 48m 15s, 500 more iterations: 9h 1m 19s. +[2026-03-26 05:35:59,518][__main__][INFO] - Starting iteration 498. +[2026-03-26 05:35:59,932][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 49 and human policies 1. +[2026-03-26 05:35:59,933][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:36:07,659][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:36:13,836][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 20 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:36:17,726][__main__][INFO] - Number of regex retries in iteration 498: 2 +[2026-03-26 05:36:17,727][__main__][INFO] - agents played in iteration 498 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:36:18,501][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:36:18,524][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:36:18,548][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:36:18,572][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:36:18,572][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:36:18,573][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:36:19,342][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:36:19,800][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:36:20,304][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:36:20,802][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:36:21,295][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:36:21,793][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:36:22,291][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:36:22,789][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:36:23,287][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:36:23,784][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:36:24,280][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:36:24,779][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:36:25,276][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:36:25,774][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:36:26,271][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:36:26,768][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:36:27,263][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:36:27,758][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:36:28,258][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:36:28,756][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:36:29,261][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:36:29,766][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:36:30,270][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:36:30,775][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:36:31,278][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:36:31,793][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:36:32,297][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:36:32,804][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:36:33,311][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:36:33,817][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:36:34,323][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:36:34,833][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:36:35,341][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:36:35,843][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:36:36,345][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:36:36,846][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:36:37,344][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:36:37,842][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:36:38,343][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:36:38,841][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:36:39,341][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:36:39,844][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:36:40,345][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:36:40,845][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:36:41,345][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:36:41,845][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:36:42,347][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:36:42,847][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:36:43,346][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:36:43,844][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:36:44,343][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:36:44,846][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:36:45,344][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:36:45,842][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:36:46,341][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:36:46,841][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:36:47,342][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:36:47,846][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:36:48,348][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:36:48,851][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:36:49,358][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:36:49,856][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:36:50,367][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:36:50,867][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:36:51,369][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10032 tokens. +[2026-03-26 05:36:52,182][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.47%, ΔTime: 00:00:32 +[2026-03-26 05:36:52,972][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:36:52,975][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:36:52,977][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:36:53,706][__main__][INFO] - Iteration 499 took 53s (33.09% Gen, 65.55% Train). Generation: 17s, Training: 35s. Estimated remaining time: 37h 12m 39s. Estimated total time: 44h 48m 45s. Time estimates for 10 more iterations: 8m 57s, 100 more iterations: 1h 29m 37s, 500 more iterations: 7h 28m 7s. +[2026-03-26 05:36:53,708][__main__][INFO] - Starting iteration 499. +[2026-03-26 05:36:54,111][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 49 and human policies 1. +[2026-03-26 05:36:54,112][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:37:10,763][__main__][INFO] - Number of regex retries in iteration 499: 0 +[2026-03-26 05:37:10,764][__main__][INFO] - agents played in iteration 499 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:37:11,537][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:37:11,561][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:37:11,584][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:37:11,608][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:37:11,608][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:37:11,609][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:37:12,377][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:37:12,827][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:37:13,331][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:37:13,831][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:37:14,337][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:37:14,832][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:37:15,333][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:37:15,833][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:37:16,332][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:37:16,830][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:37:17,328][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:37:17,829][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:37:18,330][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:37:18,830][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:37:19,329][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:37:19,827][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:37:20,331][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:37:20,830][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:37:21,329][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:37:21,827][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:37:22,327][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:37:22,826][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:37:23,332][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:37:23,832][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:37:24,336][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:37:24,839][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:37:25,341][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:37:25,844][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:37:26,349][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:37:26,855][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:37:27,356][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:37:27,857][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:37:28,361][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:37:28,863][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:37:29,363][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:37:29,862][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:37:30,360][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:37:30,865][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:37:31,367][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:37:31,866][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:37:32,364][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:37:32,861][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:37:33,359][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:37:33,857][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:37:34,353][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:37:34,850][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:37:35,346][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:37:35,843][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:37:36,339][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:37:36,839][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:37:37,334][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:37:37,836][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:37:38,343][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:37:38,845][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:37:39,344][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:37:39,842][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:37:40,340][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:37:40,840][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:37:41,335][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:37:41,836][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:37:42,338][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:37:42,842][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:37:43,346][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:37:43,850][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:37:44,353][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10044 tokens. +[2026-03-26 05:37:45,144][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.23%, Current % of VRAM taken: 60.66%, Block Peak % of device VRAM: 62.34%, ΔTime: 00:00:32 +[2026-03-26 05:37:45,894][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:37:45,896][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:37:45,898][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:37:46,590][__main__][INFO] - Iteration 500 took 52s (31.73% Gen, 66.95% Train). Generation: 16s, Training: 35s. Estimated remaining time: 36h 6m 58s. Estimated total time: 43h 43m 56s. Time estimates for 10 more iterations: 8m 44s, 100 more iterations: 1h 27m 27s, 500 more iterations: 7h 17m 19s. +[2026-03-26 05:37:46,592][__main__][INFO] - Starting iteration 500. +[2026-03-26 05:37:47,016][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 49 and human policies 1. +[2026-03-26 05:37:47,016][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:38:04,078][__main__][INFO] - Number of regex retries in iteration 500: 0 +[2026-03-26 05:38:04,079][__main__][INFO] - agents played in iteration 500 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:38:04,856][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.59%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:38:04,880][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.59%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:38:04,903][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.59%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:38:04,926][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.59%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:38:04,927][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:38:04,928][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:38:05,681][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:38:06,134][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:38:06,637][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:38:07,132][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:38:07,632][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:38:08,130][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:38:08,626][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:38:09,124][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:38:09,623][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:38:10,120][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:38:10,617][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:38:11,114][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:38:11,610][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:38:12,107][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:38:12,604][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:38:13,102][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:38:13,839][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:38:14,335][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:38:14,832][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:38:15,330][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:38:15,828][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:38:16,325][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:38:16,822][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:38:17,320][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:38:17,818][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:38:18,315][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:38:18,812][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:38:19,310][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:38:19,810][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:38:20,310][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:38:20,808][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:38:21,304][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:38:21,802][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:38:22,301][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:38:22,798][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:38:23,296][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:38:23,792][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:38:24,290][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:38:24,788][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:38:25,285][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:38:25,783][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:38:26,282][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:38:26,779][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:38:27,277][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:38:27,773][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:38:28,271][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:38:28,766][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:38:29,263][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:38:29,761][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:38:30,259][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:38:30,755][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:38:31,250][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:38:31,748][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:38:32,245][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:38:32,742][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:38:33,239][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:38:33,736][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:38:34,232][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:38:34,730][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:38:35,229][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:38:35,730][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:38:36,236][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:38:36,739][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:38:37,241][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:38:37,740][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9839 tokens. +[2026-03-26 05:38:38,551][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.00%, Current % of VRAM taken: 60.42%, Block Peak % of device VRAM: 61.83%, ΔTime: 00:00:32 +[2026-03-26 05:38:39,298][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:38:39,300][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:38:39,301][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:38:40,704][__main__][INFO] - Iteration 501 took 53s (31.78% Gen, 65.60% Train). Generation: 17s, Training: 35s. Estimated remaining time: 37h 6m 35s. Estimated total time: 44h 44m 27s. Time estimates for 10 more iterations: 8m 56s, 100 more iterations: 1h 29m 28s, 500 more iterations: 7h 27m 24s. +[2026-03-26 05:38:40,706][__main__][INFO] - Starting iteration 501. +[2026-03-26 05:38:41,116][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 50 and human policies 1. +[2026-03-26 05:38:41,116][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:39:02,668][__main__][INFO] - Number of regex retries in iteration 501: 0 +[2026-03-26 05:39:02,669][__main__][INFO] - agents played in iteration 501 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:39:03,442][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:39:03,466][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:39:03,489][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:39:03,513][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:39:03,513][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:39:03,514][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:39:04,269][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:39:04,722][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:39:05,220][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:39:05,715][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:39:06,212][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:39:06,705][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:39:07,198][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:39:07,692][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:39:08,189][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:39:08,690][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:39:09,187][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:39:09,686][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:39:10,181][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:39:10,678][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:39:11,179][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:39:11,683][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:39:12,183][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:39:12,682][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:39:13,181][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:39:13,680][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:39:14,177][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:39:14,672][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:39:15,170][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:39:15,667][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:39:16,166][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:39:16,663][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:39:17,158][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:39:17,658][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:39:18,157][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:39:18,655][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:39:19,156][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:39:19,659][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:39:20,159][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:39:20,657][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:39:21,156][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:39:21,652][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:39:22,149][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:39:22,645][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:39:23,143][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:39:23,642][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:39:24,138][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:39:24,634][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:39:25,131][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:39:25,629][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:39:26,126][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:39:26,631][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:39:27,132][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:39:27,630][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:39:28,127][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:39:28,623][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:39:29,122][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:39:29,627][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:39:30,131][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:39:30,632][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:39:31,131][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:39:31,627][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:39:32,127][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:39:32,633][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:39:33,133][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:39:33,632][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:39:34,130][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:39:34,635][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:39:35,135][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:39:35,631][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:39:36,129][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9973 tokens. +[2026-03-26 05:39:36,936][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.04%, Current % of VRAM taken: 60.47%, Block Peak % of device VRAM: 62.09%, ΔTime: 00:00:32 +[2026-03-26 05:39:37,650][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:39:37,653][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:39:37,654][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:39:38,401][__main__][INFO] - Iteration 502 took 57s (37.62% Gen, 61.07% Train). Generation: 21s, Training: 34s. Estimated remaining time: 40h 5m 27s. Estimated total time: 47h 44m 17s. Time estimates for 10 more iterations: 9m 32s, 100 more iterations: 1h 35m 28s, 500 more iterations: 7h 57m 22s. +[2026-03-26 05:39:38,403][__main__][INFO] - Starting iteration 502. +[2026-03-26 05:39:38,818][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 50 and human policies 1. +[2026-03-26 05:39:38,818][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:40:17,602][__main__][INFO] - Number of regex retries in iteration 502: 0 +[2026-03-26 05:40:17,603][__main__][INFO] - agents played in iteration 502 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:40:18,382][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:40:18,406][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:40:18,429][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:40:18,453][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:40:18,453][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:40:18,454][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:40:19,215][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:40:19,668][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:40:20,172][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:40:20,671][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:40:21,169][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:40:21,667][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:40:22,165][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:40:22,664][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:40:23,162][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:40:23,658][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:40:24,161][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:40:24,660][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:40:25,158][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:40:25,656][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:40:26,155][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:40:26,655][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:40:27,154][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:40:27,647][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:40:28,143][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:40:28,641][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:40:29,139][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:40:29,633][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:40:30,126][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:40:30,623][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:40:31,118][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:40:31,612][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:40:32,105][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:40:32,605][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:40:33,103][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:40:33,599][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:40:34,094][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:40:34,597][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:40:35,093][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:40:35,591][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:40:36,091][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:40:36,586][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:40:37,106][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:40:37,608][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:40:38,106][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:40:38,603][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:40:39,103][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:40:39,601][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:40:40,099][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:40:40,597][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:40:41,099][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:40:41,596][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:40:42,093][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:40:42,590][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:40:43,086][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:40:43,583][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:40:44,079][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:40:44,574][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:40:45,070][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:40:45,565][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:40:46,060][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:40:46,556][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:40:47,051][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:40:47,548][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:40:48,049][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:40:48,547][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:40:49,046][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:40:49,548][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:40:50,049][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:40:50,550][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:40:51,050][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9967 tokens. +[2026-03-26 05:40:51,837][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.17%, Current % of VRAM taken: 60.60%, Block Peak % of device VRAM: 62.31%, ΔTime: 00:00:32 +[2026-03-26 05:40:52,640][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:40:52,642][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:40:52,644][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:40:53,755][__main__][INFO] - Iteration 503 took 1m 14s (51.75% Gen, 46.76% Train). Generation: 38s, Training: 35s. Estimated remaining time: 54h 46m 50s. Estimated total time: 62h 26m 55s. Time estimates for 10 more iterations: 12m 29s, 100 more iterations: 2h 4m 53s, 500 more iterations: 10h 24m 29s. +[2026-03-26 05:40:53,757][__main__][INFO] - Starting iteration 503. +[2026-03-26 05:40:54,160][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 50 and human policies 1. +[2026-03-26 05:40:54,161][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:41:35,179][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:41:45,081][__main__][INFO] - Number of regex retries in iteration 503: 1 +[2026-03-26 05:41:45,082][__main__][INFO] - agents played in iteration 503 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:41:45,855][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:41:45,879][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:41:45,902][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:41:45,926][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:41:45,926][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:41:45,927][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:41:46,696][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:41:47,151][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:41:47,652][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:41:48,165][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:41:48,667][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:41:49,165][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:41:49,662][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:41:50,160][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:41:50,671][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:41:51,172][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:41:51,671][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:41:52,170][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:41:52,668][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:41:53,165][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:41:53,665][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:41:54,167][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:41:54,666][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:41:55,170][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:41:55,673][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:41:56,175][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:41:56,676][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:41:57,178][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:41:57,681][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:41:58,184][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:41:58,687][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:41:59,185][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:41:59,678][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:42:00,175][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:42:00,681][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:42:01,183][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:42:01,682][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:42:02,182][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:42:02,680][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:42:03,178][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:42:03,677][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:42:04,181][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:42:04,687][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:42:05,187][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:42:05,684][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:42:06,183][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:42:06,685][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:42:07,184][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:42:07,684][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:42:08,184][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:42:08,684][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:42:09,183][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:42:09,682][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:42:10,181][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:42:10,681][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:42:11,176][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:42:11,670][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:42:12,165][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:42:12,662][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:42:13,156][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:42:13,650][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:42:14,145][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:42:14,641][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:42:15,138][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:42:15,637][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:42:16,135][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:42:16,637][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:42:17,144][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:42:17,652][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:42:18,180][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:42:18,679][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9902 tokens. +[2026-03-26 05:42:19,501][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 12.96%, Current % of VRAM taken: 59.38%, Block Peak % of device VRAM: 62.14%, ΔTime: 00:00:32 +[2026-03-26 05:42:20,229][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:42:20,232][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:42:20,233][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:42:20,924][__main__][INFO] - Iteration 504 took 1m 26s (58.69% Gen, 40.51% Train). Generation: 50s, Training: 35s. Estimated remaining time: 64h 36m 41s. Estimated total time: 72h 18m 13s. Time estimates for 10 more iterations: 14m 27s, 100 more iterations: 2h 24m 36s, 500 more iterations: 12h 3m 2s. +[2026-03-26 05:42:20,926][__main__][INFO] - Starting iteration 504. +[2026-03-26 05:42:21,337][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 50 and human policies 1. +[2026-03-26 05:42:21,337][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:42:25,679][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:42:25,767][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:42:37,273][__main__][INFO] - Number of regex retries in iteration 504: 2 +[2026-03-26 05:42:37,274][__main__][INFO] - agents played in iteration 504 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:42:38,062][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:42:38,086][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:42:38,110][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:42:38,134][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:42:38,135][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:42:38,135][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:42:38,904][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:42:39,358][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:42:39,859][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:42:40,361][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:42:40,863][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:42:41,362][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:42:41,864][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:42:42,362][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:42:42,860][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:42:43,365][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:42:43,861][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:42:44,354][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:42:44,852][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:42:45,364][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:42:45,862][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:42:46,359][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:42:46,863][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:42:47,361][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:42:47,856][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:42:48,354][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:42:48,851][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:42:49,350][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:42:49,852][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:42:50,352][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:42:50,852][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:42:51,355][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:42:51,857][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:42:52,359][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:42:52,860][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:42:53,361][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:42:53,857][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:42:54,355][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:42:54,856][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:42:55,355][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:42:55,851][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:42:56,368][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:42:56,868][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:42:57,370][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:42:57,870][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:42:58,371][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:42:58,870][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:42:59,371][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:42:59,873][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:43:00,377][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:43:00,873][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:43:01,376][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:43:01,883][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:43:02,391][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:43:02,891][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:43:03,391][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:43:03,894][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:43:04,393][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:43:04,888][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:43:05,387][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:43:05,887][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:43:06,387][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:43:06,885][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:43:07,385][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:43:07,885][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:43:08,386][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:43:08,889][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:43:09,392][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:43:09,896][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:43:10,400][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:43:10,900][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9955 tokens. +[2026-03-26 05:43:11,764][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.96%, Current % of VRAM taken: 60.39%, Block Peak % of device VRAM: 62.10%, ΔTime: 00:00:32 +[2026-03-26 05:43:12,504][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:43:12,513][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:43:12,514][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:43:13,227][__main__][INFO] - Iteration 505 took 51s (30.71% Gen, 67.91% Train). Generation: 15s, Training: 35s. Estimated remaining time: 35h 32m 8s. Estimated total time: 43h 14m 32s. Time estimates for 10 more iterations: 8m 38s, 100 more iterations: 1h 26m 29s, 500 more iterations: 7h 12m 25s. +[2026-03-26 05:43:13,229][__main__][INFO] - Starting iteration 505. +[2026-03-26 05:43:13,635][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 50 and human policies 1. +[2026-03-26 05:43:13,635][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:43:23,283][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:43:29,408][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 20 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:43:35,862][__main__][INFO] - Number of regex retries in iteration 505: 2 +[2026-03-26 05:43:35,862][__main__][INFO] - agents played in iteration 505 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:43:36,643][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:43:36,669][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:43:36,695][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:43:36,720][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:43:36,721][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:43:36,721][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:43:37,519][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:43:37,979][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:43:38,483][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:43:38,986][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:43:39,484][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:43:39,983][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:43:40,481][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:43:40,978][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:43:41,475][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:43:41,971][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:43:42,465][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:43:42,964][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:43:43,467][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:43:43,974][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:43:44,472][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:43:44,970][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:43:45,467][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:43:45,961][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:43:46,455][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:43:46,948][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:43:47,441][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:43:47,945][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:43:48,447][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:43:48,944][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:43:49,440][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:43:49,936][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:43:50,440][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:43:50,937][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:43:51,433][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:43:51,941][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:43:52,436][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:43:52,936][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:43:53,430][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:43:53,931][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:43:54,428][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:43:54,928][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:43:55,429][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:43:55,931][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:43:56,433][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:43:56,934][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:43:57,432][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:43:57,930][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:43:58,430][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:43:58,932][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:43:59,430][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:43:59,932][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:44:00,432][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:44:00,931][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:44:01,429][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:44:01,927][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:44:02,428][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:44:02,926][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:44:03,423][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:44:03,922][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:44:04,425][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:44:04,920][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:44:05,419][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:44:05,926][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:44:06,429][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:44:06,928][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:44:07,428][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:44:07,931][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:44:08,434][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:44:08,934][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:44:09,438][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9886 tokens. +[2026-03-26 05:44:10,262][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.21%, Current % of VRAM taken: 60.63%, Block Peak % of device VRAM: 62.21%, ΔTime: 00:00:32 +[2026-03-26 05:44:11,012][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:44:11,014][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:44:11,016][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:44:11,853][__main__][INFO] - Iteration 506 took 58s (38.18% Gen, 60.38% Train). Generation: 22s, Training: 35s. Estimated remaining time: 40h 47m 34s. Estimated total time: 48h 30m 57s. Time estimates for 10 more iterations: 9m 42s, 100 more iterations: 1h 37m 1s, 500 more iterations: 8h 5m 9s. +[2026-03-26 05:44:11,856][__main__][INFO] - Starting iteration 506. +[2026-03-26 05:44:12,257][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 50 and human policies 1. +[2026-03-26 05:44:12,257][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:44:35,285][__main__][INFO] - Number of regex retries in iteration 506: 0 +[2026-03-26 05:44:35,285][__main__][INFO] - agents played in iteration 506 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:44:36,083][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:44:36,107][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:44:36,131][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:44:36,155][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:44:36,155][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:44:36,156][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:44:36,943][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:44:37,400][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:44:37,907][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:44:38,409][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:44:38,915][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:44:39,416][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:44:39,925][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:44:40,427][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:44:40,930][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:44:41,428][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:44:41,928][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:44:42,428][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:44:42,930][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:44:43,437][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:44:43,938][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:44:44,439][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:44:44,936][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:44:45,437][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:44:45,935][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:44:46,443][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:44:46,947][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:44:47,447][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:44:47,948][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:44:48,447][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:44:48,944][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:44:49,445][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:44:49,955][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:44:50,457][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:44:50,961][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:44:51,457][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:44:51,956][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:44:52,450][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:44:52,948][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:44:53,448][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:44:53,951][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:44:54,450][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:44:54,951][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:44:55,448][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:44:55,947][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:44:56,445][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:44:56,938][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:44:57,437][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:44:57,937][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:44:58,437][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:44:58,935][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:44:59,438][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:44:59,942][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:45:00,448][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:45:00,949][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:45:01,452][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:45:01,955][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:45:02,462][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:45:02,972][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:45:03,475][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:45:03,979][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:45:04,488][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:45:04,999][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:45:05,504][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:45:06,005][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:45:06,508][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:45:07,010][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:45:07,510][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:45:08,017][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:45:08,524][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:45:09,027][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9942 tokens. +[2026-03-26 05:45:09,861][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.20%, Current % of VRAM taken: 60.63%, Block Peak % of device VRAM: 62.23%, ΔTime: 00:00:32 +[2026-03-26 05:45:10,624][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:45:10,626][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:45:10,628][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:45:11,325][__main__][INFO] - Iteration 507 took 59s (38.99% Gen, 59.83% Train). Generation: 23s, Training: 35s. Estimated remaining time: 41h 29m 3s. Estimated total time: 49h 13m 25s. Time estimates for 10 more iterations: 9m 50s, 100 more iterations: 1h 38m 26s, 500 more iterations: 8h 12m 14s. +[2026-03-26 05:45:11,327][__main__][INFO] - Starting iteration 507. +[2026-03-26 05:45:11,752][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 50 and human policies 1. +[2026-03-26 05:45:11,752][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:45:18,005][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:45:27,190][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:45:28,162][__main__][INFO] - Number of regex retries in iteration 507: 2 +[2026-03-26 05:45:28,163][__main__][INFO] - agents played in iteration 507 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:45:28,951][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:45:28,977][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:45:29,002][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:45:29,027][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:45:29,028][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:45:29,028][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:45:29,806][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:45:30,265][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:45:30,767][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:45:31,265][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:45:31,766][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:45:32,266][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:45:32,768][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:45:33,267][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:45:33,766][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:45:34,271][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:45:34,769][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:45:35,265][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:45:35,770][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:45:36,277][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:45:36,776][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:45:37,275][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:45:37,775][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:45:38,273][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:45:38,770][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:45:39,267][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:45:39,764][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:45:40,264][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:45:40,765][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:45:41,267][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:45:41,767][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:45:42,267][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:45:42,765][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:45:43,264][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:45:43,762][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:45:44,264][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:45:44,764][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:45:45,265][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:45:45,763][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:45:46,268][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:45:46,773][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:45:47,276][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:45:47,775][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:45:48,273][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:45:48,772][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:45:49,272][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:45:49,775][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:45:50,277][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:45:50,778][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:45:51,278][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:45:51,776][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:45:52,274][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:45:52,773][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:45:53,272][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:45:53,774][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:45:54,278][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:45:54,776][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:45:55,277][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:45:55,778][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:45:56,279][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:45:56,776][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:45:57,273][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:45:57,771][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:45:58,268][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:45:58,764][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:45:59,262][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:45:59,760][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:46:00,267][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:46:00,766][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:46:01,265][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:46:01,759][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9884 tokens. +[2026-03-26 05:46:02,673][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.92%, Current % of VRAM taken: 60.35%, Block Peak % of device VRAM: 62.03%, ΔTime: 00:00:32 +[2026-03-26 05:46:03,419][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:46:03,421][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:46:03,423][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:46:04,149][__main__][INFO] - Iteration 508 took 52s (31.32% Gen, 67.29% Train). Generation: 16s, Training: 35s. Estimated remaining time: 35h 54m 39s. Estimated total time: 43h 39m 55s. Time estimates for 10 more iterations: 8m 43s, 100 more iterations: 1h 27m 19s, 500 more iterations: 7h 16m 39s. +[2026-03-26 05:46:04,152][__main__][INFO] - Starting iteration 508. +[2026-03-26 05:46:04,554][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 50 and human policies 1. +[2026-03-26 05:46:04,554][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:46:19,666][__main__][INFO] - Number of regex retries in iteration 508: 0 +[2026-03-26 05:46:19,667][__main__][INFO] - agents played in iteration 508 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:46:20,457][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:46:20,481][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:46:20,505][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:46:20,528][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:46:20,529][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:46:20,529][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:46:21,314][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:46:21,771][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:46:22,275][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:46:22,782][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:46:23,293][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:46:23,794][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:46:24,294][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:46:24,797][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:46:25,308][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:46:25,809][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:46:26,308][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:46:26,806][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:46:27,305][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:46:27,803][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:46:28,302][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:46:28,802][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:46:29,305][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:46:29,809][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:46:30,310][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:46:30,811][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:46:31,310][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:46:31,811][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:46:32,311][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:46:32,810][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:46:33,309][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:46:33,801][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:46:34,295][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:46:34,791][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:46:35,285][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:46:35,782][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:46:36,284][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:46:36,784][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:46:37,281][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:46:37,784][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:46:38,289][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:46:38,790][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:46:39,293][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:46:39,797][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:46:40,300][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:46:40,805][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:46:41,306][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:46:41,807][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:46:42,307][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:46:42,803][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:46:43,300][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:46:43,797][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:46:44,298][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:46:44,797][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:46:45,296][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:46:45,794][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:46:46,290][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:46:46,787][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:46:47,286][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:46:47,783][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:46:48,281][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:46:48,778][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:46:49,275][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:46:49,772][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:46:50,271][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:46:50,771][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:46:51,268][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:46:51,766][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:46:52,264][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:46:52,762][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:46:53,263][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9824 tokens. +[2026-03-26 05:46:54,058][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.00%, Current % of VRAM taken: 60.43%, Block Peak % of device VRAM: 62.06%, ΔTime: 00:00:32 +[2026-03-26 05:46:54,865][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:46:54,867][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:46:54,871][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:46:55,837][__main__][INFO] - Iteration 509 took 51s (29.47% Gen, 68.65% Train). Generation: 15s, Training: 35s. Estimated remaining time: 34h 58m 5s. Estimated total time: 42h 44m 12s. Time estimates for 10 more iterations: 8m 32s, 100 more iterations: 1h 25m 28s, 500 more iterations: 7h 7m 22s. +[2026-03-26 05:46:55,839][__main__][INFO] - Starting iteration 509. +[2026-03-26 05:46:56,248][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 50 and human policies 1. +[2026-03-26 05:46:56,249][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:47:53,218][__main__][INFO] - Number of regex retries in iteration 509: 0 +[2026-03-26 05:47:53,219][__main__][INFO] - agents played in iteration 509 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:47:53,991][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:47:54,014][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:47:54,038][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:47:54,061][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:47:54,062][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:47:54,062][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:47:54,831][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:47:55,286][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:47:55,787][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:47:56,284][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:47:56,781][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:47:57,277][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:47:57,774][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:47:58,276][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:47:58,780][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:47:59,285][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:47:59,788][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:48:00,291][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:48:00,790][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:48:01,289][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:48:01,788][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:48:02,288][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:48:02,787][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:48:03,285][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:48:03,787][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:48:04,285][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:48:04,783][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:48:05,285][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:48:05,788][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:48:06,296][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:48:06,794][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:48:07,290][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:48:07,789][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:48:08,288][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:48:08,789][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:48:09,287][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:48:09,786][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:48:10,284][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:48:10,789][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:48:11,291][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:48:11,790][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:48:12,289][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:48:12,787][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:48:13,285][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:48:13,784][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:48:14,286][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:48:14,784][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:48:15,284][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:48:15,781][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:48:16,276][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:48:16,773][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:48:17,271][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:48:17,768][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:48:18,265][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:48:18,762][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:48:19,260][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:48:19,755][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:48:20,254][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:48:20,749][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:48:21,244][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:48:21,738][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:48:22,232][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:48:22,728][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:48:23,231][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:48:23,731][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:48:24,236][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:48:24,736][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:48:25,236][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:48:25,743][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:48:26,243][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:48:26,746][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9921 tokens. +[2026-03-26 05:48:27,565][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.04%, Current % of VRAM taken: 60.47%, Block Peak % of device VRAM: 62.11%, ΔTime: 00:00:32 +[2026-03-26 05:48:28,357][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:48:28,359][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:48:28,361][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:48:29,171][__main__][INFO] - Iteration 510 took 1m 32s (61.31% Gen, 37.82% Train). Generation: 56s, Training: 35s. Estimated remaining time: 69h 38m 32s. Estimated total time: 77h 26m 13s. Time estimates for 10 more iterations: 15m 29s, 100 more iterations: 2h 34m 52s, 500 more iterations: 12h 54m 22s. +[2026-03-26 05:48:29,174][__main__][INFO] - Starting iteration 510. +[2026-03-26 05:48:29,596][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 50 and human policies 1. +[2026-03-26 05:48:29,597][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:48:45,944][__main__][INFO] - Number of regex retries in iteration 510: 0 +[2026-03-26 05:48:45,944][__main__][INFO] - agents played in iteration 510 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:48:46,717][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:48:46,742][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:48:46,766][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:48:46,791][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:48:46,791][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:48:46,792][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:48:47,568][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:48:48,024][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:48:48,533][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:48:49,034][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:48:49,537][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:48:50,039][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:48:50,539][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:48:51,042][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:48:51,544][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:48:52,048][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:48:52,551][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:48:53,058][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:48:53,563][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:48:54,066][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:48:54,566][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:48:55,068][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:48:55,568][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:48:56,066][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:48:56,565][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:48:57,064][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:48:57,562][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:48:58,062][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:48:58,559][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:48:59,056][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:48:59,553][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:49:00,050][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:49:00,547][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:49:01,045][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:49:01,545][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:49:02,050][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:49:02,550][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:49:03,049][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:49:03,551][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:49:04,050][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:49:04,544][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:49:05,043][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:49:05,539][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:49:06,040][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:49:06,548][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:49:07,063][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:49:07,561][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:49:08,060][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:49:08,558][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:49:09,057][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:49:09,555][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:49:10,054][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:49:10,552][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:49:11,050][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:49:11,547][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:49:12,043][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:49:12,540][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:49:13,034][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:49:13,531][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:49:14,035][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:49:14,542][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:49:15,040][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:49:15,539][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:49:16,039][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:49:16,539][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:49:17,038][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:49:17,538][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:49:18,039][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:49:18,538][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:49:19,038][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:49:19,539][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9766 tokens. +[2026-03-26 05:49:20,364][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 61.84%, ΔTime: 00:00:32 +[2026-03-26 05:49:21,116][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:49:21,118][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:49:21,120][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:49:22,517][__main__][INFO] - Iteration 511 took 52s (30.89% Gen, 66.47% Train). Generation: 16s, Training: 35s. Estimated remaining time: 36h 17m 32s. Estimated total time: 44h 6m 6s. Time estimates for 10 more iterations: 8m 49s, 100 more iterations: 1h 28m 12s, 500 more iterations: 7h 21m 1s. +[2026-03-26 05:49:22,520][__main__][INFO] - Starting iteration 511. +[2026-03-26 05:49:22,940][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 51 and human policies 1. +[2026-03-26 05:49:22,941][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:50:22,772][__main__][INFO] - Number of regex retries in iteration 511: 0 +[2026-03-26 05:50:22,773][__main__][INFO] - agents played in iteration 511 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:50:23,554][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:50:23,578][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:50:23,602][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:50:23,626][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:50:23,627][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:50:23,628][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:50:24,407][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:50:24,872][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:50:25,373][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:50:25,874][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:50:26,371][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:50:26,870][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:50:27,380][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:50:27,880][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:50:28,378][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:50:28,877][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:50:29,378][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:50:29,879][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:50:30,384][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:50:30,884][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:50:31,393][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:50:31,893][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:50:32,392][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:50:32,888][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:50:33,386][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:50:33,882][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:50:34,376][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:50:34,870][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:50:35,366][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:50:35,865][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:50:36,360][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:50:36,855][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:50:37,361][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:50:37,864][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:50:38,364][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:50:38,865][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:50:39,365][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:50:39,863][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:50:40,361][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:50:40,861][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:50:41,356][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:50:41,853][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:50:42,356][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:50:42,859][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:50:43,362][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:50:43,862][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:50:44,362][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:50:44,861][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:50:45,357][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:50:45,860][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:50:46,362][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:50:46,861][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:50:47,364][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:50:47,864][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:50:48,362][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:50:48,862][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:50:49,360][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:50:49,871][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:50:50,377][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:50:50,874][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:50:51,381][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:50:51,880][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:50:52,384][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:50:52,880][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:50:53,378][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:50:53,886][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:50:54,390][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:50:54,888][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:50:55,386][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:50:55,890][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:50:56,391][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9915 tokens. +[2026-03-26 05:50:57,216][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 11.96%, Current % of VRAM taken: 58.39%, Block Peak % of device VRAM: 62.23%, ΔTime: 00:00:32 +[2026-03-26 05:50:58,261][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:50:58,264][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:50:58,265][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:50:59,248][__main__][INFO] - Iteration 512 took 1m 36s (62.13% Gen, 36.85% Train). Generation: 59s, Training: 35s. Estimated remaining time: 72h 25m 14s. Estimated total time: 80h 15m 24s. Time estimates for 10 more iterations: 16m 3s, 100 more iterations: 2h 40m 30s, 500 more iterations: 13h 22m 34s. +[2026-03-26 05:50:59,250][__main__][INFO] - Starting iteration 512. +[2026-03-26 05:50:59,672][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 51 and human policies 1. +[2026-03-26 05:50:59,672][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:51:17,101][__main__][INFO] - Number of regex retries in iteration 512: 0 +[2026-03-26 05:51:17,102][__main__][INFO] - agents played in iteration 512 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:51:17,890][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 51.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:51:17,915][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 51.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:51:17,940][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 51.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:51:17,965][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 51.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:51:17,965][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:51:17,966][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:51:18,736][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:51:19,192][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:51:19,694][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:51:20,195][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:51:20,696][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:51:21,194][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:51:21,692][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:51:22,191][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:51:22,689][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:51:23,194][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:51:23,696][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:51:24,200][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:51:24,700][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:51:25,201][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:51:25,700][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:51:26,199][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:51:26,697][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:51:27,195][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:51:27,693][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:51:28,191][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:51:28,689][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:51:29,187][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:51:29,686][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:51:30,184][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:51:30,683][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:51:31,180][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:51:31,678][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:51:32,174][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:51:32,673][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:51:33,169][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:51:33,669][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:51:34,190][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:51:34,688][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:51:35,187][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:51:35,684][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:51:36,188][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:51:36,691][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:51:37,188][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:51:37,684][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:51:38,182][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:51:38,682][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:51:39,183][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:51:39,685][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:51:40,186][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:51:40,687][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:51:41,188][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:51:41,689][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:51:42,189][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:51:42,689][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:51:43,197][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:51:43,703][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:51:44,208][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:51:44,711][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:51:45,211][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:51:45,712][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:51:46,215][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:51:46,718][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:51:47,225][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:51:47,728][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:51:48,231][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:51:48,736][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:51:49,244][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:51:49,751][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:51:50,251][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:51:50,751][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9932 tokens. +[2026-03-26 05:51:51,558][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.15%, Current % of VRAM taken: 60.58%, Block Peak % of device VRAM: 62.09%, ΔTime: 00:00:32 +[2026-03-26 05:51:52,284][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:51:52,286][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:51:52,288][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:51:52,999][__main__][INFO] - Iteration 513 took 53s (32.68% Gen, 65.98% Train). Generation: 17s, Training: 35s. Estimated remaining time: 36h 35m 20s. Estimated total time: 44h 26m 24s. Time estimates for 10 more iterations: 8m 53s, 100 more iterations: 1h 28m 52s, 500 more iterations: 7h 24m 24s. +[2026-03-26 05:51:53,001][__main__][INFO] - Starting iteration 513. +[2026-03-26 05:51:53,409][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 51 and human policies 1. +[2026-03-26 05:51:53,410][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:52:10,687][__main__][INFO] - Number of regex retries in iteration 513: 0 +[2026-03-26 05:52:10,688][__main__][INFO] - agents played in iteration 513 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:52:11,467][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:52:11,491][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:52:11,515][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:52:11,539][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:52:11,540][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:52:11,541][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:52:12,315][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:52:12,770][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:52:13,270][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:52:13,769][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:52:14,268][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:52:14,771][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:52:15,276][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:52:15,774][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:52:16,273][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:52:16,775][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:52:17,283][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:52:17,795][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:52:18,299][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:52:18,801][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:52:19,305][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:52:19,805][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:52:20,344][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:52:20,850][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:52:21,351][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:52:21,850][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:52:22,348][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:52:22,847][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:52:23,346][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:52:23,845][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:52:24,345][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:52:24,843][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:52:25,346][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:52:25,845][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:52:26,345][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:52:26,848][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:52:27,351][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:52:27,851][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:52:28,349][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:52:28,847][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:52:29,347][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:52:29,846][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:52:30,351][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:52:30,857][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:52:31,356][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:52:31,857][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:52:32,351][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:52:32,847][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:52:33,351][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:52:33,852][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:52:34,355][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:52:34,854][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:52:35,350][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:52:35,847][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:52:36,351][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:52:36,856][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:52:37,353][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:52:37,850][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:52:38,348][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:52:38,845][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:52:39,345][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:52:39,848][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:52:40,352][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:52:40,854][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:52:41,359][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:52:41,862][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:52:42,364][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:52:42,866][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:52:43,365][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:52:43,865][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:52:44,366][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9812 tokens. +[2026-03-26 05:52:45,203][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.00%, Current % of VRAM taken: 60.42%, Block Peak % of device VRAM: 62.18%, ΔTime: 00:00:32 +[2026-03-26 05:52:45,955][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:52:45,957][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:52:45,958][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:52:46,681][__main__][INFO] - Iteration 514 took 53s (32.43% Gen, 66.21% Train). Generation: 17s, Training: 35s. Estimated remaining time: 36h 31m 38s. Estimated total time: 44h 23m 36s. Time estimates for 10 more iterations: 8m 52s, 100 more iterations: 1h 28m 47s, 500 more iterations: 7h 23m 56s. +[2026-03-26 05:52:46,683][__main__][INFO] - Starting iteration 514. +[2026-03-26 05:52:47,083][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 51 and human policies 1. +[2026-03-26 05:52:47,083][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:52:51,039][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:53:02,556][__main__][INFO] - Number of regex retries in iteration 514: 1 +[2026-03-26 05:53:02,557][__main__][INFO] - agents played in iteration 514 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:53:03,345][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:53:03,368][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:53:03,392][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:53:03,415][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:53:03,416][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:53:03,417][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:53:04,195][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:53:04,658][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:53:05,161][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:53:05,665][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:53:06,167][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:53:06,678][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:53:07,179][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:53:07,683][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:53:08,181][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:53:08,682][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:53:09,194][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:53:09,701][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:53:10,206][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:53:10,713][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:53:11,215][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:53:11,722][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:53:12,230][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:53:12,734][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:53:13,231][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:53:13,734][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:53:14,231][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:53:14,733][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:53:15,239][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:53:15,738][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:53:16,237][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:53:16,734][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:53:17,229][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:53:17,727][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:53:18,229][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:53:18,725][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:53:19,221][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:53:19,718][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:53:20,216][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:53:20,714][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:53:21,208][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:53:21,719][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:53:22,214][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:53:22,709][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:53:23,205][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:53:23,701][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:53:24,195][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:53:24,696][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:53:25,200][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:53:25,704][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:53:26,206][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:53:26,704][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:53:27,203][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:53:27,698][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:53:28,200][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:53:28,698][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:53:29,200][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:53:29,697][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:53:30,192][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:53:30,690][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:53:31,190][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:53:31,695][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:53:32,200][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:53:32,704][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:53:33,218][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:53:33,723][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:53:34,228][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:53:34,729][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:53:35,230][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:53:35,733][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:53:36,235][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9914 tokens. +[2026-03-26 05:53:37,068][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.48%, Block Peak % of device VRAM: 62.21%, ΔTime: 00:00:32 +[2026-03-26 05:53:37,911][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:53:37,914][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:53:37,915][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:53:38,652][__main__][INFO] - Iteration 515 took 51s (30.00% Gen, 68.56% Train). Generation: 15s, Training: 35s. Estimated remaining time: 35h 5m 40s. Estimated total time: 42h 58m 30s. Time estimates for 10 more iterations: 8m 35s, 100 more iterations: 1h 25m 57s, 500 more iterations: 7h 9m 45s. +[2026-03-26 05:53:38,654][__main__][INFO] - Starting iteration 515. +[2026-03-26 05:53:39,061][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 51 and human policies 1. +[2026-03-26 05:53:39,062][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:53:55,389][__main__][INFO] - Number of regex retries in iteration 515: 0 +[2026-03-26 05:53:55,390][__main__][INFO] - agents played in iteration 515 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:53:56,174][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:53:56,198][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:53:56,221][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:53:56,245][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:53:56,246][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:53:56,246][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:53:57,033][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:53:57,491][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:53:57,997][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:53:58,500][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:53:59,001][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:53:59,502][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:54:00,004][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:54:00,504][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:54:01,006][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:54:01,504][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:54:02,003][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:54:02,502][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:54:03,008][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:54:03,516][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:54:04,021][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:54:04,520][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:54:05,019][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:54:05,518][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:54:06,016][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:54:06,512][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:54:07,011][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:54:07,509][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:54:08,006][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:54:08,505][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:54:09,003][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:54:09,501][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:54:10,000][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:54:10,495][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:54:10,994][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:54:11,493][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:54:11,992][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:54:12,492][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:54:12,991][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:54:13,492][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:54:14,004][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:54:14,504][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:54:15,009][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:54:15,506][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:54:16,002][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:54:16,499][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:54:16,998][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:54:17,497][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:54:17,996][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:54:18,496][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:54:18,995][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:54:19,493][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:54:19,989][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:54:20,495][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:54:20,995][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:54:21,498][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:54:21,997][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:54:22,496][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:54:23,000][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:54:23,510][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:54:24,016][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:54:24,519][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:54:25,031][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:54:25,534][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:54:26,039][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:54:26,547][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:54:27,051][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:54:27,555][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:54:28,059][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:54:28,563][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:54:29,066][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9949 tokens. +[2026-03-26 05:54:29,884][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.12%, ΔTime: 00:00:32 +[2026-03-26 05:54:30,612][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:54:30,615][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:54:30,616][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:54:31,459][__main__][INFO] - Iteration 516 took 52s (31.16% Gen, 67.22% Train). Generation: 16s, Training: 35s. Estimated remaining time: 35h 46m 20s. Estimated total time: 43h 40m 3s. Time estimates for 10 more iterations: 8m 44s, 100 more iterations: 1h 27m 20s, 500 more iterations: 7h 16m 40s. +[2026-03-26 05:54:31,461][__main__][INFO] - Starting iteration 516. +[2026-03-26 05:54:31,877][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 51 and human policies 1. +[2026-03-26 05:54:31,878][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:54:37,036][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:54:40,565][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 20 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:54:44,092][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 20 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 05:54:49,224][__main__][INFO] - Number of regex retries in iteration 516: 3 +[2026-03-26 05:54:49,225][__main__][INFO] - agents played in iteration 516 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:54:50,013][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:54:50,036][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:54:50,060][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:54:50,084][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:54:50,084][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:54:50,085][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:54:50,852][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:54:51,303][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:54:51,807][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:54:52,311][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:54:52,819][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:54:53,318][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:54:53,816][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:54:54,314][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:54:54,812][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:54:55,312][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:54:55,811][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:54:56,310][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:54:56,815][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:54:57,313][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:54:57,814][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:54:58,319][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:54:58,822][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:54:59,321][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:54:59,819][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:55:00,318][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:55:00,815][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:55:01,313][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:55:01,810][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:55:02,309][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:55:02,807][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:55:03,305][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:55:03,804][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:55:04,303][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:55:04,801][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:55:05,297][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:55:05,801][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:55:06,303][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:55:06,805][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:55:07,304][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:55:07,803][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:55:08,302][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:55:08,800][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:55:09,297][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:55:09,795][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:55:10,296][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:55:10,793][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:55:11,289][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:55:11,787][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:55:12,286][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:55:12,790][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:55:13,291][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:55:13,794][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:55:14,299][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:55:14,800][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:55:15,300][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:55:15,804][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:55:16,309][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:55:16,812][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:55:17,313][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:55:17,819][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:55:18,318][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:55:18,824][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:55:19,326][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:55:19,824][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:55:20,327][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:55:20,826][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:55:21,324][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:55:21,821][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:55:22,321][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:55:22,824][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9951 tokens. +[2026-03-26 05:55:23,648][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.16%, Current % of VRAM taken: 60.58%, Block Peak % of device VRAM: 62.28%, ΔTime: 00:00:32 +[2026-03-26 05:55:24,324][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:55:24,326][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:55:24,328][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:55:25,152][__main__][INFO] - Iteration 517 took 53s (32.56% Gen, 65.89% Train). Generation: 17s, Training: 35s. Estimated remaining time: 36h 29m 10s. Estimated total time: 44h 23m 47s. Time estimates for 10 more iterations: 8m 52s, 100 more iterations: 1h 28m 47s, 500 more iterations: 7h 23m 57s. +[2026-03-26 05:55:25,155][__main__][INFO] - Starting iteration 517. +[2026-03-26 05:55:25,571][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 51 and human policies 1. +[2026-03-26 05:55:25,571][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:55:40,958][__main__][INFO] - Number of regex retries in iteration 517: 0 +[2026-03-26 05:55:40,959][__main__][INFO] - agents played in iteration 517 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:55:41,743][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:55:41,767][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:55:41,791][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:55:41,815][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:55:41,815][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:55:41,816][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:55:42,595][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:55:43,050][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:55:43,552][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:55:44,056][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:55:44,552][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:55:45,049][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:55:45,546][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:55:46,047][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:55:46,543][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:55:47,042][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:55:47,541][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:55:48,041][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:55:48,539][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:55:49,037][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:55:49,536][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:55:50,036][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:55:50,537][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:55:51,043][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:55:51,544][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:55:52,051][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:55:52,551][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:55:53,052][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:55:53,552][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:55:54,052][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:55:54,552][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:55:55,053][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:55:55,551][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:55:56,046][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:55:56,542][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:55:57,039][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:55:57,534][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:55:58,029][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:55:58,524][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:55:59,021][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:55:59,519][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:56:00,021][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:56:00,524][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:56:01,020][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:56:01,518][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:56:02,014][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:56:02,513][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:56:03,009][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:56:03,505][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:56:04,003][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:56:04,512][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:56:05,013][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:56:05,518][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:56:06,015][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:56:06,522][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:56:07,027][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:56:07,528][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:56:08,031][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:56:08,536][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:56:09,037][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:56:09,537][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:56:10,043][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:56:10,543][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:56:11,041][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:56:11,547][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:56:12,047][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:56:12,556][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:56:13,056][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:56:13,559][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:56:14,065][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:56:14,564][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9939 tokens. +[2026-03-26 05:56:15,404][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.93%, Current % of VRAM taken: 60.35%, Block Peak % of device VRAM: 62.18%, ΔTime: 00:00:32 +[2026-03-26 05:56:16,150][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:56:16,152][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:56:16,154][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:56:17,153][__main__][INFO] - Iteration 518 took 51s (29.83% Gen, 68.23% Train). Generation: 15s, Training: 35s. Estimated remaining time: 35h 3m 40s. Estimated total time: 42h 59m 8s. Time estimates for 10 more iterations: 8m 35s, 100 more iterations: 1h 25m 58s, 500 more iterations: 7h 9m 51s. +[2026-03-26 05:56:17,155][__main__][INFO] - Starting iteration 518. +[2026-03-26 05:56:17,563][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 51 and human policies 1. +[2026-03-26 05:56:17,564][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:57:03,831][__main__][INFO] - Number of regex retries in iteration 518: 0 +[2026-03-26 05:57:03,832][__main__][INFO] - agents played in iteration 518 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:57:04,609][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.40%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:57:04,632][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.40%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:57:04,656][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.40%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:57:04,679][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.40%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:57:04,680][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:57:04,680][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:57:05,459][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:57:05,909][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:57:06,410][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:57:06,906][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:57:07,409][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:57:07,906][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:57:08,406][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:57:08,903][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:57:09,400][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:57:09,895][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:57:10,394][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:57:10,891][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:57:11,390][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:57:11,891][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:57:12,392][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:57:12,891][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:57:13,389][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:57:13,892][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:57:14,394][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:57:14,899][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:57:15,396][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:57:15,894][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:57:16,394][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:57:16,893][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:57:17,389][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:57:17,889][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:57:18,387][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:57:18,885][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:57:19,384][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:57:19,883][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:57:20,383][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:57:20,879][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:57:21,376][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:57:21,872][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:57:22,366][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:57:22,860][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:57:23,361][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:57:23,854][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:57:24,347][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:57:24,841][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:57:25,335][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:57:25,832][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:57:26,330][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:57:26,827][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:57:27,323][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:57:27,820][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:57:28,317][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:57:28,813][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:57:29,309][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:57:29,808][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:57:30,306][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:57:30,806][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:57:31,304][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:57:31,801][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:57:32,300][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:57:32,809][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:57:33,311][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:57:33,811][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:57:34,309][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:57:34,811][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:57:35,311][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:57:35,808][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:57:36,305][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:57:36,803][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:57:37,304][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9865 tokens. +[2026-03-26 05:57:38,134][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.03%, Current % of VRAM taken: 60.46%, Block Peak % of device VRAM: 61.97%, ΔTime: 00:00:32 +[2026-03-26 05:57:38,932][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:57:38,935][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:57:38,937][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:57:40,377][__main__][INFO] - Iteration 519 took 1m 22s (55.87% Gen, 42.39% Train). Generation: 46s, Training: 35s. Estimated remaining time: 61h 3m 50s. Estimated total time: 69h 0m 42s. Time estimates for 10 more iterations: 13m 48s, 100 more iterations: 2h 18m 1s, 500 more iterations: 11h 30m 7s. +[2026-03-26 05:57:40,379][__main__][INFO] - Starting iteration 519. +[2026-03-26 05:57:40,783][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 51 and human policies 1. +[2026-03-26 05:57:40,783][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 05:57:59,905][__main__][INFO] - Number of regex retries in iteration 519: 0 +[2026-03-26 05:57:59,905][__main__][INFO] - agents played in iteration 519 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 05:58:00,692][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:58:00,715][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:58:00,739][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:58:00,762][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 05:58:00,763][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 05:58:00,763][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 05:58:01,547][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 05:58:02,000][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 05:58:02,502][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 05:58:03,006][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 05:58:03,504][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 05:58:04,003][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 05:58:04,506][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 05:58:05,006][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 05:58:05,507][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 05:58:06,003][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 05:58:06,499][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 05:58:06,998][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 05:58:07,495][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 05:58:07,989][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 05:58:08,484][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 05:58:08,983][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 05:58:09,480][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 05:58:09,983][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 05:58:10,484][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 05:58:10,981][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 05:58:11,482][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 05:58:11,977][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 05:58:12,477][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 05:58:12,974][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 05:58:13,470][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 05:58:13,968][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 05:58:14,466][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 05:58:14,963][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 05:58:15,463][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 05:58:15,963][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 05:58:16,461][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 05:58:16,959][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 05:58:17,456][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 05:58:17,977][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 05:58:18,483][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 05:58:18,979][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 05:58:19,483][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 05:58:19,988][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 05:58:20,492][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 05:58:20,995][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 05:58:21,491][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 05:58:21,985][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 05:58:22,480][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 05:58:22,976][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 05:58:23,471][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 05:58:23,965][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 05:58:24,458][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 05:58:24,953][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 05:58:25,447][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 05:58:25,942][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 05:58:26,444][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 05:58:26,945][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 05:58:27,442][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 05:58:27,942][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 05:58:28,442][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 05:58:28,938][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 05:58:29,434][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 05:58:29,933][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 05:58:30,434][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 05:58:30,934][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 05:58:31,429][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 05:58:31,929][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 05:58:32,426][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 05:58:32,926][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 05:58:33,425][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10024 tokens. +[2026-03-26 05:58:34,240][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.17%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.41%, ΔTime: 00:00:32 +[2026-03-26 05:58:34,982][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 05:58:34,984][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 05:58:34,986][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 05:58:35,745][__main__][INFO] - Iteration 520 took 54s (34.79% Gen, 63.82% Train). Generation: 19s, Training: 35s. Estimated remaining time: 37h 50m 21s. Estimated total time: 45h 48m 8s. Time estimates for 10 more iterations: 9m 9s, 100 more iterations: 1h 31m 36s, 500 more iterations: 7h 38m 1s. +[2026-03-26 05:58:35,747][__main__][INFO] - Starting iteration 520. +[2026-03-26 05:58:36,149][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 51 and human policies 1. +[2026-03-26 05:58:36,150][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:00:32,697][__main__][INFO] - Number of regex retries in iteration 520: 0 +[2026-03-26 06:00:32,698][__main__][INFO] - agents played in iteration 520 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:00:33,489][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:00:33,513][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:00:33,538][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:00:33,562][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:00:33,563][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:00:33,563][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:00:34,361][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:00:34,817][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:00:35,329][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:00:35,835][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:00:36,338][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:00:36,847][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:00:37,349][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:00:37,849][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:00:38,352][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:00:38,858][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:00:39,370][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:00:39,873][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:00:40,374][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:00:40,871][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:00:41,369][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:00:41,866][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:00:42,363][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:00:42,864][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:00:43,373][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:00:43,878][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:00:44,378][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:00:44,875][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:00:45,374][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:00:45,872][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:00:46,369][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:00:46,866][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:00:47,370][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:00:47,873][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:00:48,372][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:00:48,870][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:00:49,368][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:00:49,868][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:00:50,369][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:00:50,868][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:00:51,366][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:00:51,865][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:00:52,364][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:00:52,863][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:00:53,364][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:00:53,863][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:00:54,365][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:00:54,868][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:00:55,369][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:00:55,868][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:00:56,367][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:00:56,867][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:00:57,367][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:00:57,866][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:00:58,365][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:00:58,865][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:00:59,363][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:00:59,862][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:01:00,361][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:01:00,861][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:01:01,357][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:01:01,851][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:01:02,356][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:01:02,865][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:01:03,367][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:01:03,865][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:01:04,365][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:01:04,866][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:01:05,365][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:01:05,865][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:01:06,364][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9804 tokens. +[2026-03-26 06:01:07,194][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 61.83%, ΔTime: 00:00:32 +[2026-03-26 06:01:07,952][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:01:07,955][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:01:07,956][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:01:09,809][__main__][INFO] - Iteration 521 took 2m 33s (75.85% Gen, 22.95% Train). Generation: 1m 56s, Training: 35s. Estimated remaining time: 120h 2m 39s. Estimated total time: 128h 3m 0s. Time estimates for 10 more iterations: 25m 36s, 100 more iterations: 4h 16m 6s, 500 more iterations: 21h 20m 30s. +[2026-03-26 06:01:09,811][__main__][INFO] - Starting iteration 521. +[2026-03-26 06:01:10,217][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 52 and human policies 1. +[2026-03-26 06:01:10,217][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:01:14,961][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:01:15,886][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:01:26,759][__main__][INFO] - Number of regex retries in iteration 521: 2 +[2026-03-26 06:01:26,759][__main__][INFO] - agents played in iteration 521 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:01:27,562][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:01:27,587][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:01:27,617][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:01:27,645][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:01:27,646][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:01:27,646][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:01:28,454][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:01:28,912][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:01:29,415][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:01:29,919][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:01:30,421][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:01:30,938][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:01:31,438][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:01:31,937][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:01:32,436][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:01:32,935][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:01:33,434][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:01:33,930][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:01:34,441][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:01:34,942][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:01:35,444][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:01:35,941][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:01:36,444][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:01:36,950][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:01:37,454][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:01:37,962][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:01:38,462][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:01:38,961][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:01:39,464][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:01:39,983][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:01:40,483][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:01:40,986][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:01:41,494][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:01:42,006][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:01:42,512][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:01:43,017][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:01:43,517][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:01:44,021][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:01:44,520][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:01:45,017][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:01:45,513][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:01:46,019][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:01:46,523][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:01:47,020][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:01:47,516][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:01:48,014][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:01:48,511][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:01:49,008][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:01:49,508][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:01:50,005][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:01:50,504][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:01:51,002][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:01:51,498][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:01:51,993][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:01:52,489][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:01:52,987][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:01:53,485][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:01:53,982][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:01:54,478][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:01:54,972][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:01:55,469][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:01:55,971][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:01:56,477][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:01:56,980][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:01:57,485][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:01:57,988][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:01:58,486][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:01:58,988][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:01:59,489][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:01:59,991][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:02:00,492][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9937 tokens. +[2026-03-26 06:02:01,322][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.16%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.12%, ΔTime: 00:00:32 +[2026-03-26 06:02:02,050][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:02:02,052][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:02:02,054][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:02:02,938][__main__][INFO] - Iteration 522 took 52s (31.38% Gen, 66.94% Train). Generation: 16s, Training: 35s. Estimated remaining time: 35h 54m 53s. Estimated total time: 43h 56m 7s. Time estimates for 10 more iterations: 8m 47s, 100 more iterations: 1h 27m 52s, 500 more iterations: 7h 19m 21s. +[2026-03-26 06:02:02,940][__main__][INFO] - Starting iteration 522. +[2026-03-26 06:02:03,383][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 52 and human policies 1. +[2026-03-26 06:02:03,384][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:02:18,934][__main__][INFO] - Number of regex retries in iteration 522: 0 +[2026-03-26 06:02:18,935][__main__][INFO] - agents played in iteration 522 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:02:19,730][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:02:19,754][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:02:19,777][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:02:19,801][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:02:19,802][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:02:19,802][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:02:20,713][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:02:21,175][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:02:21,685][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:02:22,186][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:02:22,687][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:02:23,192][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:02:23,698][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:02:24,199][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:02:24,699][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:02:25,204][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:02:25,706][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:02:26,211][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:02:26,723][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:02:27,224][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:02:27,727][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:02:28,226][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:02:28,728][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:02:29,224][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:02:29,726][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:02:30,229][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:02:30,724][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:02:31,222][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:02:31,721][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:02:32,218][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:02:32,716][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:02:33,214][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:02:33,712][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:02:34,209][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:02:34,705][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:02:35,208][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:02:35,709][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:02:36,207][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:02:36,705][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:02:37,206][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:02:37,702][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:02:38,210][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:02:38,729][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:02:39,227][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:02:39,723][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:02:40,222][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:02:40,724][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:02:41,220][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:02:41,723][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:02:42,225][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:02:42,733][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:02:43,239][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:02:43,735][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:02:44,233][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:02:44,730][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:02:45,226][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:02:45,723][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:02:46,218][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:02:46,719][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:02:47,214][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:02:47,717][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:02:48,214][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:02:48,711][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:02:49,207][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:02:49,703][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:02:50,201][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:02:50,701][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:02:51,205][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:02:51,712][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:02:52,211][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:02:52,712][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9932 tokens. +[2026-03-26 06:02:53,546][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.15%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.15%, ΔTime: 00:00:32 +[2026-03-26 06:02:54,304][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:02:54,307][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:02:54,309][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:02:55,084][__main__][INFO] - Iteration 523 took 51s (30.08% Gen, 68.42% Train). Generation: 15s, Training: 35s. Estimated remaining time: 35h 2m 59s. Estimated total time: 43h 5m 6s. Time estimates for 10 more iterations: 8m 37s, 100 more iterations: 1h 26m 10s, 500 more iterations: 7h 10m 51s. +[2026-03-26 06:02:55,087][__main__][INFO] - Starting iteration 523. +[2026-03-26 06:02:55,522][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 52 and human policies 1. +[2026-03-26 06:02:55,522][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:03:10,523][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:03:10,688][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:03:22,112][__main__][INFO] - Number of regex retries in iteration 523: 2 +[2026-03-26 06:03:22,113][__main__][INFO] - agents played in iteration 523 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:03:22,897][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:03:22,921][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:03:22,945][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:03:22,969][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:03:22,969][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:03:22,970][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:03:23,756][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:03:24,214][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:03:24,729][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:03:25,234][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:03:25,738][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:03:26,239][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:03:26,739][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:03:27,244][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:03:27,754][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:03:28,253][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:03:28,749][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:03:29,245][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:03:29,745][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:03:30,246][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:03:30,743][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:03:31,242][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:03:31,744][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:03:32,241][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:03:32,737][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:03:33,240][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:03:33,733][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:03:34,239][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:03:34,744][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:03:35,242][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:03:35,741][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:03:36,243][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:03:36,748][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:03:37,257][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:03:37,760][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:03:38,265][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:03:38,769][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:03:39,271][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:03:39,773][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:03:40,271][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:03:40,768][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:03:41,267][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:03:41,765][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:03:42,264][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:03:42,763][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:03:43,262][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:03:43,759][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:03:44,258][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:03:44,758][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:03:45,262][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:03:45,768][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:03:46,269][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:03:46,771][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:03:47,271][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:03:47,775][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:03:48,273][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:03:48,771][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:03:49,275][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:03:49,778][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:03:50,278][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:03:50,779][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:03:51,277][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:03:51,775][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:03:52,274][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:03:52,784][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:03:53,288][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:03:53,792][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:03:54,299][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:03:54,803][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:03:55,320][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:03:55,830][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9991 tokens. +[2026-03-26 06:03:56,713][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.18%, ΔTime: 00:00:32 +[2026-03-26 06:03:57,475][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:03:57,477][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:03:57,479][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:03:58,182][__main__][INFO] - Iteration 524 took 1m 2s (42.43% Gen, 56.44% Train). Generation: 26s, Training: 35s. Estimated remaining time: 44h 9m 56s. Estimated total time: 52h 13m 5s. Time estimates for 10 more iterations: 10m 26s, 100 more iterations: 1h 44m 26s, 500 more iterations: 8h 42m 10s. +[2026-03-26 06:03:58,184][__main__][INFO] - Starting iteration 524. +[2026-03-26 06:03:58,608][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 52 and human policies 1. +[2026-03-26 06:03:58,609][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:04:26,210][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 1 y book, 10 z balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:04:27,427][__main__][INFO] - Number of regex retries in iteration 524: 1 +[2026-03-26 06:04:27,429][__main__][INFO] - agents played in iteration 524 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:04:28,220][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:04:28,247][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:04:28,273][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:04:28,299][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:04:28,300][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:04:28,300][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:04:29,101][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:04:29,558][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:04:30,065][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:04:30,565][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:04:31,064][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:04:31,564][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:04:32,064][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:04:32,562][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:04:33,060][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:04:33,562][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:04:34,064][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:04:34,559][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:04:35,057][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:04:35,556][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:04:36,054][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:04:36,560][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:04:37,060][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:04:37,567][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:04:38,068][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:04:38,569][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:04:39,070][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:04:39,569][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:04:40,068][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:04:40,567][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:04:41,066][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:04:41,561][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:04:42,058][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:04:42,556][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:04:43,056][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:04:43,582][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:04:44,085][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:04:44,583][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:04:45,082][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:04:45,588][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:04:46,098][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:04:46,598][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:04:47,097][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:04:47,597][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:04:48,097][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:04:48,599][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:04:49,098][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:04:49,592][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:04:50,090][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:04:50,588][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:04:51,087][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:04:51,586][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:04:52,089][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:04:52,591][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:04:53,093][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:04:53,595][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:04:54,100][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:04:54,606][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:04:55,114][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:04:55,624][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:04:56,131][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:04:56,641][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:04:57,149][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:04:57,651][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:04:58,149][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:04:58,647][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:04:59,146][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:04:59,645][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:05:00,144][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:05:00,645][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:05:01,142][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9995 tokens. +[2026-03-26 06:05:01,990][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.05%, Current % of VRAM taken: 60.48%, Block Peak % of device VRAM: 62.21%, ΔTime: 00:00:32 +[2026-03-26 06:05:02,767][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:05:02,769][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:05:02,771][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:05:03,582][__main__][INFO] - Iteration 525 took 1m 4s (44.35% Gen, 54.39% Train). Generation: 28s, Training: 35s. Estimated remaining time: 46h 4m 29s. Estimated total time: 54h 8m 44s. Time estimates for 10 more iterations: 10m 49s, 100 more iterations: 1h 48m 17s, 500 more iterations: 9h 1m 27s. +[2026-03-26 06:05:03,585][__main__][INFO] - Starting iteration 525. +[2026-03-26 06:05:03,992][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 52 and human policies 1. +[2026-03-26 06:05:03,993][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:05:22,975][__main__][INFO] - Number of regex retries in iteration 525: 0 +[2026-03-26 06:05:22,976][__main__][INFO] - agents played in iteration 525 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:05:23,776][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:05:23,801][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:05:23,826][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:05:23,850][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:05:23,851][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:05:23,851][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:05:24,662][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:05:25,122][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:05:25,630][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:05:26,129][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:05:26,629][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:05:27,128][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:05:27,627][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:05:28,127][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:05:28,639][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:05:29,139][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:05:29,638][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:05:30,138][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:05:30,637][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:05:31,136][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:05:31,638][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:05:32,136][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:05:32,635][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:05:33,136][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:05:33,637][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:05:34,142][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:05:34,643][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:05:35,144][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:05:35,660][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:05:36,158][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:05:36,662][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:05:37,163][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:05:37,663][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:05:38,163][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:05:38,664][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:05:39,165][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:05:39,687][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:05:40,186][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:05:40,691][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:05:41,196][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:05:41,695][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:05:42,194][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:05:42,697][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:05:43,202][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:05:43,709][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:05:44,213][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:05:44,719][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:05:45,223][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:05:45,722][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:05:46,229][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:05:46,742][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:05:47,243][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:05:47,743][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:05:48,244][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:05:48,743][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:05:49,242][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:05:49,739][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:05:50,237][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:05:50,735][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:05:51,236][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:05:51,735][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:05:52,238][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:05:52,738][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:05:53,236][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:05:53,734][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:05:54,233][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:05:54,731][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:05:55,232][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:05:55,736][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:05:56,244][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:05:56,746][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9943 tokens. +[2026-03-26 06:05:57,593][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.12%, ΔTime: 00:00:32 +[2026-03-26 06:05:58,541][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:05:58,544][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:05:58,546][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:05:59,449][__main__][INFO] - Iteration 526 took 55s (34.23% Gen, 64.14% Train). Generation: 18s, Training: 35s. Estimated remaining time: 38h 7m 43s. Estimated total time: 46h 12m 53s. Time estimates for 10 more iterations: 9m 14s, 100 more iterations: 1h 32m 25s, 500 more iterations: 7h 42m 8s. +[2026-03-26 06:05:59,453][__main__][INFO] - Starting iteration 526. +[2026-03-26 06:05:59,852][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 52 and human policies 1. +[2026-03-26 06:05:59,853][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:06:08,573][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:06:18,518][__main__][INFO] - Number of regex retries in iteration 526: 1 +[2026-03-26 06:06:18,519][__main__][INFO] - agents played in iteration 526 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:06:19,298][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:06:19,322][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:06:19,346][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:06:19,370][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:06:19,371][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:06:19,371][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:06:20,162][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:06:20,614][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:06:21,123][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:06:21,626][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:06:22,127][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:06:22,628][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:06:23,131][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:06:23,629][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:06:24,130][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:06:24,636][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:06:25,140][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:06:25,645][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:06:26,152][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:06:26,656][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:06:27,157][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:06:27,660][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:06:28,190][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:06:28,688][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:06:29,189][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:06:29,692][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:06:30,196][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:06:30,697][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:06:31,198][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:06:31,698][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:06:32,201][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:06:32,704][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:06:33,205][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:06:33,707][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:06:34,210][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:06:34,713][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:06:35,217][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:06:35,717][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:06:36,225][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:06:36,724][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:06:37,223][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:06:37,721][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:06:38,219][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:06:38,717][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:06:39,216][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:06:39,715][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:06:40,213][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:06:40,710][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:06:41,215][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:06:41,713][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:06:42,216][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:06:42,714][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:06:43,211][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:06:43,709][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:06:44,209][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:06:44,709][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:06:45,214][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:06:45,718][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:06:46,219][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:06:46,716][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:06:47,216][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:06:47,715][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:06:48,216][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:06:48,716][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:06:49,218][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:06:49,719][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:06:50,224][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:06:50,733][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:06:51,247][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:06:51,752][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:06:52,257][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9941 tokens. +[2026-03-26 06:06:53,099][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 06:06:53,839][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:06:53,841][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:06:53,843][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:06:54,569][__main__][INFO] - Iteration 527 took 54s (34.11% Gen, 64.56% Train). Generation: 18s, Training: 35s. Estimated remaining time: 37h 29m 46s. Estimated total time: 45h 35m 52s. Time estimates for 10 more iterations: 9m 7s, 100 more iterations: 1h 31m 11s, 500 more iterations: 7h 35m 58s. +[2026-03-26 06:06:54,571][__main__][INFO] - Starting iteration 527. +[2026-03-26 06:06:54,983][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 52 and human policies 1. +[2026-03-26 06:06:54,984][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:07:09,756][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:07:10,643][__main__][INFO] - Number of regex retries in iteration 527: 1 +[2026-03-26 06:07:10,643][__main__][INFO] - agents played in iteration 527 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:07:11,433][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:07:11,457][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:07:11,481][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:07:11,504][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:07:11,505][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:07:11,506][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:07:12,304][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:07:12,757][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:07:13,257][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:07:13,753][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:07:14,250][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:07:14,746][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:07:15,247][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:07:15,754][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:07:16,252][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:07:16,752][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:07:17,253][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:07:17,754][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:07:18,254][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:07:18,751][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:07:19,252][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:07:19,749][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:07:20,249][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:07:20,746][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:07:21,242][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:07:21,737][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:07:22,234][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:07:22,733][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:07:23,234][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:07:23,731][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:07:24,227][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:07:24,722][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:07:25,221][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:07:25,719][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:07:26,219][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:07:26,714][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:07:27,215][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:07:27,721][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:07:28,218][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:07:28,718][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:07:29,216][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:07:29,721][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:07:30,221][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:07:30,719][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:07:31,218][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:07:31,716][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:07:32,213][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:07:32,712][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:07:33,210][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:07:33,708][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:07:34,207][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:07:34,706][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:07:35,204][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:07:35,703][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:07:36,205][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:07:36,704][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:07:37,203][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:07:37,724][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:07:38,221][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:07:38,719][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:07:39,217][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:07:39,716][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:07:40,214][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:07:40,715][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:07:41,213][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:07:41,711][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:07:42,208][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:07:42,705][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:07:43,206][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:07:43,704][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:07:44,208][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9862 tokens. +[2026-03-26 06:07:45,081][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.15%, Current % of VRAM taken: 60.58%, Block Peak % of device VRAM: 62.11%, ΔTime: 00:00:32 +[2026-03-26 06:07:45,845][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:07:45,847][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:07:45,849][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:07:46,582][__main__][INFO] - Iteration 528 took 51s (30.35% Gen, 68.23% Train). Generation: 15s, Training: 35s. Estimated remaining time: 34h 52m 59s. Estimated total time: 42h 59m 57s. Time estimates for 10 more iterations: 8m 35s, 100 more iterations: 1h 25m 59s, 500 more iterations: 7h 9m 59s. +[2026-03-26 06:07:46,584][__main__][INFO] - Starting iteration 528. +[2026-03-26 06:07:46,990][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 52 and human policies 1. +[2026-03-26 06:07:46,990][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:08:02,555][__main__][INFO] - Number of regex retries in iteration 528: 0 +[2026-03-26 06:08:02,555][__main__][INFO] - agents played in iteration 528 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:08:03,329][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:08:03,353][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:08:03,377][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:08:03,401][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:08:03,402][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:08:03,402][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:08:04,200][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:08:04,655][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:08:05,158][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:08:05,662][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:08:06,160][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:08:06,659][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:08:07,155][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:08:07,652][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:08:08,153][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:08:08,661][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:08:09,169][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:08:09,692][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:08:10,199][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:08:10,702][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:08:11,205][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:08:11,707][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:08:12,212][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:08:12,724][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:08:13,224][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:08:13,726][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:08:14,229][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:08:14,729][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:08:15,230][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:08:15,729][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:08:16,228][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:08:16,738][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:08:17,243][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:08:17,745][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:08:18,245][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:08:18,744][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:08:19,243][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:08:19,742][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:08:20,242][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:08:20,742][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:08:21,244][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:08:21,750][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:08:22,254][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:08:22,752][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:08:23,251][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:08:23,749][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:08:24,249][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:08:24,753][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:08:25,261][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:08:25,760][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:08:26,262][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:08:26,765][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:08:27,267][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:08:27,769][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:08:28,270][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:08:28,768][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:08:29,268][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:08:29,770][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:08:30,275][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:08:30,775][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:08:31,275][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:08:31,775][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:08:32,275][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:08:32,774][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:08:33,274][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:08:33,774][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:08:34,274][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:08:34,776][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:08:35,279][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:08:35,788][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:08:36,291][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10007 tokens. +[2026-03-26 06:08:37,154][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 62.29%, ΔTime: 00:00:32 +[2026-03-26 06:08:37,897][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:08:37,900][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:08:37,902][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:08:38,620][__main__][INFO] - Iteration 529 took 51s (30.15% Gen, 68.46% Train). Generation: 15s, Training: 35s. Estimated remaining time: 34h 53m 43s. Estimated total time: 43h 1m 33s. Time estimates for 10 more iterations: 8m 36s, 100 more iterations: 1h 26m 3s, 500 more iterations: 7h 10m 15s. +[2026-03-26 06:08:38,622][__main__][INFO] - Starting iteration 529. +[2026-03-26 06:08:39,027][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 52 and human policies 1. +[2026-03-26 06:08:39,028][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:08:56,963][__main__][INFO] - Number of regex retries in iteration 529: 0 +[2026-03-26 06:08:56,964][__main__][INFO] - agents played in iteration 529 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:08:57,743][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:08:57,767][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:08:57,790][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:08:57,814][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:08:57,815][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:08:57,815][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:08:58,616][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:08:59,068][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:08:59,568][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:09:00,071][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:09:00,566][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:09:01,063][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:09:01,561][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:09:02,059][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:09:02,561][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:09:03,064][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:09:03,563][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:09:04,062][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:09:04,562][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:09:05,062][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:09:05,561][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:09:06,060][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:09:06,558][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:09:07,059][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:09:07,559][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:09:08,059][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:09:08,563][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:09:09,063][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:09:09,569][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:09:10,082][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:09:10,583][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:09:11,082][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:09:11,579][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:09:12,078][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:09:12,574][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:09:13,070][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:09:13,567][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:09:14,063][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:09:14,558][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:09:15,056][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:09:15,554][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:09:16,052][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:09:16,548][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:09:17,045][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:09:17,544][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:09:18,042][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:09:18,539][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:09:19,035][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:09:19,532][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:09:20,038][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:09:20,546][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:09:21,044][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:09:21,543][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:09:22,043][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:09:22,545][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:09:23,049][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:09:23,550][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:09:24,051][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:09:24,551][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:09:25,050][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:09:25,550][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:09:26,049][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:09:26,548][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:09:27,047][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:09:27,546][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:09:28,054][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:09:28,563][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:09:29,076][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:09:29,579][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:09:30,081][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:09:30,586][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9917 tokens. +[2026-03-26 06:09:31,486][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.18%, ΔTime: 00:00:32 +[2026-03-26 06:09:32,301][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:09:32,304][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:09:32,305][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:09:33,049][__main__][INFO] - Iteration 530 took 54s (33.20% Gen, 65.42% Train). Generation: 17s, Training: 35s. Estimated remaining time: 36h 52m 23s. Estimated total time: 45h 1m 7s. Time estimates for 10 more iterations: 9m 0s, 100 more iterations: 1h 30m 2s, 500 more iterations: 7h 30m 11s. +[2026-03-26 06:09:33,051][__main__][INFO] - Starting iteration 530. +[2026-03-26 06:09:33,464][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 52 and human policies 1. +[2026-03-26 06:09:33,465][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:09:48,541][__main__][INFO] - Number of regex retries in iteration 530: 0 +[2026-03-26 06:09:48,542][__main__][INFO] - agents played in iteration 530 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:09:49,326][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:09:49,350][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:09:49,374][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:09:49,398][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:09:49,398][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:09:49,399][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:09:50,192][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:09:50,676][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:09:51,198][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:09:51,698][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:09:52,207][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:09:52,707][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:09:53,208][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:09:53,707][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:09:54,220][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:09:54,722][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:09:55,224][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:09:55,726][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:09:56,227][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:09:56,727][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:09:57,241][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:09:57,742][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:09:58,243][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:09:58,744][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:09:59,243][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:09:59,745][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:10:00,241][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:10:00,740][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:10:01,236][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:10:01,731][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:10:02,228][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:10:02,726][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:10:03,225][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:10:03,722][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:10:04,222][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:10:04,729][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:10:05,229][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:10:05,732][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:10:06,230][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:10:06,727][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:10:07,227][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:10:07,732][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:10:08,238][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:10:08,736][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:10:09,236][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:10:09,734][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:10:10,233][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:10:10,731][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:10:11,228][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:10:11,730][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:10:12,233][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:10:12,736][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:10:13,238][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:10:13,743][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:10:14,244][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:10:14,744][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:10:15,244][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:10:15,744][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:10:16,256][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:10:16,756][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:10:17,261][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:10:17,761][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:10:18,267][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:10:18,767][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:10:19,266][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:10:19,765][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:10:20,268][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:10:20,772][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:10:21,279][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:10:21,784][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:10:22,288][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9959 tokens. +[2026-03-26 06:10:23,156][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.14%, ΔTime: 00:00:32 +[2026-03-26 06:10:23,928][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:10:23,930][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:10:23,932][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:10:25,543][__main__][INFO] - Iteration 531 took 52s (28.95% Gen, 67.95% Train). Generation: 15s, Training: 35s. Estimated remaining time: 35h 14m 19s. Estimated total time: 43h 23m 56s. Time estimates for 10 more iterations: 8m 40s, 100 more iterations: 1h 26m 47s, 500 more iterations: 7h 13m 59s. +[2026-03-26 06:10:25,545][__main__][INFO] - Starting iteration 531. +[2026-03-26 06:10:25,959][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 53 and human policies 1. +[2026-03-26 06:10:25,959][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:10:40,675][__main__][INFO] - Number of regex retries in iteration 531: 0 +[2026-03-26 06:10:40,676][__main__][INFO] - agents played in iteration 531 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:10:41,443][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:10:41,469][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:10:41,494][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:10:41,520][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:10:41,520][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:10:41,521][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:10:42,372][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:10:42,831][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:10:43,333][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:10:43,830][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:10:44,331][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:10:44,839][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:10:45,343][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:10:45,843][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:10:46,340][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:10:46,840][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:10:47,338][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:10:47,841][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:10:48,353][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:10:48,853][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:10:49,354][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:10:49,854][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:10:50,354][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:10:50,853][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:10:51,358][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:10:51,857][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:10:52,356][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:10:52,854][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:10:53,352][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:10:53,848][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:10:54,350][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:10:54,848][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:10:55,345][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:10:55,841][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:10:56,340][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:10:56,838][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:10:57,335][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:10:57,835][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:10:58,341][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:10:58,843][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:10:59,348][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:10:59,871][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:11:00,371][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:11:00,870][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:11:01,370][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:11:01,882][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:11:02,385][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:11:02,885][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:11:03,383][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:11:03,886][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:11:04,390][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:11:04,892][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:11:05,395][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:11:05,899][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:11:06,410][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:11:06,910][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:11:07,411][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:11:07,912][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:11:08,411][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:11:08,910][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:11:09,408][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:11:09,906][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:11:10,406][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:11:10,917][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:11:11,417][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:11:11,917][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:11:12,416][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:11:12,914][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:11:13,412][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:11:13,911][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:11:14,411][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9906 tokens. +[2026-03-26 06:11:15,242][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.19%, ΔTime: 00:00:32 +[2026-03-26 06:11:16,010][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:11:16,013][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:11:16,015][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:11:16,785][__main__][INFO] - Iteration 532 took 50s (28.95% Gen, 69.53% Train). Generation: 14s, Training: 35s. Estimated remaining time: 34h 10m 53s. Estimated total time: 42h 21m 21s. Time estimates for 10 more iterations: 8m 28s, 100 more iterations: 1h 24m 42s, 500 more iterations: 7h 3m 33s. +[2026-03-26 06:11:16,788][__main__][INFO] - Starting iteration 532. +[2026-03-26 06:11:17,194][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 53 and human policies 1. +[2026-03-26 06:11:17,195][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:11:32,390][__main__][INFO] - Number of regex retries in iteration 532: 0 +[2026-03-26 06:11:32,391][__main__][INFO] - agents played in iteration 532 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:11:33,162][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:11:33,186][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:11:33,210][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:11:33,233][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:11:33,234][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:11:33,234][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:11:34,031][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:11:34,487][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:11:34,992][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:11:35,492][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:11:35,992][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:11:36,493][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:11:36,991][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:11:37,489][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:11:37,987][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:11:38,486][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:11:38,985][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:11:39,487][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:11:39,983][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:11:40,477][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:11:40,974][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:11:41,478][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:11:41,980][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:11:42,476][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:11:42,971][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:11:43,469][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:11:43,969][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:11:44,467][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:11:44,966][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:11:45,465][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:11:45,964][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:11:46,463][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:11:46,963][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:11:47,461][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:11:47,962][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:11:48,461][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:11:48,959][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:11:49,456][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:11:49,953][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:11:50,452][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:11:50,947][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:11:51,443][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:11:51,940][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:11:52,439][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:11:52,934][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:11:53,431][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:11:53,926][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:11:54,424][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:11:54,920][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:11:55,421][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:11:55,926][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:11:56,440][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:11:56,942][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:11:57,441][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:11:57,939][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:11:58,439][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:11:58,939][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:11:59,437][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:11:59,936][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:12:00,435][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:12:00,934][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:12:01,434][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:12:01,933][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:12:02,435][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:12:02,931][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:12:03,426][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:12:03,924][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:12:04,428][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:12:04,939][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:12:05,447][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:12:05,954][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9855 tokens. +[2026-03-26 06:12:06,823][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 61.90%, ΔTime: 00:00:32 +[2026-03-26 06:12:07,606][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:12:07,609][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:12:07,610][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:12:08,415][__main__][INFO] - Iteration 533 took 51s (29.67% Gen, 68.76% Train). Generation: 15s, Training: 35s. Estimated remaining time: 34h 29m 44s. Estimated total time: 42h 41m 4s. Time estimates for 10 more iterations: 8m 32s, 100 more iterations: 1h 25m 22s, 500 more iterations: 7h 6m 50s. +[2026-03-26 06:12:08,417][__main__][INFO] - Starting iteration 533. +[2026-03-26 06:12:08,827][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 53 and human policies 1. +[2026-03-26 06:12:08,828][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:12:25,246][__main__][INFO] - Number of regex retries in iteration 533: 0 +[2026-03-26 06:12:25,247][__main__][INFO] - agents played in iteration 533 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:12:26,007][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:12:26,030][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:12:26,054][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:12:26,077][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:12:26,078][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:12:26,078][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:12:26,857][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:12:27,312][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:12:27,828][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:12:28,328][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:12:28,833][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:12:29,332][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:12:29,830][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:12:30,328][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:12:30,827][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:12:31,325][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:12:31,825][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:12:32,322][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:12:32,818][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:12:33,316][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:12:33,813][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:12:34,324][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:12:34,823][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:12:35,325][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:12:35,821][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:12:36,326][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:12:36,828][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:12:37,342][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:12:37,846][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:12:38,346][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:12:38,847][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:12:39,347][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:12:39,847][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:12:40,347][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:12:40,848][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:12:41,347][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:12:41,854][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:12:42,365][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:12:42,865][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:12:43,366][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:12:43,873][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:12:44,373][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:12:44,873][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:12:45,372][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:12:45,874][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:12:46,374][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:12:46,874][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:12:47,373][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:12:47,869][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:12:48,366][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:12:48,862][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:12:49,360][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:12:49,871][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:12:50,372][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:12:50,871][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:12:51,369][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:12:51,899][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:12:52,403][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:12:52,905][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:12:53,403][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:12:53,914][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:12:54,415][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:12:54,919][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:12:55,423][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:12:55,933][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:12:56,437][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:12:56,941][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:12:57,447][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:12:57,951][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:12:58,459][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:12:58,961][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9893 tokens. +[2026-03-26 06:12:59,810][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.97%, Current % of VRAM taken: 60.40%, Block Peak % of device VRAM: 62.12%, ΔTime: 00:00:32 +[2026-03-26 06:13:00,562][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:13:00,565][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:13:00,567][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:13:01,337][__main__][INFO] - Iteration 534 took 52s (31.27% Gen, 67.26% Train). Generation: 16s, Training: 35s. Estimated remaining time: 35h 33m 18s. Estimated total time: 43h 45m 31s. Time estimates for 10 more iterations: 8m 45s, 100 more iterations: 1h 27m 31s, 500 more iterations: 7h 17m 35s. +[2026-03-26 06:13:01,339][__main__][INFO] - Starting iteration 534. +[2026-03-26 06:13:01,748][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 53 and human policies 1. +[2026-03-26 06:13:01,749][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:13:17,722][__main__][INFO] - Number of regex retries in iteration 534: 0 +[2026-03-26 06:13:17,723][__main__][INFO] - agents played in iteration 534 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:13:18,494][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:13:18,519][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:13:18,544][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:13:18,569][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:13:18,569][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:13:18,570][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:13:19,383][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:13:19,839][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:13:20,348][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:13:20,848][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:13:21,348][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:13:21,848][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:13:22,353][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:13:22,860][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:13:23,367][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:13:23,865][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:13:24,362][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:13:24,860][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:13:25,358][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:13:25,856][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:13:26,353][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:13:26,851][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:13:27,349][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:13:27,848][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:13:28,345][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:13:28,843][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:13:29,341][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:13:29,838][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:13:30,334][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:13:30,832][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:13:31,331][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:13:31,830][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:13:32,329][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:13:32,831][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:13:33,334][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:13:33,836][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:13:34,336][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:13:34,835][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:13:35,334][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:13:35,835][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:13:36,336][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:13:36,836][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:13:37,339][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:13:37,843][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:13:38,350][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:13:38,853][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:13:39,359][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:13:39,858][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:13:40,357][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:13:40,854][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:13:41,352][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:13:41,850][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:13:42,347][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:13:42,844][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:13:43,342][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:13:43,837][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:13:44,333][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:13:44,830][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:13:45,326][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:13:45,823][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:13:46,321][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:13:46,819][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:13:47,321][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:13:47,825][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:13:48,332][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:13:48,839][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:13:49,352][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:13:49,854][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:13:50,357][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:13:50,861][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:13:51,369][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9981 tokens. +[2026-03-26 06:13:52,228][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.42%, ΔTime: 00:00:32 +[2026-03-26 06:13:52,984][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:13:52,989][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:13:52,993][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:13:53,719][__main__][INFO] - Iteration 535 took 51s (30.74% Gen, 67.86% Train). Generation: 15s, Training: 35s. Estimated remaining time: 35h 5m 30s. Estimated total time: 43h 18m 35s. Time estimates for 10 more iterations: 8m 39s, 100 more iterations: 1h 26m 37s, 500 more iterations: 7h 13m 5s. +[2026-03-26 06:13:53,722][__main__][INFO] - Starting iteration 535. +[2026-03-26 06:13:54,123][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 53 and human policies 1. +[2026-03-26 06:13:54,124][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:14:11,146][__main__][INFO] - Number of regex retries in iteration 535: 0 +[2026-03-26 06:14:11,147][__main__][INFO] - agents played in iteration 535 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:14:11,912][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:14:11,938][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:14:11,963][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:14:11,988][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:14:11,989][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:14:11,989][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:14:12,787][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:14:13,243][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:14:13,758][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:14:14,260][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:14:14,759][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:14:15,261][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:14:15,757][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:14:16,254][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:14:16,750][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:14:17,249][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:14:17,745][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:14:18,243][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:14:18,746][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:14:19,243][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:14:19,739][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:14:20,240][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:14:20,738][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:14:21,235][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:14:21,731][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:14:22,228][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:14:22,749][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:14:23,251][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:14:23,748][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:14:24,246][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:14:24,743][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:14:25,242][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:14:25,742][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:14:26,244][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:14:26,746][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:14:27,250][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:14:27,752][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:14:28,250][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:14:28,749][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:14:29,244][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:14:29,742][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:14:30,239][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:14:30,736][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:14:31,234][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:14:31,729][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:14:32,227][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:14:32,725][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:14:33,228][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:14:33,729][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:14:34,227][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:14:34,728][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:14:35,226][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:14:35,725][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:14:36,224][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:14:36,724][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:14:37,224][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:14:37,728][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:14:38,232][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:14:38,737][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:14:39,243][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:14:39,743][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:14:40,247][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:14:40,750][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:14:41,254][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:14:41,755][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:14:42,260][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:14:42,766][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:14:43,270][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:14:43,775][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:14:44,280][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:14:44,783][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9905 tokens. +[2026-03-26 06:14:45,590][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.16%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.13%, ΔTime: 00:00:32 +[2026-03-26 06:14:46,339][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:14:46,341][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:14:46,343][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:14:47,073][__main__][INFO] - Iteration 536 took 52s (32.15% Gen, 66.47% Train). Generation: 17s, Training: 35s. Estimated remaining time: 35h 53m 32s. Estimated total time: 44h 7m 30s. Time estimates for 10 more iterations: 8m 49s, 100 more iterations: 1h 28m 15s, 500 more iterations: 7h 21m 15s. +[2026-03-26 06:14:47,075][__main__][INFO] - Starting iteration 536. +[2026-03-26 06:14:47,488][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 53 and human policies 1. +[2026-03-26 06:14:47,489][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:14:52,630][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:15:04,166][__main__][INFO] - Number of regex retries in iteration 536: 1 +[2026-03-26 06:15:04,167][__main__][INFO] - agents played in iteration 536 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:15:04,941][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:15:04,967][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:15:04,992][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:15:05,017][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:15:05,018][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:15:05,018][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:15:05,806][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:15:06,269][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:15:06,772][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:15:07,274][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:15:07,773][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:15:08,273][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:15:08,785][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:15:09,287][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:15:09,788][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:15:10,291][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:15:10,794][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:15:11,295][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:15:11,796][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:15:12,296][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:15:12,795][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:15:13,295][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:15:13,793][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:15:14,292][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:15:14,788][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:15:15,294][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:15:15,796][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:15:16,293][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:15:16,790][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:15:17,288][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:15:17,822][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:15:18,322][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:15:18,829][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:15:19,324][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:15:19,818][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:15:20,319][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:15:20,818][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:15:21,320][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:15:21,816][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:15:22,316][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:15:22,815][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:15:23,314][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:15:23,813][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:15:24,312][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:15:24,812][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:15:25,320][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:15:25,825][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:15:26,326][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:15:26,828][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:15:27,330][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:15:27,837][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:15:28,336][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:15:28,833][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:15:29,346][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:15:29,848][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:15:30,358][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:15:30,862][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:15:31,364][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:15:31,866][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:15:32,374][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:15:32,875][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:15:33,382][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:15:33,894][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:15:34,401][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:15:34,908][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:15:35,411][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:15:35,913][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:15:36,415][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:15:36,914][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:15:37,414][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:15:37,918][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9859 tokens. +[2026-03-26 06:15:38,759][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 06:15:39,498][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:15:39,500][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:15:39,502][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:15:40,272][__main__][INFO] - Iteration 537 took 52s (31.60% Gen, 66.94% Train). Generation: 16s, Training: 35s. Estimated remaining time: 35h 44m 24s. Estimated total time: 43h 59m 16s. Time estimates for 10 more iterations: 8m 47s, 100 more iterations: 1h 27m 58s, 500 more iterations: 7h 19m 52s. +[2026-03-26 06:15:40,275][__main__][INFO] - Starting iteration 537. +[2026-03-26 06:15:40,685][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 53 and human policies 1. +[2026-03-26 06:15:40,686][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:15:56,269][__main__][INFO] - Number of regex retries in iteration 537: 0 +[2026-03-26 06:15:56,270][__main__][INFO] - agents played in iteration 537 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:15:57,038][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:15:57,062][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:15:57,086][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:15:57,111][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:15:57,111][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:15:57,112][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:15:57,894][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:15:58,348][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:15:58,848][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:15:59,346][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:15:59,849][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:16:00,360][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:16:00,859][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:16:01,364][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:16:01,862][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:16:02,360][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:16:02,863][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:16:03,365][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:16:03,861][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:16:04,358][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:16:04,855][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:16:05,354][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:16:05,851][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:16:06,351][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:16:06,854][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:16:07,353][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:16:07,853][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:16:08,359][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:16:08,861][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:16:09,372][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:16:09,869][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:16:10,371][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:16:10,869][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:16:11,367][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:16:11,865][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:16:12,365][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:16:12,864][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:16:13,364][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:16:13,863][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:16:14,362][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:16:14,863][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:16:15,360][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:16:15,855][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:16:16,354][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:16:16,852][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:16:17,351][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:16:17,852][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:16:18,351][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:16:18,854][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:16:19,361][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:16:19,860][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:16:20,362][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:16:20,870][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:16:21,372][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:16:21,874][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:16:22,379][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:16:22,905][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:16:23,412][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:16:23,918][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:16:24,422][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:16:24,928][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:16:25,432][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:16:25,937][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:16:26,446][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:16:26,957][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:16:27,459][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:16:27,959][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:16:28,460][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:16:28,966][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:16:29,467][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:16:29,969][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9865 tokens. +[2026-03-26 06:16:30,791][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.11%, ΔTime: 00:00:32 +[2026-03-26 06:16:31,557][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:16:31,559][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:16:31,561][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:16:32,284][__main__][INFO] - Iteration 538 took 51s (30.20% Gen, 68.39% Train). Generation: 15s, Training: 35s. Estimated remaining time: 34h 44m 17s. Estimated total time: 43h 0m 1s. Time estimates for 10 more iterations: 8m 36s, 100 more iterations: 1h 26m 0s, 500 more iterations: 7h 10m 0s. +[2026-03-26 06:16:32,287][__main__][INFO] - Starting iteration 538. +[2026-03-26 06:16:32,687][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 53 and human policies 1. +[2026-03-26 06:16:32,688][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:16:49,072][__main__][INFO] - Number of regex retries in iteration 538: 0 +[2026-03-26 06:16:49,072][__main__][INFO] - agents played in iteration 538 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:16:49,853][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:16:49,877][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:16:49,901][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:16:49,925][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:16:49,926][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:16:49,927][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:16:50,717][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:16:51,171][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:16:51,673][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:16:52,177][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:16:52,681][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:16:53,187][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:16:53,686][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:16:54,186][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:16:54,686][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:16:55,183][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:16:55,683][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:16:56,176][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:16:56,672][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:16:57,170][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:16:57,668][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:16:58,166][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:16:58,666][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:16:59,171][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:16:59,672][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:17:00,173][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:17:00,669][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:17:01,164][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:17:01,661][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:17:02,165][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:17:02,664][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:17:03,169][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:17:03,666][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:17:04,166][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:17:04,663][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:17:05,158][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:17:05,655][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:17:06,150][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:17:06,646][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:17:07,146][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:17:07,646][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:17:08,146][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:17:08,646][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:17:09,147][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:17:09,652][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:17:10,152][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:17:10,655][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:17:11,154][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:17:11,656][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:17:12,157][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:17:12,654][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:17:13,153][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:17:13,655][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:17:14,157][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:17:14,657][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:17:15,158][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:17:15,657][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:17:16,164][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:17:16,666][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:17:17,169][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:17:17,676][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:17:18,186][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:17:18,686][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:17:19,183][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:17:19,681][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:17:20,179][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:17:20,678][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:17:21,177][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:17:21,675][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:17:22,177][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:17:22,672][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9862 tokens. +[2026-03-26 06:17:23,506][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.91%, Current % of VRAM taken: 60.34%, Block Peak % of device VRAM: 62.19%, ΔTime: 00:00:32 +[2026-03-26 06:17:24,246][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:17:24,248][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:17:24,250][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:17:25,109][__main__][INFO] - Iteration 539 took 52s (31.25% Gen, 67.10% Train). Generation: 16s, Training: 35s. Estimated remaining time: 35h 24m 30s. Estimated total time: 43h 41m 6s. Time estimates for 10 more iterations: 8m 44s, 100 more iterations: 1h 27m 22s, 500 more iterations: 7h 16m 51s. +[2026-03-26 06:17:25,111][__main__][INFO] - Starting iteration 539. +[2026-03-26 06:17:25,519][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 53 and human policies 1. +[2026-03-26 06:17:25,519][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:17:29,342][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:17:40,542][__main__][INFO] - Number of regex retries in iteration 539: 1 +[2026-03-26 06:17:40,543][__main__][INFO] - agents played in iteration 539 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:17:41,313][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:17:41,336][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:17:41,360][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:17:41,383][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:17:41,384][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:17:41,384][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:17:42,174][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:17:42,625][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:17:43,132][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:17:43,636][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:17:44,133][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:17:44,631][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:17:45,128][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:17:45,626][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:17:46,125][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:17:46,623][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:17:47,120][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:17:47,619][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:17:48,119][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:17:48,618][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:17:49,113][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:17:49,612][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:17:50,129][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:17:50,633][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:17:51,147][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:17:51,646][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:17:52,148][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:17:52,653][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:17:53,150][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:17:53,648][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:17:54,146][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:17:54,643][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:17:55,140][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:17:55,636][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:17:56,138][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:17:56,633][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:17:57,131][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:17:57,627][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:17:58,123][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:17:58,619][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:17:59,116][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:17:59,613][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:18:00,111][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:18:00,608][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:18:01,107][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:18:01,605][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:18:02,102][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:18:02,598][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:18:03,094][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:18:03,598][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:18:04,108][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:18:04,613][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:18:05,120][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:18:05,622][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:18:06,126][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:18:06,628][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:18:07,132][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:18:07,636][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:18:08,137][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:18:08,639][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:18:09,140][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:18:09,637][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:18:10,136][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:18:10,634][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:18:11,129][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:18:11,627][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:18:12,123][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:18:12,624][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:18:13,120][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:18:13,617][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:18:14,113][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9839 tokens. +[2026-03-26 06:18:14,945][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.97%, Current % of VRAM taken: 60.39%, Block Peak % of device VRAM: 62.15%, ΔTime: 00:00:32 +[2026-03-26 06:18:15,679][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:18:15,681][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:18:15,683][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:18:16,388][__main__][INFO] - Iteration 540 took 50s (29.53% Gen, 69.08% Train). Generation: 15s, Training: 35s. Estimated remaining time: 34h 6m 4s. Estimated total time: 42h 23m 31s. Time estimates for 10 more iterations: 8m 28s, 100 more iterations: 1h 24m 47s, 500 more iterations: 7h 3m 55s. +[2026-03-26 06:18:16,390][__main__][INFO] - Starting iteration 540. +[2026-03-26 06:18:16,803][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 53 and human policies 1. +[2026-03-26 06:18:16,804][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:18:33,402][__main__][INFO] - Number of regex retries in iteration 540: 0 +[2026-03-26 06:18:33,403][__main__][INFO] - agents played in iteration 540 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:18:34,171][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:18:34,194][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:18:34,218][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:18:34,241][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:18:34,242][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:18:34,242][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:18:35,016][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:18:35,481][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:18:35,988][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:18:36,487][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:18:36,983][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:18:37,477][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:18:37,981][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:18:38,491][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:18:38,993][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:18:39,493][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:18:39,991][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:18:40,489][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:18:41,003][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:18:41,509][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:18:42,017][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:18:42,517][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:18:43,015][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:18:43,513][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:18:44,022][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:18:44,523][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:18:45,021][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:18:45,518][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:18:46,017][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:18:46,516][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:18:47,012][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:18:47,512][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:18:48,010][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:18:48,509][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:18:49,011][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:18:49,509][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:18:50,006][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:18:50,512][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:18:51,016][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:18:51,513][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:18:52,014][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:18:52,512][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:18:53,009][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:18:53,507][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:18:54,010][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:18:54,523][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:18:55,031][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:18:55,531][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:18:56,039][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:18:56,541][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:18:57,042][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:18:57,547][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:18:58,053][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:18:58,564][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:18:59,073][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:18:59,578][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:19:00,080][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:19:00,582][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:19:01,078][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:19:01,575][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:19:02,074][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:19:02,585][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:19:03,085][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:19:03,583][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:19:04,088][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:19:04,590][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:19:05,090][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:19:05,589][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:19:06,089][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:19:06,592][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:19:07,094][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9771 tokens. +[2026-03-26 06:19:07,921][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.00%, Current % of VRAM taken: 60.43%, Block Peak % of device VRAM: 62.06%, ΔTime: 00:00:32 +[2026-03-26 06:19:08,687][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:19:08,689][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:19:08,691][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:19:10,002][__main__][INFO] - Iteration 541 took 53s (31.20% Gen, 66.33% Train). Generation: 16s, Training: 35s. Estimated remaining time: 36h 1m 39s. Estimated total time: 44h 20m 0s. Time estimates for 10 more iterations: 8m 52s, 100 more iterations: 1h 28m 40s, 500 more iterations: 7h 23m 20s. +[2026-03-26 06:19:10,005][__main__][INFO] - Starting iteration 541. +[2026-03-26 06:19:10,406][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 54 and human policies 1. +[2026-03-26 06:19:10,406][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:19:14,276][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:19:24,910][__main__][INFO] - Number of regex retries in iteration 541: 1 +[2026-03-26 06:19:24,911][__main__][INFO] - agents played in iteration 541 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:19:25,669][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:19:25,693][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:19:25,716][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:19:25,740][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:19:25,740][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:19:25,741][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:19:26,549][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:19:27,008][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:19:27,513][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:19:28,012][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:19:28,524][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:19:29,025][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:19:29,520][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:19:30,017][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:19:30,518][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:19:31,016][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:19:31,516][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:19:32,013][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:19:32,527][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:19:33,028][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:19:33,529][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:19:34,028][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:19:34,527][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:19:35,031][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:19:35,576][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:19:36,082][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:19:36,584][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:19:37,087][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:19:37,588][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:19:38,087][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:19:38,591][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:19:39,100][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:19:39,602][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:19:40,100][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:19:40,598][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:19:41,096][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:19:41,595][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:19:42,091][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:19:42,593][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:19:43,097][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:19:43,597][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:19:44,097][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:19:44,593][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:19:45,092][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:19:45,589][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:19:46,089][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:19:46,601][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:19:47,115][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:19:47,618][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:19:48,122][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:19:48,627][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:19:49,132][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:19:49,636][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:19:50,140][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:19:50,643][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:19:51,146][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:19:51,648][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:19:52,150][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:19:52,654][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:19:53,155][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:19:53,656][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:19:54,158][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:19:54,660][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:19:55,163][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:19:55,663][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:19:56,161][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:19:56,659][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:19:57,155][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:19:57,651][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:19:58,148][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:19:58,650][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9864 tokens. +[2026-03-26 06:19:59,490][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.01%, Current % of VRAM taken: 60.44%, Block Peak % of device VRAM: 62.15%, ΔTime: 00:00:32 +[2026-03-26 06:20:00,228][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:20:00,231][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:20:00,232][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:20:00,988][__main__][INFO] - Iteration 542 took 50s (28.67% Gen, 69.83% Train). Generation: 14s, Training: 35s. Estimated remaining time: 33h 49m 56s. Estimated total time: 42h 9m 8s. Time estimates for 10 more iterations: 8m 25s, 100 more iterations: 1h 24m 18s, 500 more iterations: 7h 1m 31s. +[2026-03-26 06:20:00,990][__main__][INFO] - Starting iteration 542. +[2026-03-26 06:20:01,389][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 54 and human policies 1. +[2026-03-26 06:20:01,390][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:20:16,633][__main__][INFO] - Number of regex retries in iteration 542: 0 +[2026-03-26 06:20:16,634][__main__][INFO] - agents played in iteration 542 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:20:17,407][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:20:17,433][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:20:17,458][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:20:17,483][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:20:17,484][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:20:17,485][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:20:18,288][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:20:18,744][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:20:19,244][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:20:19,744][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:20:20,246][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:20:20,747][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:20:21,248][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:20:21,748][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:20:22,253][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:20:22,756][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:20:23,252][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:20:23,752][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:20:24,251][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:20:24,747][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:20:25,248][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:20:25,747][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:20:26,250][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:20:26,762][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:20:27,266][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:20:27,767][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:20:28,267][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:20:28,766][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:20:29,267][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:20:29,766][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:20:30,264][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:20:30,760][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:20:31,256][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:20:31,759][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:20:32,265][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:20:32,762][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:20:33,259][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:20:33,755][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:20:34,258][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:20:34,754][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:20:35,250][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:20:35,746][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:20:36,245][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:20:36,744][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:20:37,241][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:20:37,739][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:20:38,236][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:20:38,743][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:20:39,266][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:20:39,777][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:20:40,280][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:20:40,786][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:20:41,298][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:20:41,809][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:20:42,316][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:20:42,816][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:20:43,317][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:20:43,818][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:20:44,322][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:20:44,822][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:20:45,322][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:20:45,821][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:20:46,320][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:20:46,817][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:20:47,313][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:20:47,814][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:20:48,312][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:20:48,811][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:20:49,310][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:20:49,811][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:20:50,317][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9820 tokens. +[2026-03-26 06:20:51,159][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.05%, Current % of VRAM taken: 60.48%, Block Peak % of device VRAM: 62.44%, ΔTime: 00:00:32 +[2026-03-26 06:20:51,925][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:20:51,927][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:20:51,929][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:20:52,641][__main__][INFO] - Iteration 543 took 51s (29.74% Gen, 68.87% Train). Generation: 15s, Training: 35s. Estimated remaining time: 34h 22m 33s. Estimated total time: 42h 42m 37s. Time estimates for 10 more iterations: 8m 32s, 100 more iterations: 1h 25m 25s, 500 more iterations: 7h 7m 6s. +[2026-03-26 06:20:52,643][__main__][INFO] - Starting iteration 543. +[2026-03-26 06:20:53,065][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 54 and human policies 1. +[2026-03-26 06:20:53,066][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:21:07,918][__main__][INFO] - Number of regex retries in iteration 543: 0 +[2026-03-26 06:21:07,919][__main__][INFO] - agents played in iteration 543 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:21:08,681][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:21:08,704][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:21:08,728][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:21:08,752][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:21:08,752][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:21:08,753][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:21:09,542][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:21:09,994][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:21:10,495][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:21:10,989][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:21:11,492][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:21:11,988][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:21:12,486][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:21:12,983][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:21:13,479][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:21:13,979][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:21:14,480][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:21:14,982][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:21:15,480][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:21:15,983][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:21:16,485][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:21:16,984][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:21:17,483][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:21:17,986][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:21:18,489][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:21:19,003][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:21:19,507][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:21:20,009][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:21:20,510][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:21:21,010][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:21:21,511][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:21:22,010][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:21:22,508][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:21:23,005][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:21:23,505][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:21:24,001][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:21:24,495][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:21:24,990][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:21:25,487][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:21:25,983][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:21:26,482][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:21:26,982][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:21:27,481][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:21:27,980][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:21:28,484][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:21:28,986][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:21:29,498][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:21:30,006][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:21:30,513][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:21:31,018][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:21:31,525][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:21:32,031][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:21:32,536][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:21:33,041][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:21:33,546][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:21:34,046][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:21:34,550][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:21:35,051][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:21:35,550][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:21:36,046][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:21:36,542][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:21:37,042][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:21:37,538][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:21:38,035][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:21:38,532][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:21:39,029][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:21:39,525][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:21:40,025][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:21:40,522][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:21:41,019][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:21:41,515][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9857 tokens. +[2026-03-26 06:21:42,345][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 10.88%, Current % of VRAM taken: 57.31%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 06:21:43,103][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:21:43,105][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:21:43,107][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:21:43,987][__main__][INFO] - Iteration 544 took 50s (29.17% Gen, 69.10% Train). Generation: 14s, Training: 35s. Estimated remaining time: 34h 5m 13s. Estimated total time: 42h 26m 9s. Time estimates for 10 more iterations: 8m 29s, 100 more iterations: 1h 24m 52s, 500 more iterations: 7h 4m 21s. +[2026-03-26 06:21:43,989][__main__][INFO] - Starting iteration 544. +[2026-03-26 06:21:44,391][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 54 and human policies 1. +[2026-03-26 06:21:44,391][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:22:01,886][__main__][INFO] - Number of regex retries in iteration 544: 0 +[2026-03-26 06:22:01,887][__main__][INFO] - agents played in iteration 544 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:22:02,652][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 50.40%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:22:02,675][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 50.40%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:22:02,699][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 50.40%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:22:02,722][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 50.40%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:22:02,723][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:22:02,723][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:22:03,509][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:22:03,961][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:22:04,462][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:22:04,958][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:22:05,459][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:22:05,970][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:22:06,477][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:22:06,974][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:22:07,471][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:22:07,970][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:22:08,472][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:22:08,970][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:22:09,474][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:22:09,985][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:22:10,487][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:22:10,990][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:22:11,490][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:22:11,992][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:22:12,490][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:22:12,988][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:22:13,487][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:22:13,989][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:22:14,489][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:22:14,987][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:22:15,487][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:22:15,984][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:22:16,483][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:22:16,980][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:22:17,478][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:22:17,978][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:22:18,478][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:22:18,981][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:22:19,481][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:22:19,977][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:22:20,479][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:22:20,982][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:22:21,489][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:22:21,993][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:22:22,497][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:22:22,998][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:22:23,502][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:22:24,009][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:22:24,518][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:22:25,021][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:22:25,523][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:22:26,023][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:22:26,523][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:22:27,018][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:22:27,514][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:22:28,011][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:22:28,511][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:22:29,014][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:22:29,510][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:22:30,009][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:22:30,508][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:22:31,011][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:22:31,509][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:22:32,006][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:22:32,501][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:22:32,996][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:22:33,492][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:22:33,991][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:22:34,493][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:22:34,988][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:22:35,490][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9867 tokens. +[2026-03-26 06:22:36,309][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.21%, Current % of VRAM taken: 60.63%, Block Peak % of device VRAM: 62.14%, ΔTime: 00:00:32 +[2026-03-26 06:22:37,072][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:22:37,075][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:22:37,076][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:22:37,986][__main__][INFO] - Iteration 545 took 53s (32.64% Gen, 65.66% Train). Generation: 17s, Training: 35s. Estimated remaining time: 36h 17m 59s. Estimated total time: 44h 39m 48s. Time estimates for 10 more iterations: 8m 55s, 100 more iterations: 1h 29m 19s, 500 more iterations: 7h 26m 38s. +[2026-03-26 06:22:37,988][__main__][INFO] - Starting iteration 545. +[2026-03-26 06:22:38,395][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 54 and human policies 1. +[2026-03-26 06:22:38,396][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:22:53,202][__main__][INFO] - Number of regex retries in iteration 545: 0 +[2026-03-26 06:22:53,203][__main__][INFO] - agents played in iteration 545 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:22:53,962][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:22:53,987][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:22:54,011][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:22:54,036][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:22:54,036][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:22:54,037][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:22:54,856][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:22:55,312][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:22:55,818][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:22:56,323][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:22:56,817][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:22:57,319][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:22:57,823][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:22:58,324][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:22:58,820][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:22:59,323][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:22:59,840][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:23:00,364][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:23:00,874][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:23:01,380][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:23:01,884][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:23:02,390][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:23:02,912][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:23:03,408][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:23:03,907][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:23:04,408][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:23:04,906][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:23:05,407][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:23:05,912][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:23:06,419][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:23:06,921][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:23:07,422][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:23:07,919][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:23:08,419][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:23:08,919][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:23:09,424][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:23:09,931][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:23:10,430][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:23:10,933][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:23:11,440][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:23:11,947][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:23:12,451][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:23:12,958][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:23:13,472][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:23:13,975][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:23:14,479][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:23:14,985][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:23:15,493][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:23:15,995][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:23:16,499][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:23:17,002][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:23:17,511][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:23:18,018][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:23:18,519][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:23:19,020][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:23:19,519][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:23:20,020][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:23:20,523][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:23:21,026][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:23:21,527][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:23:22,037][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:23:22,547][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:23:23,050][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:23:23,553][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:23:24,055][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:23:24,558][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:23:25,059][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:23:25,561][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:23:26,062][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:23:26,563][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:23:27,067][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9934 tokens. +[2026-03-26 06:23:27,991][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.17%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.25%, ΔTime: 00:00:33 +[2026-03-26 06:23:28,761][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:23:28,763][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:23:28,765][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:23:29,499][__main__][INFO] - Iteration 546 took 51s (28.97% Gen, 69.59% Train). Generation: 14s, Training: 35s. Estimated remaining time: 34h 12m 33s. Estimated total time: 42h 35m 14s. Time estimates for 10 more iterations: 8m 31s, 100 more iterations: 1h 25m 10s, 500 more iterations: 7h 5m 52s. +[2026-03-26 06:23:29,502][__main__][INFO] - Starting iteration 546. +[2026-03-26 06:23:29,900][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 54 and human policies 1. +[2026-03-26 06:23:29,900][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:23:44,510][__main__][INFO] - Number of regex retries in iteration 546: 0 +[2026-03-26 06:23:44,511][__main__][INFO] - agents played in iteration 546 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:23:45,269][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:23:45,292][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:23:45,317][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:23:45,340][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:23:45,341][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:23:45,341][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:23:46,136][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:23:46,596][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:23:47,102][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:23:47,603][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:23:48,108][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:23:48,612][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:23:49,113][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:23:49,613][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:23:50,116][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:23:50,621][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:23:51,124][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:23:51,626][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:23:52,132][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:23:52,637][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:23:53,148][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:23:53,648][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:23:54,153][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:23:54,658][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:23:55,156][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:23:55,656][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:23:56,155][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:23:56,657][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:23:57,156][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:23:57,655][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:23:58,154][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:23:58,652][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:23:59,148][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:23:59,644][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:24:00,144][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:24:00,645][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:24:01,142][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:24:01,644][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:24:02,148][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:24:02,650][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:24:03,149][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:24:03,652][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:24:04,153][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:24:04,654][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:24:05,156][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:24:05,655][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:24:06,155][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:24:06,656][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:24:07,158][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:24:07,657][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:24:08,159][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:24:08,655][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:24:09,154][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:24:09,650][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:24:10,146][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:24:10,645][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:24:11,143][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:24:11,642][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:24:12,139][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:24:12,637][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:24:13,136][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:24:13,633][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:24:14,131][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:24:14,630][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:24:15,127][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:24:15,626][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:24:16,124][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:24:16,624][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:24:17,122][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:24:17,619][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:24:18,118][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9820 tokens. +[2026-03-26 06:24:18,941][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 61.93%, ΔTime: 00:00:32 +[2026-03-26 06:24:19,699][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:24:19,701][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:24:19,703][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:24:20,508][__main__][INFO] - Iteration 547 took 50s (28.87% Gen, 69.54% Train). Generation: 14s, Training: 35s. Estimated remaining time: 33h 46m 55s. Estimated total time: 42h 10m 27s. Time estimates for 10 more iterations: 8m 26s, 100 more iterations: 1h 24m 20s, 500 more iterations: 7h 1m 44s. +[2026-03-26 06:24:20,510][__main__][INFO] - Starting iteration 547. +[2026-03-26 06:24:20,910][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 54 and human policies 1. +[2026-03-26 06:24:20,911][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:24:36,419][__main__][INFO] - Number of regex retries in iteration 547: 0 +[2026-03-26 06:24:36,420][__main__][INFO] - agents played in iteration 547 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:24:37,179][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:24:37,203][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:24:37,227][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:24:37,250][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:24:37,251][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:24:37,251][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:24:38,031][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:24:38,482][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:24:38,989][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:24:39,489][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:24:39,988][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:24:40,486][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:24:40,986][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:24:41,484][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:24:41,980][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:24:42,479][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:24:42,980][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:24:43,480][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:24:43,978][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:24:44,472][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:24:44,969][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:24:45,468][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:24:45,967][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:24:46,465][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:24:46,969][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:24:47,478][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:24:47,975][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:24:48,474][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:24:48,971][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:24:49,467][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:24:49,965][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:24:50,468][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:24:50,968][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:24:51,468][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:24:51,969][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:24:52,469][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:24:52,972][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:24:53,478][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:24:53,984][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:24:54,485][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:24:54,989][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:24:55,489][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:24:55,992][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:24:56,494][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:24:57,001][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:24:57,501][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:24:58,001][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:24:58,505][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:24:59,013][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:24:59,513][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:25:00,023][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:25:00,524][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:25:01,025][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:25:01,527][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:25:02,027][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:25:02,527][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:25:03,025][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:25:03,522][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:25:04,018][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:25:04,515][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:25:05,017][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:25:05,513][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:25:06,010][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:25:06,511][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:25:07,009][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:25:07,508][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:25:08,006][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:25:08,508][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:25:09,015][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:25:09,512][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:25:10,012][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9831 tokens. +[2026-03-26 06:25:10,849][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.09%, ΔTime: 00:00:32 +[2026-03-26 06:25:14,670][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:25:14,672][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:25:14,674][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:25:16,288][__main__][INFO] - Iteration 548 took 55s (28.01% Gen, 69.08% Train). Generation: 15s, Training: 38s. Estimated remaining time: 37h 44m 27s. Estimated total time: 46h 8m 54s. Time estimates for 10 more iterations: 9m 13s, 100 more iterations: 1h 32m 17s, 500 more iterations: 7h 41m 29s. +[2026-03-26 06:25:16,290][__main__][INFO] - Starting iteration 548. +[2026-03-26 06:25:16,708][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 54 and human policies 1. +[2026-03-26 06:25:16,709][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:25:21,720][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:25:31,509][__main__][INFO] - Number of regex retries in iteration 548: 1 +[2026-03-26 06:25:31,510][__main__][INFO] - agents played in iteration 548 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:25:32,324][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:25:32,348][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:25:32,371][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:25:32,394][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:25:32,395][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:25:32,396][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:25:33,180][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:25:33,641][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:25:34,158][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:25:34,662][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:25:35,164][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:25:35,664][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:25:36,167][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:25:36,688][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:25:37,193][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:25:37,696][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:25:38,192][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:25:38,691][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:25:39,194][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:25:39,708][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:25:40,206][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:25:40,707][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:25:41,204][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:25:41,706][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:25:42,207][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:25:42,712][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:25:43,216][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:25:43,716][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:25:44,217][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:25:44,717][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:25:45,216][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:25:45,718][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:25:46,235][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:25:46,743][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:25:47,253][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:25:47,757][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:25:48,262][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:25:48,766][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:25:49,270][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:25:49,775][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:25:50,279][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:25:50,785][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:25:51,290][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:25:51,794][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:25:52,299][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:25:52,800][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:25:53,314][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:25:53,816][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:25:54,315][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:25:54,815][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:25:55,315][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:25:55,812][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:25:56,312][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:25:56,812][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:25:57,311][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:25:57,812][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:25:58,314][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:25:58,814][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:25:59,320][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:25:59,830][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:26:00,333][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:26:00,837][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:26:01,340][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:26:01,841][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:26:02,342][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:26:02,846][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:26:03,349][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:26:03,865][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:26:04,367][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:26:04,870][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:26:05,374][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9939 tokens. +[2026-03-26 06:26:06,212][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.14%, ΔTime: 00:00:33 +[2026-03-26 06:26:06,980][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:26:06,982][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:26:06,984][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:26:07,765][__main__][INFO] - Iteration 549 took 51s (28.99% Gen, 69.48% Train). Generation: 14s, Training: 35s. Estimated remaining time: 34h 7m 34s. Estimated total time: 42h 32m 53s. Time estimates for 10 more iterations: 8m 30s, 100 more iterations: 1h 25m 5s, 500 more iterations: 7h 5m 28s. +[2026-03-26 06:26:07,767][__main__][INFO] - Starting iteration 549. +[2026-03-26 06:26:08,166][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 54 and human policies 1. +[2026-03-26 06:26:08,166][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:26:12,544][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:26:24,604][__main__][INFO] - Number of regex retries in iteration 549: 1 +[2026-03-26 06:26:24,605][__main__][INFO] - agents played in iteration 549 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:26:25,435][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:26:25,460][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:26:25,484][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:26:25,509][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:26:25,510][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:26:25,510][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:26:26,306][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:26:26,764][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:26:27,268][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:26:27,772][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:26:28,283][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:26:28,781][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:26:29,280][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:26:29,779][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:26:30,274][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:26:30,774][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:26:31,275][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:26:31,774][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:26:32,273][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:26:32,775][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:26:33,271][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:26:33,766][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:26:34,266][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:26:34,765][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:26:35,286][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:26:35,787][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:26:36,290][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:26:36,791][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:26:37,292][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:26:37,791][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:26:38,295][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:26:38,801][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:26:39,312][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:26:39,816][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:26:40,317][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:26:40,822][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:26:41,322][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:26:41,823][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:26:42,325][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:26:42,826][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:26:43,328][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:26:43,829][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:26:44,334][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:26:44,837][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:26:45,344][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:26:45,847][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:26:46,350][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:26:46,852][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:26:47,352][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:26:47,850][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:26:48,350][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:26:48,850][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:26:49,350][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:26:49,848][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:26:50,347][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:26:50,846][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:26:51,345][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:26:51,845][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:26:52,345][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:26:52,847][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:26:53,349][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:26:53,847][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:26:54,347][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:26:54,844][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:26:55,345][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:26:55,847][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:26:56,348][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:26:56,848][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:26:57,347][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:26:57,849][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:26:58,350][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9967 tokens. +[2026-03-26 06:26:59,177][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.15%, Current % of VRAM taken: 60.58%, Block Peak % of device VRAM: 62.11%, ΔTime: 00:00:32 +[2026-03-26 06:26:59,984][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:26:59,986][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:26:59,988][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:27:00,728][__main__][INFO] - Iteration 550 took 52s (31.27% Gen, 67.32% Train). Generation: 16s, Training: 35s. Estimated remaining time: 35h 21m 56s. Estimated total time: 43h 48m 8s. Time estimates for 10 more iterations: 8m 45s, 100 more iterations: 1h 27m 36s, 500 more iterations: 7h 18m 1s. +[2026-03-26 06:27:00,730][__main__][INFO] - Starting iteration 550. +[2026-03-26 06:27:01,137][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 54 and human policies 1. +[2026-03-26 06:27:01,138][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:27:23,812][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 20 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:27:29,376][__main__][INFO] - Number of regex retries in iteration 550: 1 +[2026-03-26 06:27:29,377][__main__][INFO] - agents played in iteration 550 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:27:30,150][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:27:30,174][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:27:30,198][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:27:30,222][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:27:30,222][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:27:30,223][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:27:31,002][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:27:31,454][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:27:31,952][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:27:32,448][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:27:32,946][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:27:33,446][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:27:33,951][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:27:34,449][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:27:34,948][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:27:35,451][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:27:35,951][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:27:36,459][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:27:36,959][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:27:37,459][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:27:37,965][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:27:38,466][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:27:38,967][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:27:39,468][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:27:39,971][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:27:40,480][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:27:40,987][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:27:41,489][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:27:41,992][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:27:42,492][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:27:43,009][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:27:43,510][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:27:44,011][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:27:44,511][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:27:45,008][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:27:45,505][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:27:46,006][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:27:46,513][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:27:47,013][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:27:47,516][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:27:48,017][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:27:48,518][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:27:49,017][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:27:49,529][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:27:50,032][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:27:50,534][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:27:51,035][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:27:51,536][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:27:52,035][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:27:52,534][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:27:53,044][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:27:53,543][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:27:54,072][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:27:54,574][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:27:55,073][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:27:55,571][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:27:56,071][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:27:56,572][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:27:57,069][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:27:57,567][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:27:58,065][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:27:58,566][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:27:59,066][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:27:59,566][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:28:00,065][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:28:00,576][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:28:01,074][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:28:01,575][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:28:02,075][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:28:02,573][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:28:03,072][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9717 tokens. +[2026-03-26 06:28:03,895][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.03%, ΔTime: 00:00:32 +[2026-03-26 06:28:04,661][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:28:04,663][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:28:04,665][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:28:06,071][__main__][INFO] - Iteration 551 took 1m 4s (43.49% Gen, 54.34% Train). Generation: 28s, Training: 35s. Estimated remaining time: 45h 39m 25s. Estimated total time: 54h 6m 42s. Time estimates for 10 more iterations: 10m 49s, 100 more iterations: 1h 48m 13s, 500 more iterations: 9h 1m 7s. +[2026-03-26 06:28:06,073][__main__][INFO] - Starting iteration 551. +[2026-03-26 06:28:06,473][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 55 and human policies 1. +[2026-03-26 06:28:06,474][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:28:21,012][__main__][INFO] - Number of regex retries in iteration 551: 0 +[2026-03-26 06:28:21,013][__main__][INFO] - agents played in iteration 551 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:28:21,780][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:28:21,804][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:28:21,827][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:28:21,851][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:28:21,851][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:28:21,852][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:28:22,665][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:28:23,125][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:28:23,629][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:28:24,130][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:28:24,633][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:28:25,134][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:28:25,633][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:28:26,133][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:28:26,632][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:28:27,132][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:28:27,631][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:28:28,130][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:28:28,628][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:28:29,127][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:28:29,624][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:28:30,120][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:28:30,619][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:28:31,119][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:28:31,619][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:28:32,120][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:28:32,620][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:28:33,125][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:28:33,635][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:28:34,137][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:28:34,640][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:28:35,145][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:28:35,650][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:28:36,154][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:28:36,660][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:28:37,162][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:28:37,663][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:28:38,168][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:28:38,672][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:28:39,177][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:28:39,682][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:28:40,186][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:28:40,690][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:28:41,192][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:28:41,697][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:28:42,197][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:28:42,697][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:28:43,222][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:28:43,728][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:28:44,230][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:28:44,740][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:28:45,253][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:28:45,757][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:28:46,260][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:28:46,762][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:28:47,264][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:28:47,767][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:28:48,268][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:28:48,770][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:28:49,269][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:28:49,774][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:28:50,274][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:28:50,778][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:28:51,280][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:28:51,781][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:28:52,281][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:28:52,785][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:28:53,287][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:28:53,795][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:28:54,305][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:28:54,807][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9965 tokens. +[2026-03-26 06:28:55,651][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.30%, ΔTime: 00:00:32 +[2026-03-26 06:28:56,443][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:28:56,446][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:28:56,447][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:28:57,187][__main__][INFO] - Iteration 552 took 50s (28.67% Gen, 69.87% Train). Generation: 14s, Training: 35s. Estimated remaining time: 33h 47m 36s. Estimated total time: 42h 15m 44s. Time estimates for 10 more iterations: 8m 27s, 100 more iterations: 1h 24m 31s, 500 more iterations: 7h 2m 37s. +[2026-03-26 06:28:57,191][__main__][INFO] - Starting iteration 552. +[2026-03-26 06:28:57,593][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 55 and human policies 1. +[2026-03-26 06:28:57,593][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:29:56,384][__main__][INFO] - Number of regex retries in iteration 552: 0 +[2026-03-26 06:29:56,384][__main__][INFO] - agents played in iteration 552 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:29:57,154][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:29:57,178][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:29:57,202][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:29:57,225][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:29:57,226][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:29:57,226][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:29:58,024][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:29:58,473][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:29:58,976][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:29:59,471][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:29:59,966][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:30:00,473][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:30:00,983][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:30:01,481][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:30:01,978][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:30:02,476][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:30:02,972][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:30:03,472][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:30:03,976][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:30:04,500][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:30:04,999][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:30:05,497][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:30:05,995][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:30:06,491][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:30:06,987][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:30:07,485][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:30:07,987][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:30:08,487][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:30:08,983][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:30:09,478][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:30:09,975][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:30:10,471][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:30:10,968][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:30:11,466][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:30:11,970][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:30:12,469][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:30:12,969][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:30:13,466][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:30:13,965][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:30:14,466][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:30:14,965][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:30:15,465][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:30:15,969][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:30:16,469][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:30:16,970][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:30:17,497][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:30:18,004][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:30:18,511][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:30:19,012][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:30:19,517][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:30:20,021][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:30:20,526][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:30:21,029][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:30:21,533][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:30:22,036][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:30:22,538][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:30:23,039][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:30:23,538][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:30:24,037][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:30:24,536][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:30:25,035][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:30:25,534][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:30:26,033][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:30:26,531][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:30:27,029][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:30:27,526][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:30:28,024][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:30:28,522][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:30:29,019][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:30:29,518][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:30:30,015][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9839 tokens. +[2026-03-26 06:30:30,846][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.08%, ΔTime: 00:00:32 +[2026-03-26 06:30:31,602][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:30:31,605][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:30:31,606][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:30:32,346][__main__][INFO] - Iteration 553 took 1m 34s (62.05% Gen, 37.17% Train). Generation: 58s, Training: 35s. Estimated remaining time: 70h 27m 57s. Estimated total time: 78h 57m 41s. Time estimates for 10 more iterations: 15m 47s, 100 more iterations: 2h 37m 55s, 500 more iterations: 13h 9m 36s. +[2026-03-26 06:30:32,348][__main__][INFO] - Starting iteration 553. +[2026-03-26 06:30:32,748][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 55 and human policies 1. +[2026-03-26 06:30:32,749][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:30:47,528][__main__][INFO] - Number of regex retries in iteration 553: 0 +[2026-03-26 06:30:47,529][__main__][INFO] - agents played in iteration 553 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:30:48,305][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:30:48,329][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:30:48,353][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:30:48,377][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:30:48,378][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:30:48,379][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:30:49,164][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:30:49,626][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:30:50,138][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:30:50,643][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:30:51,165][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:30:51,667][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:30:52,173][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:30:52,674][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:30:53,178][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:30:53,678][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:30:54,184][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:30:54,686][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:30:55,187][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:30:55,690][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:30:56,212][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:30:56,718][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:30:57,220][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:30:57,718][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:30:58,224][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:30:58,733][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:30:59,234][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:30:59,735][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:31:00,235][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:31:00,738][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:31:01,237][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:31:01,749][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:31:02,248][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:31:02,749][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:31:03,246][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:31:03,742][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:31:04,238][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:31:04,731][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:31:05,227][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:31:05,726][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:31:06,226][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:31:06,726][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:31:07,226][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:31:07,725][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:31:08,234][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:31:08,749][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:31:09,256][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:31:09,759][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:31:10,260][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:31:10,764][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:31:11,269][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:31:11,774][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:31:12,290][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:31:12,795][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:31:13,297][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:31:13,797][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:31:14,296][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:31:14,795][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:31:15,294][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:31:15,793][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:31:16,298][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:31:16,804][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:31:17,304][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:31:17,805][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:31:18,305][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:31:18,805][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:31:19,305][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:31:19,806][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:31:20,306][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:31:20,806][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:31:21,309][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9899 tokens. +[2026-03-26 06:31:22,145][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.14%, ΔTime: 00:00:32 +[2026-03-26 06:31:23,005][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:31:23,008][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:31:23,009][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:31:23,754][__main__][INFO] - Iteration 554 took 51s (28.98% Gen, 69.56% Train). Generation: 14s, Training: 35s. Estimated remaining time: 33h 59m 46s. Estimated total time: 42h 30m 21s. Time estimates for 10 more iterations: 8m 30s, 100 more iterations: 1h 25m 0s, 500 more iterations: 7h 5m 3s. +[2026-03-26 06:31:23,757][__main__][INFO] - Starting iteration 554. +[2026-03-26 06:31:24,156][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 55 and human policies 1. +[2026-03-26 06:31:24,157][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:31:28,756][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:31:38,088][__main__][INFO] - Number of regex retries in iteration 554: 1 +[2026-03-26 06:31:38,089][__main__][INFO] - agents played in iteration 554 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:31:38,852][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:31:38,876][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:31:38,899][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:31:38,923][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:31:38,923][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:31:38,924][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:31:39,746][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:31:40,206][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:31:40,713][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:31:41,214][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:31:41,715][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:31:42,234][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:31:42,743][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:31:43,249][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:31:43,747][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:31:44,246][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:31:44,746][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:31:45,246][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:31:45,748][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:31:46,248][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:31:46,747][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:31:47,246][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:31:47,746][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:31:48,245][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:31:48,744][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:31:49,242][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:31:49,755][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:31:50,259][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:31:50,760][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:31:51,260][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:31:51,760][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:31:52,260][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:31:52,765][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:31:53,273][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:31:53,774][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:31:54,278][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:31:54,784][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:31:55,279][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:31:55,779][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:31:56,287][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:31:56,795][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:31:57,299][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:31:57,807][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:31:58,317][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:31:58,826][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:31:59,334][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:31:59,846][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:32:00,363][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:32:00,871][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:32:01,400][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:32:01,913][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:32:02,419][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:32:02,926][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:32:03,434][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:32:03,944][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:32:04,462][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:32:04,968][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:32:05,471][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:32:05,973][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:32:06,475][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:32:06,975][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:32:07,476][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:32:07,977][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:32:08,477][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:32:08,978][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:32:09,478][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:32:09,977][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:32:10,477][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:32:10,981][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:32:11,489][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:32:11,991][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9977 tokens. +[2026-03-26 06:32:12,860][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.15%, ΔTime: 00:00:33 +[2026-03-26 06:32:13,634][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:32:13,637][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:32:13,638][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:32:14,470][__main__][INFO] - Iteration 555 took 50s (27.69% Gen, 70.65% Train). Generation: 13s, Training: 35s. Estimated remaining time: 33h 24m 19s. Estimated total time: 41h 55m 45s. Time estimates for 10 more iterations: 8m 23s, 100 more iterations: 1h 23m 51s, 500 more iterations: 6h 59m 17s. +[2026-03-26 06:32:14,473][__main__][INFO] - Starting iteration 555. +[2026-03-26 06:32:14,877][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 55 and human policies 1. +[2026-03-26 06:32:14,877][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:32:34,840][__main__][INFO] - Number of regex retries in iteration 555: 0 +[2026-03-26 06:32:34,841][__main__][INFO] - agents played in iteration 555 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:32:35,605][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:32:35,629][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:32:35,653][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:32:35,676][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:32:35,677][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:32:35,677][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:32:36,470][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:32:36,927][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:32:37,429][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:32:37,929][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:32:38,435][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:32:38,947][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:32:39,459][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:32:39,961][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:32:40,464][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:32:40,965][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:32:41,468][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:32:41,968][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:32:42,470][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:32:42,971][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:32:43,471][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:32:43,972][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:32:44,471][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:32:44,970][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:32:45,469][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:32:45,970][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:32:46,480][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:32:46,977][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:32:47,477][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:32:47,978][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:32:48,478][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:32:48,978][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:32:49,479][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:32:49,980][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:32:50,482][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:32:50,981][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:32:51,482][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:32:51,982][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:32:52,492][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:32:53,003][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:32:53,509][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:32:54,011][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:32:54,516][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:32:55,020][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:32:55,523][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:32:56,027][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:32:56,534][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:32:57,046][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:32:57,552][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:32:58,061][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:32:58,563][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:32:59,069][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:32:59,573][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:33:00,072][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:33:00,573][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:33:01,074][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:33:01,572][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:33:02,072][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:33:02,571][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:33:03,073][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:33:03,571][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:33:04,071][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:33:04,573][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:33:05,071][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:33:05,568][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:33:06,066][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:33:06,565][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:33:07,062][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:33:07,559][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:33:08,057][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:33:08,555][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10014 tokens. +[2026-03-26 06:33:09,377][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.25%, ΔTime: 00:00:32 +[2026-03-26 06:33:10,153][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:33:10,156][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:33:10,157][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:33:11,032][__main__][INFO] - Iteration 556 took 56s (35.55% Gen, 62.89% Train). Generation: 19s, Training: 35s. Estimated remaining time: 38h 15m 24s. Estimated total time: 46h 47m 46s. Time estimates for 10 more iterations: 9m 21s, 100 more iterations: 1h 33m 35s, 500 more iterations: 7h 47m 57s. +[2026-03-26 06:33:11,034][__main__][INFO] - Starting iteration 556. +[2026-03-26 06:33:11,434][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 55 and human policies 1. +[2026-03-26 06:33:11,435][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:33:25,770][__main__][INFO] - Number of regex retries in iteration 556: 0 +[2026-03-26 06:33:25,770][__main__][INFO] - agents played in iteration 556 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:33:26,530][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:33:26,556][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:33:26,582][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:33:26,608][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:33:26,609][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:33:26,609][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:33:27,419][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:33:27,876][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:33:28,380][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:33:28,885][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:33:29,393][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:33:29,897][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:33:30,403][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:33:30,906][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:33:31,409][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:33:31,918][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:33:32,436][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:33:32,940][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:33:33,443][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:33:33,945][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:33:34,445][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:33:34,946][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:33:35,450][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:33:35,950][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:33:36,450][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:33:36,949][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:33:37,449][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:33:37,948][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:33:38,447][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:33:38,953][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:33:39,463][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:33:39,964][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:33:40,466][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:33:40,966][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:33:41,466][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:33:41,965][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:33:42,466][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:33:42,971][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:33:43,481][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:33:43,991][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:33:44,499][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:33:45,009][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:33:45,510][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:33:46,016][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:33:46,521][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:33:47,025][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:33:47,531][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:33:48,036][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:33:48,539][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:33:49,041][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:33:49,543][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:33:50,045][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:33:50,543][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:33:51,042][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:33:51,540][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:33:52,038][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:33:52,545][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:33:53,060][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:33:53,560][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:33:54,060][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:33:54,560][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:33:55,059][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:33:55,560][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:33:56,059][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:33:56,558][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:33:57,056][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:33:57,558][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:33:58,064][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:33:58,566][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:33:59,066][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:33:59,563][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9922 tokens. +[2026-03-26 06:34:00,389][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.95%, Current % of VRAM taken: 60.37%, Block Peak % of device VRAM: 62.15%, ΔTime: 00:00:32 +[2026-03-26 06:34:01,161][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:34:01,164][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:34:01,165][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:34:01,927][__main__][INFO] - Iteration 557 took 50s (28.39% Gen, 70.10% Train). Generation: 14s, Training: 35s. Estimated remaining time: 33h 31m 28s. Estimated total time: 42h 4m 42s. Time estimates for 10 more iterations: 8m 24s, 100 more iterations: 1h 24m 9s, 500 more iterations: 7h 0m 47s. +[2026-03-26 06:34:01,930][__main__][INFO] - Starting iteration 557. +[2026-03-26 06:34:02,330][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 55 and human policies 1. +[2026-03-26 06:34:02,331][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:34:06,918][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:34:17,515][__main__][INFO] - Number of regex retries in iteration 557: 1 +[2026-03-26 06:34:17,516][__main__][INFO] - agents played in iteration 557 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:34:18,276][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:34:18,300][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:34:18,323][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:34:18,347][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:34:18,348][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:34:18,348][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:34:19,139][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:34:19,591][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:34:20,097][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:34:20,599][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:34:21,110][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:34:21,611][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:34:22,112][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:34:22,613][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:34:23,113][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:34:23,612][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:34:24,115][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:34:24,619][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:34:25,125][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:34:25,621][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:34:26,119][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:34:26,617][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:34:27,120][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:34:27,619][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:34:28,116][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:34:28,615][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:34:29,112][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:34:29,611][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:34:30,110][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:34:30,613][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:34:31,125][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:34:31,624][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:34:32,122][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:34:32,623][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:34:33,123][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:34:33,620][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:34:34,127][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:34:34,625][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:34:35,123][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:34:35,622][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:34:36,126][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:34:36,632][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:34:37,134][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:34:37,633][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:34:38,136][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:34:38,638][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:34:39,141][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:34:39,644][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:34:40,147][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:34:40,651][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:34:41,154][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:34:41,660][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:34:42,158][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:34:42,656][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:34:43,154][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:34:43,651][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:34:44,150][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:34:44,648][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:34:45,149][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:34:45,647][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:34:46,145][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:34:46,644][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:34:47,144][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:34:47,642][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:34:48,139][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:34:48,641][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:34:49,138][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:34:49,634][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:34:50,136][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:34:50,633][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:34:51,135][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9956 tokens. +[2026-03-26 06:34:51,960][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.04%, Current % of VRAM taken: 60.46%, Block Peak % of device VRAM: 62.07%, ΔTime: 00:00:32 +[2026-03-26 06:34:52,755][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:34:52,757][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:34:52,759][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:34:53,560][__main__][INFO] - Iteration 558 took 51s (29.64% Gen, 68.79% Train). Generation: 15s, Training: 35s. Estimated remaining time: 34h 7m 27s. Estimated total time: 42h 41m 31s. Time estimates for 10 more iterations: 8m 32s, 100 more iterations: 1h 25m 23s, 500 more iterations: 7h 6m 55s. +[2026-03-26 06:34:53,562][__main__][INFO] - Starting iteration 558. +[2026-03-26 06:34:53,965][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 55 and human policies 1. +[2026-03-26 06:34:53,966][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:35:09,193][__main__][INFO] - Number of regex retries in iteration 558: 0 +[2026-03-26 06:35:09,194][__main__][INFO] - agents played in iteration 558 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:35:09,954][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:35:09,978][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:35:10,002][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:35:10,025][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:35:10,026][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:35:10,026][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:35:10,806][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:35:11,264][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:35:11,777][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:35:12,278][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:35:12,776][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:35:13,274][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:35:13,777][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:35:14,289][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:35:14,788][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:35:15,295][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:35:15,802][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:35:16,309][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:35:16,825][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:35:17,336][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:35:17,836][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:35:18,340][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:35:18,843][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:35:19,340][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:35:19,837][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:35:20,344][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:35:20,847][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:35:21,344][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:35:21,840][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:35:22,338][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:35:22,837][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:35:23,349][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:35:23,851][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:35:24,352][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:35:24,851][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:35:25,353][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:35:25,853][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:35:26,370][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:35:26,884][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:35:27,389][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:35:27,892][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:35:28,396][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:35:28,897][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:35:29,400][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:35:29,908][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:35:30,420][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:35:30,925][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:35:31,430][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:35:31,935][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:35:32,434][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:35:32,934][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:35:33,432][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:35:33,934][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:35:34,438][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:35:34,937][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:35:35,437][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:35:35,937][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:35:36,437][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:35:36,937][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:35:37,436][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:35:37,939][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:35:38,454][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:35:38,955][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:35:39,455][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:35:39,955][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:35:40,453][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:35:40,954][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:35:41,454][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:35:41,955][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:35:42,457][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:35:42,957][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9869 tokens. +[2026-03-26 06:35:43,878][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.06%, ΔTime: 00:00:33 +[2026-03-26 06:35:44,644][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:35:44,646][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:35:44,648][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:35:45,422][__main__][INFO] - Iteration 559 took 51s (29.59% Gen, 68.90% Train). Generation: 15s, Training: 35s. Estimated remaining time: 34h 17m 55s. Estimated total time: 42h 52m 52s. Time estimates for 10 more iterations: 8m 34s, 100 more iterations: 1h 25m 45s, 500 more iterations: 7h 8m 48s. +[2026-03-26 06:35:45,424][__main__][INFO] - Starting iteration 559. +[2026-03-26 06:35:45,826][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 55 and human policies 1. +[2026-03-26 06:35:45,827][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:36:01,707][__main__][INFO] - Number of regex retries in iteration 559: 0 +[2026-03-26 06:36:01,708][__main__][INFO] - agents played in iteration 559 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:36:02,473][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:36:02,497][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:36:02,521][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:36:02,545][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:36:02,546][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:36:02,547][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:36:03,336][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:36:03,790][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:36:04,305][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:36:04,808][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:36:05,308][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:36:05,808][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:36:06,305][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:36:06,801][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:36:07,297][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:36:07,810][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:36:08,311][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:36:08,813][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:36:09,317][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:36:09,812][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:36:10,317][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:36:10,818][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:36:11,331][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:36:11,831][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:36:12,333][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:36:12,831][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:36:13,332][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:36:13,829][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:36:14,327][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:36:14,823][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:36:15,332][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:36:15,836][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:36:16,359][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:36:16,867][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:36:17,379][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:36:17,889][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:36:18,393][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:36:18,898][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:36:19,411][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:36:19,922][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:36:20,426][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:36:20,931][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:36:21,432][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:36:21,936][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:36:22,437][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:36:22,937][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:36:23,434][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:36:23,939][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:36:24,447][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:36:24,949][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:36:25,450][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:36:25,952][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:36:26,453][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:36:26,952][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:36:27,452][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:36:27,952][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:36:28,448][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:36:28,946][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:36:29,458][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:36:29,960][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:36:30,460][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:36:30,960][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:36:31,461][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:36:31,967][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:36:32,469][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:36:32,972][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:36:33,474][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:36:33,977][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:36:34,479][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:36:34,980][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:36:35,479][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9864 tokens. +[2026-03-26 06:36:36,304][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.26%, ΔTime: 00:00:32 +[2026-03-26 06:36:37,100][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:36:37,102][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:36:37,104][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:36:37,808][__main__][INFO] - Iteration 560 took 51s (30.55% Gen, 68.09% Train). Generation: 15s, Training: 35s. Estimated remaining time: 34h 43m 19s. Estimated total time: 43h 19m 8s. Time estimates for 10 more iterations: 8m 39s, 100 more iterations: 1h 26m 38s, 500 more iterations: 7h 13m 11s. +[2026-03-26 06:36:37,810][__main__][INFO] - Starting iteration 560. +[2026-03-26 06:36:38,210][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 55 and human policies 1. +[2026-03-26 06:36:38,210][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:36:44,264][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:36:54,289][__main__][INFO] - Number of regex retries in iteration 560: 1 +[2026-03-26 06:36:54,291][__main__][INFO] - agents played in iteration 560 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:36:55,055][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:36:55,079][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:36:55,107][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:36:55,132][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:36:55,133][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:36:55,133][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:36:55,942][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:36:56,400][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:36:56,904][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:36:57,402][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:36:57,902][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:36:58,406][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:36:58,912][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:36:59,414][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:36:59,918][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:37:00,421][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:37:00,922][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:37:01,422][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:37:01,933][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:37:02,441][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:37:02,944][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:37:03,446][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:37:03,947][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:37:04,451][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:37:04,953][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:37:05,461][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:37:05,977][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:37:06,491][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:37:07,000][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:37:07,514][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:37:08,027][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:37:08,534][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:37:09,047][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:37:09,553][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:37:10,061][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:37:10,573][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:37:11,077][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:37:11,585][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:37:12,088][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:37:12,599][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:37:13,107][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:37:13,617][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:37:14,120][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:37:14,630][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:37:15,149][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:37:15,652][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:37:16,159][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:37:16,662][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:37:17,169][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:37:17,674][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:37:18,180][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:37:18,683][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:37:19,187][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:37:19,695][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:37:20,206][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:37:20,706][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:37:21,206][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:37:21,703][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:37:22,200][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:37:22,699][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:37:23,200][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:37:23,700][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:37:24,205][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:37:24,708][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:37:25,211][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:37:25,715][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:37:26,217][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:37:26,719][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:37:27,219][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:37:27,719][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:37:28,216][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9891 tokens. +[2026-03-26 06:37:29,050][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.93%, Current % of VRAM taken: 60.35%, Block Peak % of device VRAM: 62.33%, ΔTime: 00:00:33 +[2026-03-26 06:37:29,817][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:37:29,819][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:37:29,821][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:37:31,349][__main__][INFO] - Iteration 561 took 53s (30.26% Gen, 66.86% Train). Generation: 16s, Training: 35s. Estimated remaining time: 35h 40m 18s. Estimated total time: 44h 17m 0s. Time estimates for 10 more iterations: 8m 51s, 100 more iterations: 1h 28m 34s, 500 more iterations: 7h 22m 50s. +[2026-03-26 06:37:31,352][__main__][INFO] - Starting iteration 561. +[2026-03-26 06:37:31,754][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 56 and human policies 1. +[2026-03-26 06:37:31,755][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:37:45,997][__main__][INFO] - Number of regex retries in iteration 561: 0 +[2026-03-26 06:37:45,998][__main__][INFO] - agents played in iteration 561 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:37:46,753][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:37:46,777][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:37:46,800][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:37:46,824][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:37:46,824][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:37:46,825][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:37:47,626][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:37:48,080][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:37:48,583][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:37:49,082][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:37:49,588][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:37:50,106][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:37:50,609][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:37:51,110][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:37:51,610][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:37:52,108][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:37:52,611][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:37:53,115][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:37:53,617][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:37:54,125][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:37:54,623][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:37:55,122][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:37:55,619][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:37:56,117][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:37:56,614][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:37:57,112][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:37:57,613][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:37:58,120][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:37:58,622][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:37:59,121][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:37:59,622][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:38:00,124][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:38:00,627][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:38:01,129][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:38:01,628][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:38:02,126][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:38:02,627][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:38:03,128][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:38:03,629][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:38:04,134][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:38:04,639][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:38:05,143][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:38:05,644][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:38:06,143][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:38:06,642][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:38:07,141][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:38:07,640][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:38:08,144][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:38:08,643][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:38:09,142][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:38:09,640][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:38:10,138][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:38:10,636][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:38:11,135][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:38:11,634][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:38:12,133][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:38:12,629][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:38:13,124][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:38:13,622][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:38:14,119][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:38:14,619][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:38:15,121][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:38:15,620][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:38:16,120][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:38:16,621][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:38:17,123][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:38:17,625][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:38:18,126][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:38:18,626][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:38:19,125][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:38:19,624][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9858 tokens. +[2026-03-26 06:38:20,458][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.01%, ΔTime: 00:00:32 +[2026-03-26 06:38:21,248][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:38:21,250][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:38:21,252][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:38:21,968][__main__][INFO] - Iteration 562 took 50s (28.36% Gen, 70.20% Train). Generation: 14s, Training: 35s. Estimated remaining time: 33h 13m 11s. Estimated total time: 41h 50m 44s. Time estimates for 10 more iterations: 8m 22s, 100 more iterations: 1h 23m 41s, 500 more iterations: 6h 58m 27s. +[2026-03-26 06:38:21,971][__main__][INFO] - Starting iteration 562. +[2026-03-26 06:38:22,373][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 56 and human policies 1. +[2026-03-26 06:38:22,373][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:38:36,408][__main__][INFO] - Number of regex retries in iteration 562: 0 +[2026-03-26 06:38:36,409][__main__][INFO] - agents played in iteration 562 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:38:37,166][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:38:37,190][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:38:37,213][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:38:37,237][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:38:37,237][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:38:37,238][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:38:38,020][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:38:38,474][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:38:38,994][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:38:39,498][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:38:40,000][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:38:40,502][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:38:41,003][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:38:41,514][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:38:42,025][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:38:42,526][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:38:43,026][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:38:43,530][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:38:44,034][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:38:44,543][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:38:45,053][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:38:45,559][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:38:46,062][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:38:46,562][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:38:47,076][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:38:47,579][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:38:48,084][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:38:48,588][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:38:49,090][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:38:49,592][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:38:50,101][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:38:50,604][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:38:51,109][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:38:51,610][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:38:52,113][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:38:52,614][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:38:53,120][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:38:53,631][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:38:54,135][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:38:54,640][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:38:55,145][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:38:55,649][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:38:56,154][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:38:56,658][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:38:57,172][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:38:57,674][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:38:58,176][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:38:58,679][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:38:59,180][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:38:59,681][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:39:00,182][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:39:00,685][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:39:01,199][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:39:01,705][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:39:02,209][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:39:02,710][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:39:03,210][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:39:03,709][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:39:04,209][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:39:04,706][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:39:05,210][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:39:05,720][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:39:06,221][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:39:06,720][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:39:07,223][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:39:07,723][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:39:08,226][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:39:08,725][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:39:09,224][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:39:09,725][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:39:10,231][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9979 tokens. +[2026-03-26 06:39:11,073][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.02%, Current % of VRAM taken: 60.45%, Block Peak % of device VRAM: 62.17%, ΔTime: 00:00:33 +[2026-03-26 06:39:11,879][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:39:11,881][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:39:11,883][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:39:12,752][__main__][INFO] - Iteration 563 took 50s (27.86% Gen, 70.41% Train). Generation: 14s, Training: 35s. Estimated remaining time: 33h 20m 37s. Estimated total time: 41h 59m 1s. Time estimates for 10 more iterations: 8m 23s, 100 more iterations: 1h 23m 58s, 500 more iterations: 6h 59m 50s. +[2026-03-26 06:39:12,755][__main__][INFO] - Starting iteration 563. +[2026-03-26 06:39:13,158][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 56 and human policies 1. +[2026-03-26 06:39:13,159][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:39:17,619][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:39:28,354][__main__][INFO] - Number of regex retries in iteration 563: 1 +[2026-03-26 06:39:28,355][__main__][INFO] - agents played in iteration 563 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:39:29,121][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:39:29,145][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:39:29,170][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:39:29,195][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:39:29,196][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:39:29,196][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:39:30,000][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:39:30,459][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:39:30,961][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:39:31,463][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:39:31,963][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:39:32,463][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:39:32,965][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:39:33,463][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:39:33,962][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:39:34,462][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:39:34,959][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:39:35,457][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:39:35,955][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:39:36,454][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:39:36,958][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:39:37,470][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:39:37,977][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:39:38,481][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:39:38,983][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:39:39,486][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:39:39,987][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:39:40,488][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:39:40,989][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:39:41,489][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:39:41,991][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:39:42,491][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:39:42,995][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:39:43,499][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:39:44,000][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:39:44,500][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:39:45,002][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:39:45,505][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:39:46,006][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:39:46,508][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:39:47,011][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:39:47,511][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:39:48,012][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:39:48,515][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:39:49,013][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:39:49,510][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:39:50,008][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:39:50,510][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:39:51,008][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:39:51,508][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:39:52,008][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:39:52,511][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:39:53,019][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:39:53,521][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:39:54,022][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:39:54,525][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:39:55,026][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:39:55,527][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:39:56,026][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:39:56,526][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:39:57,023][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:39:57,523][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:39:58,058][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:39:58,562][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:39:59,064][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:39:59,566][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:40:00,069][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:40:00,572][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:40:01,080][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:40:01,582][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:40:02,086][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9854 tokens. +[2026-03-26 06:40:02,925][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 61.79%, ΔTime: 00:00:32 +[2026-03-26 06:40:03,736][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:40:03,739][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:40:03,740][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:40:04,487][__main__][INFO] - Iteration 564 took 51s (29.60% Gen, 68.94% Train). Generation: 15s, Training: 35s. Estimated remaining time: 34h 7m 14s. Estimated total time: 42h 46m 30s. Time estimates for 10 more iterations: 8m 33s, 100 more iterations: 1h 25m 33s, 500 more iterations: 7h 7m 45s. +[2026-03-26 06:40:04,490][__main__][INFO] - Starting iteration 564. +[2026-03-26 06:40:04,894][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 56 and human policies 1. +[2026-03-26 06:40:04,895][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:40:22,212][__main__][INFO] - Number of regex retries in iteration 564: 0 +[2026-03-26 06:40:22,213][__main__][INFO] - agents played in iteration 564 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:40:23,088][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:40:23,112][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:40:23,136][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:40:23,160][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:40:23,161][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:40:23,161][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:40:23,955][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:40:24,417][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:40:24,929][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:40:25,432][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:40:25,934][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:40:26,435][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:40:26,936][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:40:27,450][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:40:27,954][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:40:28,461][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:40:28,960][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:40:29,473][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:40:29,979][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:40:30,485][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:40:30,998][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:40:31,500][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:40:32,002][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:40:32,503][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:40:33,004][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:40:33,507][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:40:34,011][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:40:34,517][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:40:35,021][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:40:35,523][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:40:36,026][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:40:36,532][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:40:37,036][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:40:37,543][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:40:38,054][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:40:38,556][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:40:39,063][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:40:39,569][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:40:40,075][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:40:40,583][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:40:41,086][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:40:41,591][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:40:42,097][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:40:42,599][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:40:43,109][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:40:43,613][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:40:44,114][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:40:44,618][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:40:45,117][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:40:45,617][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:40:46,121][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:40:46,621][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:40:47,122][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:40:47,623][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:40:48,127][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:40:48,627][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:40:49,133][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:40:49,640][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:40:50,142][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:40:50,645][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:40:51,147][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:40:51,648][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:40:52,150][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:40:52,651][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:40:53,149][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:40:53,656][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:40:54,167][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:40:54,672][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:40:55,173][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:40:55,676][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:40:56,178][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9996 tokens. +[2026-03-26 06:40:57,017][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:33 +[2026-03-26 06:40:57,789][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:40:57,792][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:40:57,793][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:40:58,524][__main__][INFO] - Iteration 565 took 53s (32.29% Gen, 66.34% Train). Generation: 17s, Training: 35s. Estimated remaining time: 36h 1m 21s. Estimated total time: 44h 41m 31s. Time estimates for 10 more iterations: 8m 56s, 100 more iterations: 1h 29m 23s, 500 more iterations: 7h 26m 55s. +[2026-03-26 06:40:58,527][__main__][INFO] - Starting iteration 565. +[2026-03-26 06:40:58,928][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 56 and human policies 1. +[2026-03-26 06:40:58,929][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:41:24,050][__main__][INFO] - Number of regex retries in iteration 565: 0 +[2026-03-26 06:41:24,051][__main__][INFO] - agents played in iteration 565 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:41:24,815][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:41:24,839][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:41:24,863][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:41:24,887][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:41:24,888][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:41:24,888][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:41:25,691][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:41:26,147][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:41:26,653][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:41:27,149][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:41:27,649][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:41:28,148][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:41:28,651][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:41:29,158][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:41:29,663][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:41:30,181][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:41:30,688][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:41:31,194][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:41:31,698][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:41:32,198][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:41:32,702][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:41:33,201][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:41:33,704][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:41:34,218][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:41:34,720][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:41:35,225][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:41:35,730][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:41:36,235][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:41:36,741][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:41:37,244][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:41:37,744][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:41:38,243][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:41:38,745][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:41:39,246][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:41:39,745][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:41:40,245][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:41:40,743][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:41:41,247][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:41:41,746][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:41:42,246][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:41:42,747][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:41:43,253][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:41:43,767][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:41:44,271][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:41:44,776][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:41:45,280][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:41:45,784][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:41:46,288][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:41:46,788][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:41:47,289][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:41:47,790][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:41:48,292][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:41:48,791][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:41:49,291][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:41:49,792][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:41:50,291][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:41:50,792][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:41:51,308][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:41:51,812][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:41:52,312][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:41:52,812][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:41:53,315][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:41:53,813][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:41:54,313][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:41:54,813][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:41:55,313][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:41:55,812][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:41:56,325][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:41:56,830][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:41:57,332][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:41:57,832][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9975 tokens. +[2026-03-26 06:41:58,667][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 62.05%, ΔTime: 00:00:32 +[2026-03-26 06:41:59,483][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:41:59,486][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:41:59,488][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:42:00,225][__main__][INFO] - Iteration 566 took 1m 1s (40.98% Gen, 57.81% Train). Generation: 25s, Training: 35s. Estimated remaining time: 42h 23m 42s. Estimated total time: 51h 4m 54s. Time estimates for 10 more iterations: 10m 12s, 100 more iterations: 1h 42m 9s, 500 more iterations: 8h 30m 49s. +[2026-03-26 06:42:00,227][__main__][INFO] - Starting iteration 566. +[2026-03-26 06:42:00,629][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 56 and human policies 1. +[2026-03-26 06:42:00,629][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:42:17,186][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 20 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:42:22,164][__main__][INFO] - Number of regex retries in iteration 566: 1 +[2026-03-26 06:42:22,165][__main__][INFO] - agents played in iteration 566 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:42:22,937][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:42:22,962][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:42:22,986][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:42:23,011][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:42:23,012][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:42:23,012][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:42:23,821][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:42:24,275][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:42:24,781][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:42:25,285][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:42:25,785][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:42:26,287][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:42:26,786][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:42:27,281][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:42:27,782][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:42:28,282][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:42:28,805][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:42:29,316][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:42:29,818][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:42:30,327][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:42:30,831][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:42:31,332][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:42:31,835][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:42:32,334][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:42:32,837][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:42:33,342][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:42:33,847][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:42:34,350][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:42:34,867][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:42:35,371][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:42:35,869][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:42:36,366][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:42:36,866][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:42:37,364][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:42:37,862][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:42:38,364][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:42:38,873][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:42:39,386][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:42:39,889][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:42:40,395][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:42:40,899][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:42:41,395][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:42:41,894][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:42:42,395][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:42:42,893][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:42:43,392][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:42:43,895][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:42:44,398][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:42:44,899][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:42:45,399][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:42:45,899][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:42:46,400][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:42:46,902][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:42:47,405][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:42:47,902][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:42:48,406][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:42:48,912][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:42:49,421][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:42:49,933][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:42:50,437][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:42:50,941][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:42:51,445][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:42:51,947][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:42:52,446][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:42:52,948][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:42:53,447][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:42:53,951][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:42:54,448][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:42:54,946][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:42:55,449][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:42:55,946][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9850 tokens. +[2026-03-26 06:42:56,777][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 12.98%, Current % of VRAM taken: 59.40%, Block Peak % of device VRAM: 62.20%, ΔTime: 00:00:32 +[2026-03-26 06:42:57,550][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:42:57,552][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:42:57,554][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:42:58,354][__main__][INFO] - Iteration 567 took 57s (37.31% Gen, 61.30% Train). Generation: 21s, Training: 35s. Estimated remaining time: 39h 24m 7s. Estimated total time: 48h 6m 17s. Time estimates for 10 more iterations: 9m 37s, 100 more iterations: 1h 36m 12s, 500 more iterations: 8h 1m 2s. +[2026-03-26 06:42:58,356][__main__][INFO] - Starting iteration 567. +[2026-03-26 06:42:58,764][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 56 and human policies 1. +[2026-03-26 06:42:58,764][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:43:25,692][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:43:26,551][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:43:30,008][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 20 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:43:36,930][__main__][INFO] - Number of regex retries in iteration 567: 3 +[2026-03-26 06:43:36,931][__main__][INFO] - agents played in iteration 567 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:43:37,805][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:43:37,829][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:43:37,853][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:43:37,877][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:43:37,878][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:43:37,878][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:43:38,679][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:43:39,131][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:43:39,630][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:43:40,128][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:43:40,629][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:43:41,128][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:43:41,627][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:43:42,131][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:43:42,643][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:43:43,146][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:43:43,648][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:43:44,150][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:43:44,656][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:43:45,183][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:43:45,683][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:43:46,183][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:43:46,683][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:43:47,183][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:43:47,682][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:43:48,183][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:43:48,687][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:43:49,189][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:43:49,696][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:43:50,202][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:43:50,704][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:43:51,211][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:43:51,719][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:43:52,222][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:43:52,722][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:43:53,225][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:43:53,728][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:43:54,230][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:43:54,734][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:43:55,235][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:43:55,735][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:43:56,235][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:43:56,736][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:43:57,238][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:43:57,736][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:43:58,235][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:43:58,735][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:43:59,240][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:43:59,743][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:44:00,242][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:44:00,742][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:44:01,241][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:44:01,741][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:44:02,240][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:44:02,740][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:44:03,238][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:44:03,738][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:44:04,238][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:44:04,740][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:44:05,237][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:44:05,735][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:44:06,236][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:44:06,746][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:44:07,247][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:44:07,746][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:44:08,246][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:44:08,744][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:44:09,245][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:44:09,745][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:44:10,244][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:44:10,742][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9955 tokens. +[2026-03-26 06:44:11,575][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 06:44:12,331][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:44:12,333][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:44:12,335][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:44:13,108][__main__][INFO] - Iteration 568 took 1m 14s (51.34% Gen, 47.62% Train). Generation: 38s, Training: 35s. Estimated remaining time: 53h 13m 51s. Estimated total time: 61h 57m 16s. Time estimates for 10 more iterations: 12m 23s, 100 more iterations: 2h 3m 54s, 500 more iterations: 10h 19m 32s. +[2026-03-26 06:44:13,111][__main__][INFO] - Starting iteration 568. +[2026-03-26 06:44:13,531][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 56 and human policies 1. +[2026-03-26 06:44:13,531][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:44:57,982][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:45:05,767][__main__][INFO] - Number of regex retries in iteration 568: 1 +[2026-03-26 06:45:05,768][__main__][INFO] - agents played in iteration 568 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:45:06,527][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:45:06,551][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:45:06,575][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:45:06,599][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:45:06,599][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:45:06,600][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:45:07,404][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:45:07,858][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:45:08,362][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:45:08,861][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:45:09,360][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:45:09,860][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:45:10,362][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:45:10,867][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:45:11,369][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:45:11,871][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:45:12,373][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:45:12,872][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:45:13,371][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:45:13,869][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:45:14,370][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:45:14,868][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:45:15,366][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:45:15,864][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:45:16,367][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:45:16,878][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:45:17,381][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:45:17,882][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:45:18,384][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:45:18,884][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:45:19,388][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:45:19,887][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:45:20,390][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:45:20,888][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:45:21,387][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:45:21,885][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:45:22,384][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:45:22,884][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:45:23,384][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:45:23,890][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:45:24,387][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:45:24,890][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:45:25,390][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:45:25,888][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:45:26,387][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:45:26,886][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:45:27,384][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:45:27,881][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:45:28,377][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:45:28,872][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:45:29,369][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:45:29,865][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:45:30,362][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:45:30,861][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:45:31,359][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:45:31,862][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:45:32,365][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:45:32,865][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:45:33,364][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:45:33,863][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:45:34,364][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:45:34,871][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:45:35,373][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:45:35,873][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:45:36,382][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:45:36,882][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:45:37,383][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:45:37,881][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:45:38,379][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:45:38,880][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:45:39,377][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9897 tokens. +[2026-03-26 06:45:40,228][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.04%, Current % of VRAM taken: 60.46%, Block Peak % of device VRAM: 62.10%, ΔTime: 00:00:32 +[2026-03-26 06:45:41,079][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:45:41,081][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:45:41,083][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:45:41,808][__main__][INFO] - Iteration 569 took 1m 28s (59.17% Gen, 40.00% Train). Generation: 52s, Training: 35s. Estimated remaining time: 64h 49m 2s. Estimated total time: 73h 33m 55s. Time estimates for 10 more iterations: 14m 42s, 100 more iterations: 2h 27m 7s, 500 more iterations: 12h 15m 39s. +[2026-03-26 06:45:41,811][__main__][INFO] - Starting iteration 569. +[2026-03-26 06:45:42,211][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 56 and human policies 1. +[2026-03-26 06:45:42,212][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:45:56,762][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:45:57,004][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:45:58,534][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:46:00,879][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 20 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:46:07,821][__main__][INFO] - Number of regex retries in iteration 569: 4 +[2026-03-26 06:46:07,822][__main__][INFO] - agents played in iteration 569 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:46:08,606][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:46:08,630][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:46:08,654][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:46:08,678][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:46:08,678][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:46:08,679][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:46:09,466][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:46:09,920][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:46:10,421][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:46:10,918][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:46:11,417][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:46:11,916][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:46:12,422][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:46:12,929][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:46:13,429][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:46:13,937][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:46:14,437][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:46:14,944][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:46:15,445][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:46:15,945][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:46:16,444][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:46:16,941][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:46:17,438][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:46:17,940][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:46:18,441][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:46:18,938][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:46:19,437][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:46:19,936][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:46:20,437][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:46:20,937][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:46:21,436][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:46:21,934][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:46:22,429][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:46:22,927][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:46:23,425][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:46:23,923][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:46:24,421][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:46:24,920][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:46:25,419][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:46:25,918][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:46:26,424][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:46:26,924][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:46:27,428][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:46:27,928][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:46:28,428][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:46:28,927][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:46:29,425][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:46:29,920][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:46:30,417][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:46:30,919][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:46:31,417][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:46:31,924][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:46:32,423][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:46:32,924][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:46:33,424][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:46:33,921][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:46:34,419][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:46:34,917][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:46:35,415][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:46:35,912][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:46:36,410][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:46:36,907][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:46:37,406][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:46:37,905][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:46:38,406][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:46:38,907][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:46:39,407][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:46:39,910][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:46:40,416][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:46:40,917][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:46:41,425][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9877 tokens. +[2026-03-26 06:46:42,261][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.15%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.15%, ΔTime: 00:00:32 +[2026-03-26 06:46:43,036][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:46:43,038][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:46:43,040][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:46:43,774][__main__][INFO] - Iteration 570 took 1m 1s (41.60% Gen, 57.21% Train). Generation: 25s, Training: 35s. Estimated remaining time: 42h 32m 15s. Estimated total time: 51h 18m 10s. Time estimates for 10 more iterations: 10m 15s, 100 more iterations: 1h 42m 36s, 500 more iterations: 8h 33m 1s. +[2026-03-26 06:46:43,776][__main__][INFO] - Starting iteration 570. +[2026-03-26 06:46:44,178][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 56 and human policies 1. +[2026-03-26 06:46:44,179][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:47:05,522][__main__][INFO] - Number of regex retries in iteration 570: 0 +[2026-03-26 06:47:05,522][__main__][INFO] - agents played in iteration 570 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:47:06,293][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:47:06,317][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:47:06,340][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:47:06,364][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:47:06,364][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:47:06,365][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:47:07,159][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:47:07,616][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:47:08,117][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:47:08,614][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:47:09,113][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:47:09,636][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:47:10,142][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:47:10,644][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:47:11,143][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:47:11,648][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:47:12,147][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:47:12,650][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:47:13,160][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:47:13,669][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:47:14,171][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:47:14,674][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:47:15,174][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:47:15,677][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:47:16,180][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:47:16,687][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:47:17,190][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:47:17,693][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:47:18,193][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:47:18,694][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:47:19,194][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:47:19,697][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:47:20,200][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:47:20,710][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:47:21,216][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:47:21,722][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:47:22,224][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:47:22,725][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:47:23,226][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:47:23,723][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:47:24,222][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:47:24,731][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:47:25,236][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:47:25,734][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:47:26,234][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:47:26,737][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:47:27,233][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:47:27,730][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:47:28,229][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:47:28,725][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:47:29,221][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:47:29,719][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:47:30,216][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:47:30,710][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:47:31,208][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:47:31,715][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:47:32,215][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:47:32,713][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:47:33,212][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:47:33,709][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:47:34,207][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:47:34,705][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:47:35,201][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:47:35,703][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:47:36,207][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:47:36,710][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:47:37,218][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:47:37,718][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:47:38,224][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:47:38,724][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:47:39,225][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9932 tokens. +[2026-03-26 06:47:40,084][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.18%, ΔTime: 00:00:32 +[2026-03-26 06:47:40,854][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:47:40,856][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:47:40,858][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:47:44,094][__main__][INFO] - Iteration 571 took 59s (35.62% Gen, 58.97% Train). Generation: 21s, Training: 35s. Estimated remaining time: 41h 8m 52s. Estimated total time: 49h 55m 47s. Time estimates for 10 more iterations: 9m 59s, 100 more iterations: 1h 39m 51s, 500 more iterations: 8h 19m 17s. +[2026-03-26 06:47:44,096][__main__][INFO] - Starting iteration 571. +[2026-03-26 06:47:44,499][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 57 and human policies 1. +[2026-03-26 06:47:44,500][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:48:00,818][__main__][INFO] - Number of regex retries in iteration 571: 0 +[2026-03-26 06:48:00,819][__main__][INFO] - agents played in iteration 571 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:48:01,577][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:48:01,600][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:48:01,624][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:48:01,648][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:48:01,648][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:48:01,649][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:48:02,461][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:48:02,916][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:48:03,425][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:48:03,920][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:48:04,429][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:48:04,942][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:48:05,450][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:48:05,954][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:48:06,456][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:48:06,962][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:48:07,466][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:48:07,969][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:48:08,470][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:48:08,972][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:48:09,472][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:48:09,973][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:48:10,475][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:48:10,980][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:48:11,481][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:48:11,981][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:48:12,486][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:48:12,991][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:48:13,506][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:48:14,014][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:48:14,516][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:48:15,018][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:48:15,518][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:48:16,022][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:48:16,527][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:48:17,027][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:48:17,528][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:48:18,031][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:48:18,532][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:48:19,036][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:48:19,535][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:48:20,034][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:48:20,536][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:48:21,036][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:48:21,534][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:48:22,034][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:48:22,532][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:48:23,034][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:48:23,531][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:48:24,033][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:48:24,535][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:48:25,040][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:48:25,541][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:48:26,047][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:48:26,555][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:48:27,055][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:48:27,555][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:48:28,057][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:48:28,554][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:48:29,050][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:48:29,548][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:48:30,048][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:48:30,549][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:48:31,049][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:48:31,549][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:48:32,048][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:48:32,547][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:48:33,046][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:48:33,544][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:48:34,039][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:48:34,538][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9956 tokens. +[2026-03-26 06:48:35,397][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 06:48:36,169][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:48:36,172][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:48:36,174][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:48:37,020][__main__][INFO] - Iteration 572 took 52s (31.07% Gen, 67.31% Train). Generation: 16s, Training: 35s. Estimated remaining time: 34h 58m 16s. Estimated total time: 43h 46m 5s. Time estimates for 10 more iterations: 8m 45s, 100 more iterations: 1h 27m 32s, 500 more iterations: 7h 17m 40s. +[2026-03-26 06:48:37,022][__main__][INFO] - Starting iteration 572. +[2026-03-26 06:48:37,463][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 57 and human policies 1. +[2026-03-26 06:48:37,464][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:48:57,847][__main__][INFO] - Number of regex retries in iteration 572: 0 +[2026-03-26 06:48:57,849][__main__][INFO] - agents played in iteration 572 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:48:58,617][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:48:58,641][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:48:58,665][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:48:58,688][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:48:58,689][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:48:58,689][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:48:59,494][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:48:59,945][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:49:00,451][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:49:00,964][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:49:01,469][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:49:01,969][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:49:02,479][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:49:02,980][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:49:03,481][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:49:03,982][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:49:04,491][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:49:05,002][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:49:05,504][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:49:06,009][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:49:06,512][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:49:07,014][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:49:07,518][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:49:08,022][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:49:08,521][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:49:09,017][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:49:09,515][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:49:10,025][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:49:10,531][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:49:11,030][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:49:11,530][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:49:12,038][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:49:12,539][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:49:13,038][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:49:13,537][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:49:14,036][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:49:14,535][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:49:15,034][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:49:15,532][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:49:16,030][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:49:16,530][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:49:17,040][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:49:17,541][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:49:18,043][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:49:18,545][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:49:19,044][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:49:19,544][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:49:20,041][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:49:20,542][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:49:21,039][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:49:21,544][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:49:22,043][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:49:22,539][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:49:23,041][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:49:23,538][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:49:24,032][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:49:24,534][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:49:25,034][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:49:25,535][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:49:26,037][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:49:26,542][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:49:27,036][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:49:27,535][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:49:28,038][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:49:28,538][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:49:29,038][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:49:29,538][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:49:30,037][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:49:30,539][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:49:31,041][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:49:31,542][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9867 tokens. +[2026-03-26 06:49:32,404][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.04%, Current % of VRAM taken: 60.46%, Block Peak % of device VRAM: 62.13%, ΔTime: 00:00:32 +[2026-03-26 06:49:33,204][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:49:33,206][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:49:33,208][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:49:33,910][__main__][INFO] - Iteration 573 took 56s (36.11% Gen, 62.64% Train). Generation: 20s, Training: 35s. Estimated remaining time: 38h 13m 35s. Estimated total time: 47h 2m 20s. Time estimates for 10 more iterations: 9m 24s, 100 more iterations: 1h 34m 4s, 500 more iterations: 7h 50m 23s. +[2026-03-26 06:49:33,912][__main__][INFO] - Starting iteration 573. +[2026-03-26 06:49:34,337][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 57 and human policies 1. +[2026-03-26 06:49:34,338][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:49:50,421][__main__][INFO] - Number of regex retries in iteration 573: 0 +[2026-03-26 06:49:50,423][__main__][INFO] - agents played in iteration 573 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:49:51,195][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:49:51,219][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:49:51,243][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:49:51,267][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:49:51,268][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:49:51,268][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:49:52,084][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:49:52,539][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:49:53,053][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:49:53,554][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:49:54,056][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:49:54,556][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:49:55,061][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:49:55,559][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:49:56,055][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:49:56,562][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:49:57,067][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:49:57,569][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:49:58,071][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:49:58,572][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:49:59,073][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:49:59,572][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:50:00,074][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:50:00,586][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:50:01,090][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:50:01,591][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:50:02,096][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:50:02,599][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:50:03,102][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:50:03,602][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:50:04,102][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:50:04,604][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:50:05,101][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:50:05,602][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:50:06,106][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:50:06,630][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:50:07,132][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:50:07,633][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:50:08,136][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:50:08,641][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:50:09,144][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:50:09,642][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:50:10,139][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:50:10,638][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:50:11,140][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:50:11,634][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:50:12,132][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:50:12,630][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:50:13,129][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:50:13,629][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:50:14,129][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:50:14,628][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:50:15,122][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:50:15,621][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:50:16,125][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:50:16,628][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:50:17,130][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:50:17,637][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:50:18,131][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:50:18,634][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:50:19,135][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:50:19,632][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:50:20,132][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:50:20,630][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:50:21,128][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:50:21,632][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:50:22,139][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:50:22,643][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:50:23,141][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:50:23,641][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:50:24,140][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9917 tokens. +[2026-03-26 06:50:25,096][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.28%, ΔTime: 00:00:33 +[2026-03-26 06:50:25,856][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:50:25,859][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:50:25,860][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:50:26,540][__main__][INFO] - Iteration 574 took 52s (30.81% Gen, 67.88% Train). Generation: 16s, Training: 35s. Estimated remaining time: 34h 40m 31s. Estimated total time: 43h 30m 9s. Time estimates for 10 more iterations: 8m 42s, 100 more iterations: 1h 27m 0s, 500 more iterations: 7h 15m 1s. +[2026-03-26 06:50:26,542][__main__][INFO] - Starting iteration 574. +[2026-03-26 06:50:26,949][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 57 and human policies 1. +[2026-03-26 06:50:26,950][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:50:58,529][__main__][INFO] - Number of regex retries in iteration 574: 0 +[2026-03-26 06:50:58,530][__main__][INFO] - agents played in iteration 574 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:50:59,313][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:50:59,337][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:50:59,360][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:50:59,383][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:50:59,384][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:50:59,384][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:51:00,191][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:51:00,652][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:51:01,155][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:51:01,658][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:51:02,154][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:51:02,651][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:51:03,149][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:51:03,657][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:51:04,156][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:51:04,654][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:51:05,158][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:51:05,651][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:51:06,148][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:51:06,644][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:51:07,140][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:51:07,642][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:51:08,138][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:51:08,646][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:51:09,141][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:51:09,640][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:51:10,140][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:51:10,639][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:51:11,146][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:51:11,645][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:51:12,145][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:51:12,643][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:51:13,141][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:51:13,636][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:51:14,133][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:51:14,634][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:51:15,135][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:51:15,630][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:51:16,128][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:51:16,628][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:51:17,127][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:51:17,627][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:51:18,127][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:51:18,636][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:51:19,138][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:51:19,639][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:51:20,141][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:51:20,640][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:51:21,136][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:51:21,631][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:51:22,126][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:51:22,631][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:51:23,130][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:51:23,627][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:51:24,123][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:51:24,621][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:51:25,120][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:51:25,616][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:51:26,114][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:51:26,616][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:51:27,121][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:51:27,622][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:51:28,119][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:51:28,616][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:51:29,119][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:51:29,624][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:51:30,128][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:51:30,626][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:51:31,123][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:51:31,617][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:51:32,122][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9787 tokens. +[2026-03-26 06:51:32,974][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.05%, Current % of VRAM taken: 60.47%, Block Peak % of device VRAM: 62.03%, ΔTime: 00:00:32 +[2026-03-26 06:51:33,754][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:51:33,757][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:51:33,758][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:51:34,492][__main__][INFO] - Iteration 575 took 1m 7s (46.76% Gen, 52.16% Train). Generation: 31s, Training: 35s. Estimated remaining time: 47h 26m 26s. Estimated total time: 56h 17m 12s. Time estimates for 10 more iterations: 11m 15s, 100 more iterations: 1h 52m 34s, 500 more iterations: 9h 22m 52s. +[2026-03-26 06:51:34,494][__main__][INFO] - Starting iteration 575. +[2026-03-26 06:51:34,896][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 57 and human policies 1. +[2026-03-26 06:51:34,897][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:51:59,923][__main__][INFO] - Number of regex retries in iteration 575: 0 +[2026-03-26 06:51:59,924][__main__][INFO] - agents played in iteration 575 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:52:00,698][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:52:00,722][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:52:00,745][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:52:00,768][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:52:00,769][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:52:00,770][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:52:01,579][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:52:02,036][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:52:02,549][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:52:03,050][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:52:03,555][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:52:04,055][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:52:04,556][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:52:05,058][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:52:05,559][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:52:06,055][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:52:06,552][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:52:07,055][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:52:07,553][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:52:08,051][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:52:08,549][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:52:09,050][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:52:09,545][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:52:10,047][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:52:10,549][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:52:11,047][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:52:11,545][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:52:12,047][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:52:12,545][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:52:13,045][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:52:13,546][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:52:14,047][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:52:16,045][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:52:16,543][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:52:17,042][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:52:17,539][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:52:18,038][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:52:18,537][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:52:19,035][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:52:19,531][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:52:20,032][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:52:20,530][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:52:21,030][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:52:21,526][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:52:22,022][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:52:22,521][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:52:23,017][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:52:23,517][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:52:24,017][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:52:24,517][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:52:25,016][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:52:25,514][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:52:26,015][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:52:26,515][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:52:27,016][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:52:27,508][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:52:28,008][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:52:28,508][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:52:29,001][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:52:29,492][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:52:29,988][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:52:30,486][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:52:30,984][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:52:31,483][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:52:31,990][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:52:32,490][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:52:32,993][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:52:33,490][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:52:33,989][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:52:34,489][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:52:34,993][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10013 tokens. +[2026-03-26 06:52:35,844][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.11%, ΔTime: 00:00:34 +[2026-03-26 06:52:36,619][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:52:36,621][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:52:36,623][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:52:37,387][__main__][INFO] - Iteration 576 took 1m 2s (40.05% Gen, 58.73% Train). Generation: 25s, Training: 36s. Estimated remaining time: 43h 12m 46s. Estimated total time: 52h 4m 35s. Time estimates for 10 more iterations: 10m 24s, 100 more iterations: 1h 44m 9s, 500 more iterations: 8h 40m 45s. +[2026-03-26 06:52:37,390][__main__][INFO] - Starting iteration 576. +[2026-03-26 06:52:37,792][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 57 and human policies 1. +[2026-03-26 06:52:37,793][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:52:52,950][__main__][INFO] - Number of regex retries in iteration 576: 0 +[2026-03-26 06:52:52,951][__main__][INFO] - agents played in iteration 576 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:52:53,727][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:52:53,751][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:52:53,774][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:52:53,798][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:52:53,798][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:52:53,799][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:52:54,617][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:52:55,070][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:52:55,573][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:52:56,081][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:52:56,581][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:52:57,080][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:52:57,576][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:52:58,085][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:52:58,586][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:52:59,090][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:52:59,590][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:53:00,091][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:53:00,591][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:53:01,089][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:53:01,586][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:53:02,085][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:53:02,588][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:53:03,085][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:53:03,583][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:53:04,077][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:53:04,573][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:53:05,069][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:53:05,566][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:53:06,062][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:53:06,559][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:53:07,056][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:53:07,557][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:53:08,054][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:53:08,554][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:53:09,050][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:53:09,555][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:53:10,053][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:53:10,552][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:53:11,049][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:53:11,545][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:53:12,041][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:53:12,537][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:53:13,031][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:53:13,533][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:53:14,029][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:53:14,525][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:53:15,023][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:53:15,520][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:53:16,016][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:53:16,514][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:53:17,012][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:53:17,510][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:53:18,013][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:53:18,510][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:53:19,010][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:53:19,512][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:53:20,011][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:53:20,508][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:53:21,004][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:53:21,499][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:53:21,998][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:53:22,496][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:53:22,995][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:53:23,494][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:53:23,991][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:53:24,490][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:53:24,987][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:53:25,484][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:53:25,981][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:53:26,478][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9872 tokens. +[2026-03-26 06:53:27,328][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.97%, Current % of VRAM taken: 60.40%, Block Peak % of device VRAM: 62.03%, ΔTime: 00:00:32 +[2026-03-26 06:53:28,103][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:53:28,106][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:53:28,107][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:53:28,837][__main__][INFO] - Iteration 577 took 51s (29.70% Gen, 68.87% Train). Generation: 15s, Training: 35s. Estimated remaining time: 33h 39m 38s. Estimated total time: 42h 32m 19s. Time estimates for 10 more iterations: 8m 30s, 100 more iterations: 1h 25m 4s, 500 more iterations: 7h 5m 23s. +[2026-03-26 06:53:28,840][__main__][INFO] - Starting iteration 577. +[2026-03-26 06:53:29,242][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 57 and human policies 1. +[2026-03-26 06:53:29,243][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:53:34,340][mllm.models.large_language_model_local][WARNING] - Response Proposal: 5 hats, 5 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:53:45,638][__main__][INFO] - Number of regex retries in iteration 577: 1 +[2026-03-26 06:53:45,639][__main__][INFO] - agents played in iteration 577 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:53:46,410][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:53:46,434][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:53:46,457][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:53:46,480][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:53:46,481][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:53:46,482][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:53:47,317][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:53:47,777][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:53:48,278][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:53:48,777][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:53:49,274][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:53:49,776][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:53:50,275][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:53:50,773][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:53:51,271][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:53:51,768][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:53:52,269][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:53:52,774][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:53:53,274][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:53:53,775][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:53:54,272][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:53:54,771][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:53:55,270][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:53:55,780][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:53:56,285][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:53:56,783][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:53:57,282][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:53:57,779][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:53:58,279][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:53:58,784][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:53:59,281][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:53:59,782][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:54:00,284][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:54:00,783][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:54:01,282][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:54:01,778][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:54:02,277][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:54:02,776][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:54:03,279][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:54:03,775][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:54:04,273][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:54:04,768][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:54:05,263][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:54:05,759][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:54:06,253][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:54:06,749][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:54:07,246][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:54:07,743][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:54:08,240][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:54:08,740][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:54:09,234][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:54:09,734][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:54:10,237][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:54:10,735][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:54:11,235][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:54:11,733][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:54:12,231][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:54:12,728][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:54:13,224][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:54:13,719][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:54:14,241][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:54:14,736][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:54:15,232][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:54:15,734][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:54:16,231][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:54:16,728][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:54:17,224][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:54:17,720][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:54:18,217][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:54:18,713][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:54:19,210][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9910 tokens. +[2026-03-26 06:54:20,071][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 61.96%, ΔTime: 00:00:32 +[2026-03-26 06:54:20,842][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:54:20,845][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:54:20,847][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:54:21,617][__main__][INFO] - Iteration 578 took 52s (31.31% Gen, 67.22% Train). Generation: 16s, Training: 35s. Estimated remaining time: 34h 45m 16s. Estimated total time: 43h 38m 48s. Time estimates for 10 more iterations: 8m 43s, 100 more iterations: 1h 27m 17s, 500 more iterations: 7h 16m 28s. +[2026-03-26 06:54:21,620][__main__][INFO] - Starting iteration 578. +[2026-03-26 06:54:22,019][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 57 and human policies 1. +[2026-03-26 06:54:22,019][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:54:30,226][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:54:37,504][__main__][INFO] - Number of regex retries in iteration 578: 1 +[2026-03-26 06:54:37,504][__main__][INFO] - agents played in iteration 578 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:54:38,270][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:54:38,294][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:54:38,317][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:54:38,341][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:54:38,341][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:54:38,342][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:54:39,157][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:54:39,613][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:54:40,116][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:54:40,616][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:54:41,116][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:54:41,614][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:54:42,111][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:54:42,610][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:54:43,109][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:54:43,607][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:54:44,107][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:54:44,602][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:54:45,100][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:54:45,596][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:54:46,095][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:54:46,592][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:54:47,090][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:54:47,589][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:54:48,090][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:54:48,588][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:54:49,083][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:54:49,584][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:54:50,082][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:54:50,581][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:54:51,077][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:54:51,574][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:54:52,071][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:54:52,567][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:54:53,064][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:54:53,562][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:54:54,060][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:54:54,557][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:54:55,054][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:54:55,553][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:54:56,052][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:54:56,551][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:54:57,049][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:54:57,546][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:54:58,044][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:54:58,541][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:54:59,037][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:54:59,534][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:55:00,031][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:55:00,528][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:55:01,025][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:55:01,523][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:55:02,020][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:55:02,518][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:55:03,016][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:55:03,516][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:55:04,011][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:55:04,508][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:55:05,006][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:55:05,500][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:55:05,996][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:55:06,493][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:55:06,990][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:55:07,493][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:55:07,990][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:55:08,487][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:55:08,984][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:55:09,479][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:55:09,975][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:55:10,472][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:55:10,972][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9900 tokens. +[2026-03-26 06:55:11,815][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.61%, Block Peak % of device VRAM: 61.98%, ΔTime: 00:00:32 +[2026-03-26 06:55:12,580][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:55:12,582][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:55:12,584][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:55:13,357][__main__][INFO] - Iteration 579 took 51s (30.16% Gen, 68.33% Train). Generation: 15s, Training: 35s. Estimated remaining time: 33h 52m 31s. Estimated total time: 42h 46m 56s. Time estimates for 10 more iterations: 8m 33s, 100 more iterations: 1h 25m 33s, 500 more iterations: 7h 7m 49s. +[2026-03-26 06:55:13,359][__main__][INFO] - Starting iteration 579. +[2026-03-26 06:55:13,760][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 57 and human policies 1. +[2026-03-26 06:55:13,760][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:55:20,908][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 06:55:30,646][__main__][INFO] - Number of regex retries in iteration 579: 1 +[2026-03-26 06:55:30,647][__main__][INFO] - agents played in iteration 579 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:55:31,419][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:55:31,443][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:55:31,466][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:55:31,489][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:55:31,490][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:55:31,490][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:55:32,299][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:55:32,753][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:55:33,258][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:55:33,755][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:55:34,254][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:55:34,751][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:55:35,248][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:55:35,746][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:55:36,246][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:55:36,748][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:55:37,248][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:55:37,744][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:55:38,245][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:55:38,749][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:55:39,253][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:55:39,753][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:55:40,255][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:55:40,754][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:55:41,254][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:55:41,754][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:55:42,251][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:55:42,747][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:55:43,244][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:55:43,744][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:55:44,241][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:55:44,738][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:55:45,235][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:55:45,736][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:55:46,233][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:55:46,730][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:55:47,232][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:55:47,730][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:55:48,228][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:55:48,725][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:55:49,221][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:55:49,716][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:55:50,211][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:55:50,708][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:55:51,206][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:55:51,702][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:55:52,193][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:55:52,689][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:55:53,184][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:55:53,681][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:55:54,177][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:55:54,673][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:55:55,172][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:55:55,668][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:55:56,165][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:55:56,655][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:55:57,150][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:55:57,646][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:55:58,139][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:55:58,634][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:55:59,125][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:55:59,619][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:56:00,111][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:56:00,609][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:56:01,104][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:56:01,601][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:56:02,099][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:56:02,599][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:56:03,096][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:56:03,594][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:56:04,089][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9865 tokens. +[2026-03-26 06:56:04,968][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.98%, Current % of VRAM taken: 60.41%, Block Peak % of device VRAM: 61.99%, ΔTime: 00:00:32 +[2026-03-26 06:56:05,729][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:56:05,731][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:56:05,733][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:56:06,452][__main__][INFO] - Iteration 580 took 52s (32.05% Gen, 66.59% Train). Generation: 16s, Training: 35s. Estimated remaining time: 34h 59m 21s. Estimated total time: 43h 54m 38s. Time estimates for 10 more iterations: 8m 46s, 100 more iterations: 1h 27m 49s, 500 more iterations: 7h 19m 6s. +[2026-03-26 06:56:06,454][__main__][INFO] - Starting iteration 580. +[2026-03-26 06:56:06,856][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 57 and human policies 1. +[2026-03-26 06:56:06,856][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:56:21,960][__main__][INFO] - Number of regex retries in iteration 580: 0 +[2026-03-26 06:56:21,961][__main__][INFO] - agents played in iteration 580 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:56:22,729][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:56:22,752][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:56:22,776][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:56:22,799][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:56:22,800][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:56:22,800][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:56:23,614][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:56:24,068][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:56:24,570][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:56:25,067][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:56:25,563][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:56:26,060][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:56:26,557][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:56:27,054][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:56:27,551][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:56:28,055][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:56:28,562][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:56:29,063][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:56:29,567][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:56:30,066][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:56:30,566][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:56:31,068][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:56:31,570][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:56:32,070][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:56:32,568][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:56:33,067][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:56:33,559][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:56:34,055][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:56:34,552][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:56:35,049][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:56:35,547][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:56:36,045][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:56:36,541][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:56:37,037][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:56:37,533][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:56:38,030][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:56:38,528][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:56:39,025][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:56:39,523][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:56:40,021][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:56:40,520][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:56:41,018][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:56:41,517][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:56:42,017][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:56:42,515][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:56:43,014][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:56:43,516][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:56:44,013][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:56:44,508][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:56:45,009][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:56:45,506][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:56:46,007][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:56:46,505][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:56:47,005][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:56:47,502][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:56:48,004][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:56:48,507][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:56:49,008][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:56:49,507][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:56:50,008][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:56:50,508][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:56:51,008][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:56:51,509][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:56:52,009][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:56:52,507][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:56:53,008][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:56:53,506][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:56:54,002][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:56:54,499][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:56:54,998][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:56:55,496][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10074 tokens. +[2026-03-26 06:56:56,354][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.16%, Current % of VRAM taken: 60.58%, Block Peak % of device VRAM: 62.32%, ΔTime: 00:00:32 +[2026-03-26 06:56:57,120][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:56:57,122][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:56:57,124][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:56:58,414][__main__][INFO] - Iteration 581 took 51s (29.30% Gen, 68.20% Train). Generation: 15s, Training: 35s. Estimated remaining time: 34h 1m 48s. Estimated total time: 42h 57m 58s. Time estimates for 10 more iterations: 8m 35s, 100 more iterations: 1h 25m 55s, 500 more iterations: 7h 9m 39s. +[2026-03-26 06:56:58,417][__main__][INFO] - Starting iteration 581. +[2026-03-26 06:56:58,819][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 58 and human policies 1. +[2026-03-26 06:56:58,819][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:57:14,265][__main__][INFO] - Number of regex retries in iteration 581: 0 +[2026-03-26 06:57:14,266][__main__][INFO] - agents played in iteration 581 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:57:15,029][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:57:15,053][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:57:15,077][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:57:15,101][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:57:15,102][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:57:15,102][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:57:15,930][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:57:16,383][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:57:16,886][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:57:17,385][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:57:17,885][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:57:18,383][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:57:18,884][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:57:19,388][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:57:19,887][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:57:20,387][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:57:20,887][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:57:21,386][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:57:21,882][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:57:22,378][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:57:22,881][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:57:23,379][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:57:23,876][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:57:24,372][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:57:24,869][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:57:25,372][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:57:25,867][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:57:26,362][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:57:26,861][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:57:27,357][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:57:27,852][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:57:28,349][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:57:28,845][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:57:29,340][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:57:29,837][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:57:30,334][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:57:30,830][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:57:31,326][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:57:31,822][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:57:32,318][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:57:32,811][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:57:33,308][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:57:33,809][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:57:34,303][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:57:34,795][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:57:35,289][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:57:35,780][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:57:36,272][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:57:36,767][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:57:37,261][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:57:37,756][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:57:38,252][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:57:38,747][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:57:39,242][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:57:39,738][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:57:40,233][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:57:40,728][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:57:41,222][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:57:41,717][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:57:42,210][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:57:42,710][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:57:43,205][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:57:43,696][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:57:44,190][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:57:44,681][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:57:45,175][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:57:45,669][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:57:46,164][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:57:46,660][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:57:47,154][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:57:47,650][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9793 tokens. +[2026-03-26 06:57:48,527][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.99%, Current % of VRAM taken: 60.41%, Block Peak % of device VRAM: 61.96%, ΔTime: 00:00:32 +[2026-03-26 06:57:49,294][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:57:49,296][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:57:49,298][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:57:50,023][__main__][INFO] - Iteration 582 took 51s (30.17% Gen, 68.41% Train). Generation: 15s, Training: 35s. Estimated remaining time: 33h 43m 13s. Estimated total time: 42h 40m 14s. Time estimates for 10 more iterations: 8m 32s, 100 more iterations: 1h 25m 20s, 500 more iterations: 7h 6m 42s. +[2026-03-26 06:57:50,025][__main__][INFO] - Starting iteration 582. +[2026-03-26 06:57:50,428][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 58 and human policies 1. +[2026-03-26 06:57:50,428][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:58:09,740][__main__][INFO] - Number of regex retries in iteration 582: 0 +[2026-03-26 06:58:09,741][__main__][INFO] - agents played in iteration 582 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:58:10,523][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:58:10,546][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:58:10,569][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:58:10,593][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:58:10,593][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:58:10,594][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:58:11,407][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:58:11,859][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:58:12,362][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:58:12,861][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:58:13,359][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:58:13,856][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:58:14,354][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:58:14,851][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:58:15,353][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:58:15,852][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:58:16,344][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:58:16,835][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:58:17,331][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:58:17,824][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:58:18,318][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:58:18,812][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:58:19,314][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:58:19,820][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:58:20,319][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:58:20,817][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:58:21,314][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:58:21,809][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:58:22,306][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:58:22,801][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:58:23,295][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:58:23,792][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:58:24,293][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:58:24,791][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:58:25,290][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:58:25,787][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:58:26,286][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:58:26,779][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:58:27,275][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:58:27,770][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:58:28,266][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:58:28,762][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:58:29,259][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:58:29,754][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:58:30,251][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:58:30,747][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:58:31,244][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:58:31,743][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:58:32,258][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:58:32,754][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:58:33,248][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:58:33,746][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:58:34,245][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:58:34,745][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:58:35,242][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:58:35,739][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:58:36,233][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:58:36,730][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:58:37,225][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:58:37,722][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:58:38,220][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:58:38,716][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:58:39,216][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:58:39,715][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:58:40,211][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:58:40,709][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:58:41,208][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:58:41,706][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:58:42,205][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:58:42,702][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:58:43,199][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9816 tokens. +[2026-03-26 06:58:44,032][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 61.95%, ΔTime: 00:00:32 +[2026-03-26 06:58:44,808][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:58:44,810][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:58:44,812][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:58:45,545][__main__][INFO] - Iteration 583 took 55s (35.04% Gen, 63.63% Train). Generation: 19s, Training: 35s. Estimated remaining time: 36h 57m 57s. Estimated total time: 45h 55m 54s. Time estimates for 10 more iterations: 9m 11s, 100 more iterations: 1h 31m 51s, 500 more iterations: 7h 39m 19s. +[2026-03-26 06:58:45,548][__main__][INFO] - Starting iteration 583. +[2026-03-26 06:58:45,945][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 58 and human policies 1. +[2026-03-26 06:58:45,946][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:59:01,630][__main__][INFO] - Number of regex retries in iteration 583: 0 +[2026-03-26 06:59:01,631][__main__][INFO] - agents played in iteration 583 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:59:02,404][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:59:02,427][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:59:02,451][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:59:02,474][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:59:02,475][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:59:02,475][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:59:03,283][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:59:03,731][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:59:04,235][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:59:04,733][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:59:05,230][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:59:05,724][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:59:06,217][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:59:06,710][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:59:07,203][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:59:07,706][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:59:08,207][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:59:08,707][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 06:59:09,204][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 06:59:09,704][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 06:59:10,201][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 06:59:10,700][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 06:59:11,200][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 06:59:11,701][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 06:59:12,201][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 06:59:12,698][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 06:59:13,196][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 06:59:13,693][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 06:59:14,190][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 06:59:14,688][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 06:59:15,186][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 06:59:15,688][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 06:59:16,186][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 06:59:16,684][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 06:59:17,182][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 06:59:17,679][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 06:59:18,176][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 06:59:18,671][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 06:59:19,199][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 06:59:19,699][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 06:59:20,196][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 06:59:20,694][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 06:59:21,190][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 06:59:21,689][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 06:59:22,187][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 06:59:22,691][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 06:59:23,186][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 06:59:23,680][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 06:59:24,176][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 06:59:24,671][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 06:59:25,168][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 06:59:25,664][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 06:59:26,159][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 06:59:26,654][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 06:59:27,150][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 06:59:27,650][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 06:59:28,150][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 06:59:28,649][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 06:59:29,148][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 06:59:29,647][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 06:59:30,147][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 06:59:30,647][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 06:59:31,147][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 06:59:31,645][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 06:59:32,143][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 06:59:32,642][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 06:59:33,140][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 06:59:33,638][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 06:59:34,136][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 06:59:34,633][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 06:59:35,131][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9868 tokens. +[2026-03-26 06:59:35,972][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 61.97%, ΔTime: 00:00:32 +[2026-03-26 06:59:36,738][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 06:59:36,741][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 06:59:36,743][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 06:59:37,457][__main__][INFO] - Iteration 584 took 51s (30.45% Gen, 68.16% Train). Generation: 15s, Training: 35s. Estimated remaining time: 33h 56m 46s. Estimated total time: 42h 55m 35s. Time estimates for 10 more iterations: 8m 35s, 100 more iterations: 1h 25m 51s, 500 more iterations: 7h 9m 15s. +[2026-03-26 06:59:37,459][__main__][INFO] - Starting iteration 584. +[2026-03-26 06:59:37,865][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 58 and human policies 1. +[2026-03-26 06:59:37,866][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 06:59:52,442][__main__][INFO] - Number of regex retries in iteration 584: 0 +[2026-03-26 06:59:52,443][__main__][INFO] - agents played in iteration 584 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 06:59:53,218][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:59:53,242][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:59:53,265][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:59:53,289][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 06:59:53,289][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 06:59:53,290][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 06:59:54,088][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 06:59:54,542][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 06:59:55,042][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 06:59:55,537][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 06:59:56,036][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 06:59:56,538][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 06:59:57,036][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 06:59:57,532][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 06:59:58,031][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 06:59:58,533][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 06:59:59,031][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 06:59:59,532][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:00:00,035][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:00:00,533][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:00:01,033][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:00:01,534][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:00:02,036][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:00:02,531][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:00:03,025][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:00:03,517][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:00:04,010][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:00:04,504][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:00:04,998][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:00:05,491][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:00:05,991][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:00:06,487][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:00:06,984][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:00:07,481][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:00:07,977][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:00:08,473][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:00:08,970][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:00:09,466][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:00:09,962][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:00:10,459][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:00:10,955][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:00:11,453][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:00:11,950][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:00:12,448][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:00:12,946][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:00:13,445][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:00:13,942][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:00:14,436][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:00:14,933][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:00:15,428][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:00:15,925][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:00:16,420][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:00:16,916][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:00:17,413][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:00:17,905][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:00:18,398][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:00:18,892][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:00:19,387][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:00:19,881][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:00:20,378][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:00:20,873][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:00:21,370][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:00:21,864][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:00:22,361][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:00:22,860][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:00:23,357][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:00:23,855][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:00:24,353][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:00:24,850][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:00:25,351][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:00:25,847][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9877 tokens. +[2026-03-26 07:00:26,682][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.04%, Current % of VRAM taken: 60.47%, Block Peak % of device VRAM: 61.95%, ΔTime: 00:00:32 +[2026-03-26 07:00:27,443][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:00:27,445][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:00:27,447][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:00:28,146][__main__][INFO] - Iteration 585 took 50s (28.99% Gen, 69.62% Train). Generation: 14s, Training: 35s. Estimated remaining time: 32h 54m 25s. Estimated total time: 41h 54m 5s. Time estimates for 10 more iterations: 8m 22s, 100 more iterations: 1h 23m 48s, 500 more iterations: 6h 59m 0s. +[2026-03-26 07:00:28,148][__main__][INFO] - Starting iteration 585. +[2026-03-26 07:00:28,548][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 58 and human policies 1. +[2026-03-26 07:00:28,548][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:00:42,662][__main__][INFO] - Number of regex retries in iteration 585: 0 +[2026-03-26 07:00:42,663][__main__][INFO] - agents played in iteration 585 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:00:43,535][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:00:43,559][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:00:43,583][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:00:43,608][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:00:43,608][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:00:43,609][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:00:44,443][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:00:44,899][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:00:45,409][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:00:45,909][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:00:46,415][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:00:46,914][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:00:47,413][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:00:47,912][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:00:48,412][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:00:48,914][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:00:49,412][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:00:49,914][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:00:50,416][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:00:50,921][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:00:51,417][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:00:51,914][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:00:52,416][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:00:52,912][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:00:53,412][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:00:53,909][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:00:54,406][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:00:54,901][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:00:55,408][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:00:55,907][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:00:56,407][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:00:56,908][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:00:57,411][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:00:57,915][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:00:58,415][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:00:58,916][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:00:59,416][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:00:59,916][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:01:00,416][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:01:00,913][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:01:01,410][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:01:01,906][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:01:02,402][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:01:02,896][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:01:03,395][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:01:03,892][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:01:04,388][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:01:04,884][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:01:05,380][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:01:05,880][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:01:06,370][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:01:06,866][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:01:07,362][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:01:07,858][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:01:08,355][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:01:08,852][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:01:09,349][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:01:09,847][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:01:10,345][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:01:10,842][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:01:11,337][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:01:11,831][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:01:12,328][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:01:12,825][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:01:13,322][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:01:13,814][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:01:14,310][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:01:14,808][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:01:15,306][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:01:15,805][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:01:16,301][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9933 tokens. +[2026-03-26 07:01:17,130][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 62.29%, ΔTime: 00:00:32 +[2026-03-26 07:01:17,975][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:01:17,977][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:01:17,979][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:01:18,713][__main__][INFO] - Iteration 586 took 50s (28.14% Gen, 70.40% Train). Generation: 14s, Training: 35s. Estimated remaining time: 32h 47m 46s. Estimated total time: 41h 48m 16s. Time estimates for 10 more iterations: 8m 21s, 100 more iterations: 1h 23m 36s, 500 more iterations: 6h 58m 2s. +[2026-03-26 07:01:18,715][__main__][INFO] - Starting iteration 586. +[2026-03-26 07:01:19,116][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 58 and human policies 1. +[2026-03-26 07:01:19,117][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:01:23,392][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 07:01:33,755][__main__][INFO] - Number of regex retries in iteration 586: 1 +[2026-03-26 07:01:33,756][__main__][INFO] - agents played in iteration 586 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:01:34,562][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:01:34,586][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:01:34,609][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:01:34,632][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:01:34,633][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:01:34,633][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:01:35,453][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:01:35,907][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:01:36,412][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:01:36,907][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:01:37,407][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:01:37,907][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:01:38,407][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:01:38,906][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:01:39,406][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:01:39,901][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:01:40,395][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:01:40,889][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:01:41,384][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:01:41,878][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:01:42,372][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:01:42,867][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:01:43,361][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:01:43,859][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:01:44,357][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:01:44,859][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:01:45,364][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:01:45,861][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:01:46,358][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:01:46,854][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:01:47,353][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:01:47,853][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:01:48,354][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:01:48,851][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:01:49,348][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:01:49,845][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:01:50,344][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:01:50,840][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:01:51,335][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:01:51,832][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:01:52,329][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:01:52,826][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:01:53,323][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:01:53,820][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:01:54,317][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:01:54,813][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:01:55,311][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:01:55,807][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:01:56,305][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:01:56,800][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:01:57,295][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:01:57,791][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:01:58,286][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:01:58,783][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:01:59,278][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:01:59,775][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:02:00,272][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:02:00,769][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:02:01,266][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:02:01,763][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:02:02,261][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:02:02,759][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:02:03,256][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:02:03,751][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:02:04,247][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:02:04,742][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:02:05,238][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:02:05,733][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:02:06,228][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:02:06,723][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:02:07,219][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9853 tokens. +[2026-03-26 07:02:08,044][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.99%, Current % of VRAM taken: 60.42%, Block Peak % of device VRAM: 62.03%, ΔTime: 00:00:32 +[2026-03-26 07:02:08,807][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:02:08,809][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:02:08,811][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:02:09,519][__main__][INFO] - Iteration 587 took 50s (29.04% Gen, 69.55% Train). Generation: 14s, Training: 35s. Estimated remaining time: 32h 58m 49s. Estimated total time: 42h 0m 10s. Time estimates for 10 more iterations: 8m 24s, 100 more iterations: 1h 24m 0s, 500 more iterations: 7h 0m 1s. +[2026-03-26 07:02:09,521][__main__][INFO] - Starting iteration 587. +[2026-03-26 07:02:09,934][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 58 and human policies 1. +[2026-03-26 07:02:09,935][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:02:25,408][__main__][INFO] - Number of regex retries in iteration 587: 0 +[2026-03-26 07:02:25,408][__main__][INFO] - agents played in iteration 587 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:02:26,201][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:02:26,226][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:02:26,251][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:02:26,274][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:02:26,275][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:02:26,276][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:02:27,097][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:02:27,553][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:02:28,058][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:02:28,562][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:02:29,062][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:02:29,564][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:02:30,068][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:02:30,567][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:02:31,066][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:02:31,564][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:02:32,062][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:02:32,561][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:02:33,065][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:02:33,564][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:02:34,063][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:02:34,565][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:02:35,063][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:02:35,566][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:02:36,066][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:02:36,566][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:02:37,065][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:02:37,572][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:02:38,072][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:02:38,572][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:02:39,070][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:02:39,568][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:02:40,064][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:02:40,559][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:02:41,055][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:02:41,552][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:02:42,047][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:02:42,541][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:02:43,035][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:02:43,532][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:02:44,029][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:02:44,526][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:02:45,023][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:02:45,519][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:02:46,016][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:02:46,513][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:02:47,011][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:02:47,512][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:02:48,009][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:02:48,507][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:02:49,003][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:02:49,497][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:02:49,995][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:02:50,493][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:02:50,990][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:02:51,484][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:02:51,979][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:02:52,470][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:02:52,963][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:02:53,457][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:02:53,950][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:02:54,444][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:02:54,937][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:02:55,432][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:02:55,924][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:02:56,418][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:02:56,909][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:02:57,402][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:02:57,895][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:02:58,391][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:02:58,892][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9868 tokens. +[2026-03-26 07:02:59,729][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.17%, Current % of VRAM taken: 60.60%, Block Peak % of device VRAM: 62.25%, ΔTime: 00:00:32 +[2026-03-26 07:03:00,488][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:03:00,490][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:03:00,492][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:03:01,241][__main__][INFO] - Iteration 588 took 51s (30.16% Gen, 68.38% Train). Generation: 15s, Training: 35s. Estimated remaining time: 33h 43m 8s. Estimated total time: 42h 45m 20s. Time estimates for 10 more iterations: 8m 33s, 100 more iterations: 1h 25m 30s, 500 more iterations: 7h 7m 33s. +[2026-03-26 07:03:01,243][__main__][INFO] - Starting iteration 588. +[2026-03-26 07:03:01,645][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 58 and human policies 1. +[2026-03-26 07:03:01,646][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:03:17,948][__main__][INFO] - Number of regex retries in iteration 588: 0 +[2026-03-26 07:03:17,949][__main__][INFO] - agents played in iteration 588 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:03:18,731][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:03:18,754][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:03:18,778][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:03:18,803][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:03:18,803][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:03:18,804][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:03:19,727][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:03:20,219][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:03:20,749][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:03:21,248][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:03:21,745][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:03:22,243][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:03:22,739][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:03:23,238][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:03:23,747][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:03:24,260][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:03:24,762][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:03:25,299][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:03:25,798][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:03:26,296][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:03:26,794][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:03:27,294][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:03:27,793][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:03:28,291][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:03:28,792][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:03:29,296][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:03:29,796][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:03:30,297][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:03:30,799][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:03:31,296][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:03:31,794][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:03:32,296][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:03:32,797][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:03:33,298][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:03:33,796][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:03:34,294][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:03:34,814][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:03:35,309][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:03:35,807][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:03:36,304][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:03:36,799][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:03:37,295][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:03:37,793][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:03:38,290][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:03:38,787][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:03:39,292][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:03:39,789][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:03:40,288][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:03:40,787][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:03:41,290][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:03:41,788][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:03:42,287][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:03:42,786][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:03:43,288][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:03:43,787][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:03:44,282][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:03:44,778][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:03:45,271][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:03:45,767][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:03:46,261][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:03:46,758][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:03:47,252][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:03:47,752][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:03:48,246][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:03:48,741][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:03:49,241][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:03:49,735][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:03:50,231][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:03:50,729][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:03:51,226][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:03:51,722][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9860 tokens. +[2026-03-26 07:03:52,539][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 07:03:53,375][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:03:53,377][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:03:53,379][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:03:54,091][__main__][INFO] - Iteration 589 took 52s (31.09% Gen, 67.55% Train). Generation: 16s, Training: 35s. Estimated remaining time: 34h 39m 13s. Estimated total time: 43h 42m 19s. Time estimates for 10 more iterations: 8m 44s, 100 more iterations: 1h 27m 24s, 500 more iterations: 7h 17m 3s. +[2026-03-26 07:03:54,094][__main__][INFO] - Starting iteration 589. +[2026-03-26 07:03:54,494][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 58 and human policies 1. +[2026-03-26 07:03:54,495][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:04:09,602][__main__][INFO] - Number of regex retries in iteration 589: 0 +[2026-03-26 07:04:09,603][__main__][INFO] - agents played in iteration 589 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:04:10,384][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:04:10,408][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:04:10,431][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:04:10,455][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:04:10,456][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:04:10,456][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:04:11,288][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:04:11,740][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:04:12,239][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:04:12,739][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:04:13,237][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:04:13,735][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:04:14,234][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:04:14,733][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:04:15,232][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:04:15,726][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:04:16,228][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:04:16,725][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:04:17,222][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:04:17,720][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:04:18,218][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:04:18,715][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:04:19,213][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:04:19,716][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:04:20,217][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:04:20,711][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:04:21,209][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:04:21,710][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:04:22,212][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:04:22,715][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:04:23,211][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:04:23,706][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:04:24,203][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:04:24,700][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:04:25,195][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:04:25,691][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:04:26,194][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:04:26,691][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:04:27,187][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:04:27,686][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:04:28,184][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:04:28,682][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:04:29,180][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:04:29,678][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:04:30,175][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:04:30,674][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:04:31,172][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:04:31,669][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:04:32,167][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:04:32,664][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:04:33,163][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:04:33,660][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:04:34,157][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:04:34,654][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:04:35,152][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:04:35,648][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:04:36,145][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:04:36,641][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:04:37,137][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:04:37,631][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:04:38,125][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:04:38,621][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:04:39,118][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:04:39,615][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:04:40,112][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:04:40,609][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:04:41,109][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:04:41,606][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:04:42,101][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:04:42,599][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:04:43,096][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9893 tokens. +[2026-03-26 07:04:43,935][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 61.97%, ΔTime: 00:00:32 +[2026-03-26 07:04:44,692][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:04:44,694][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:04:44,696][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:04:45,447][__main__][INFO] - Iteration 590 took 50s (29.65% Gen, 68.87% Train). Generation: 15s, Training: 35s. Estimated remaining time: 33h 23m 43s. Estimated total time: 42h 27m 40s. Time estimates for 10 more iterations: 8m 29s, 100 more iterations: 1h 24m 55s, 500 more iterations: 7h 4m 36s. +[2026-03-26 07:04:45,450][__main__][INFO] - Starting iteration 590. +[2026-03-26 07:04:45,853][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 58 and human policies 1. +[2026-03-26 07:04:45,853][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:04:52,390][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 07:05:00,775][__main__][INFO] - Number of regex retries in iteration 590: 1 +[2026-03-26 07:05:00,776][__main__][INFO] - agents played in iteration 590 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:05:01,547][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:05:01,571][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:05:01,594][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:05:01,617][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:05:01,618][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:05:01,618][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:05:02,444][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:05:02,901][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:05:03,410][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:05:03,911][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:05:04,415][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:05:04,914][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:05:05,413][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:05:05,914][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:05:06,414][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:05:06,914][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:05:07,414][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:05:07,914][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:05:08,414][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:05:08,914][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:05:09,414][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:05:09,914][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:05:10,413][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:05:10,915][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:05:11,416][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:05:11,916][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:05:12,432][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:05:12,933][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:05:13,434][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:05:13,934][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:05:14,427][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:05:14,922][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:05:15,417][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:05:15,913][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:05:16,413][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:05:16,908][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:05:17,403][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:05:17,897][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:05:18,394][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:05:18,892][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:05:19,388][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:05:19,887][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:05:20,387][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:05:20,885][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:05:21,383][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:05:21,879][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:05:22,378][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:05:22,872][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:05:23,366][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:05:23,859][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:05:24,355][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:05:24,849][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:05:25,347][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:05:25,844][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:05:26,338][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:05:26,836][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:05:27,334][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:05:27,832][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:05:28,331][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:05:28,831][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:05:29,331][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:05:29,833][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:05:30,331][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:05:30,826][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:05:31,320][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:05:31,813][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:05:32,307][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:05:32,803][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:05:33,294][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:05:33,787][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:05:34,283][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9880 tokens. +[2026-03-26 07:05:35,112][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.06%, ΔTime: 00:00:32 +[2026-03-26 07:05:35,887][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:05:35,889][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:05:35,891][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:05:37,253][__main__][INFO] - Iteration 591 took 51s (29.03% Gen, 68.32% Train). Generation: 14s, Training: 35s. Estimated remaining time: 33h 45m 13s. Estimated total time: 42h 50m 1s. Time estimates for 10 more iterations: 8m 34s, 100 more iterations: 1h 25m 40s, 500 more iterations: 7h 8m 20s. +[2026-03-26 07:05:37,255][__main__][INFO] - Starting iteration 591. +[2026-03-26 07:05:37,655][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 59 and human policies 1. +[2026-03-26 07:05:37,656][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:05:50,577][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 07:05:52,969][__main__][INFO] - Number of regex retries in iteration 591: 1 +[2026-03-26 07:05:52,970][__main__][INFO] - agents played in iteration 591 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:05:53,730][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:05:53,754][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:05:53,777][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:05:53,800][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:05:53,801][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:05:53,802][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:05:54,601][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:05:55,058][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:05:55,561][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:05:56,061][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:05:56,560][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:05:57,059][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:05:57,558][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:05:58,058][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:05:58,557][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:05:59,056][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:05:59,555][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:06:00,054][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:06:00,553][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:06:01,052][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:06:01,552][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:06:02,051][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:06:02,551][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:06:03,049][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:06:03,548][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:06:04,048][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:06:04,547][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:06:05,047][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:06:05,544][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:06:06,045][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:06:06,546][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:06:07,044][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:06:07,540][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:06:08,038][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:06:08,535][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:06:09,028][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:06:09,526][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:06:10,022][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:06:10,520][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:06:11,016][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:06:11,512][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:06:12,008][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:06:12,505][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:06:13,000][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:06:13,496][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:06:13,993][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:06:14,488][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:06:14,985][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:06:15,478][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:06:15,971][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:06:16,464][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:06:16,966][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:06:17,462][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:06:17,953][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:06:18,447][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:06:18,943][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:06:19,443][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:06:19,939][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:06:20,438][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:06:20,936][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:06:21,434][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:06:21,931][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:06:22,431][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:06:22,927][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:06:23,422][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:06:23,918][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:06:24,413][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:06:24,910][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:06:25,407][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:06:25,904][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:06:26,399][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9869 tokens. +[2026-03-26 07:06:27,222][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 62.05%, ΔTime: 00:00:32 +[2026-03-26 07:06:27,986][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:06:27,989][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:06:27,990][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:06:28,736][__main__][INFO] - Iteration 592 took 51s (29.98% Gen, 68.56% Train). Generation: 15s, Training: 35s. Estimated remaining time: 33h 28m 24s. Estimated total time: 42h 34m 4s. Time estimates for 10 more iterations: 8m 30s, 100 more iterations: 1h 25m 8s, 500 more iterations: 7h 5m 40s. +[2026-03-26 07:06:28,738][__main__][INFO] - Starting iteration 592. +[2026-03-26 07:06:29,136][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 59 and human policies 1. +[2026-03-26 07:06:29,137][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:06:44,793][__main__][INFO] - Number of regex retries in iteration 592: 0 +[2026-03-26 07:06:44,794][__main__][INFO] - agents played in iteration 592 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:06:45,574][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:06:45,598][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:06:45,621][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:06:45,645][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:06:45,645][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:06:45,646][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:06:46,452][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:06:46,909][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:06:47,415][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:06:47,917][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:06:48,420][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:06:48,921][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:06:49,423][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:06:49,926][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:06:50,428][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:06:50,925][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:06:51,422][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:06:51,919][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:06:52,416][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:06:52,914][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:06:53,412][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:06:53,910][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:06:54,408][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:06:54,908][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:06:55,405][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:06:55,907][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:06:56,405][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:06:56,906][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:06:57,407][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:06:57,906][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:06:58,404][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:06:58,905][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:06:59,406][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:06:59,908][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:07:00,407][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:07:00,906][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:07:01,404][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:07:01,900][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:07:02,394][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:07:02,894][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:07:03,390][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:07:03,890][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:07:04,389][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:07:04,889][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:07:05,389][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:07:05,889][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:07:06,390][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:07:06,884][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:07:07,377][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:07:07,870][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:07:08,367][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:07:08,859][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:07:09,354][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:07:09,853][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:07:10,350][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:07:10,845][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:07:11,339][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:07:11,836][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:07:12,334][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:07:12,832][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:07:13,328][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:07:13,826][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:07:14,323][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:07:14,820][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:07:15,316][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:07:15,811][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:07:16,307][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:07:16,804][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:07:17,300][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:07:17,798][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:07:18,297][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9938 tokens. +[2026-03-26 07:07:19,119][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.43%, ΔTime: 00:00:32 +[2026-03-26 07:07:19,887][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:07:19,890][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:07:19,891][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:07:20,604][__main__][INFO] - Iteration 593 took 51s (30.42% Gen, 68.19% Train). Generation: 15s, Training: 35s. Estimated remaining time: 33h 46m 52s. Estimated total time: 42h 53m 24s. Time estimates for 10 more iterations: 8m 34s, 100 more iterations: 1h 25m 46s, 500 more iterations: 7h 8m 54s. +[2026-03-26 07:07:20,606][__main__][INFO] - Starting iteration 593. +[2026-03-26 07:07:21,007][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 59 and human policies 1. +[2026-03-26 07:07:21,008][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:07:35,786][__main__][INFO] - Number of regex retries in iteration 593: 0 +[2026-03-26 07:07:35,787][__main__][INFO] - agents played in iteration 593 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:07:36,559][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:07:36,583][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:07:36,606][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:07:36,630][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:07:36,630][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:07:36,631][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:07:37,456][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:07:37,910][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:07:38,418][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:07:38,919][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:07:39,420][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:07:39,919][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:07:40,420][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:07:40,920][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:07:41,419][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:07:41,923][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:07:42,423][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:07:42,924][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:07:43,424][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:07:43,923][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:07:44,424][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:07:44,926][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:07:45,427][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:07:45,928][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:07:46,427][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:07:46,929][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:07:47,431][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:07:47,956][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:07:48,455][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:07:48,955][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:07:49,454][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:07:49,956][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:07:50,458][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:07:50,958][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:07:51,458][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:07:51,957][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:07:52,454][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:07:52,952][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:07:53,453][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:07:53,948][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:07:54,446][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:07:54,941][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:07:55,435][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:07:55,930][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:07:56,424][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:07:56,919][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:07:57,419][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:07:57,912][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:07:58,408][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:07:58,898][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:07:59,397][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:07:59,891][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:08:00,384][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:08:00,879][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:08:01,374][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:08:01,872][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:08:02,367][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:08:02,869][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:08:03,366][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:08:03,863][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:08:04,357][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:08:04,853][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:08:05,351][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:08:05,848][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:08:06,346][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:08:06,843][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:08:07,339][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:08:07,836][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:08:08,333][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:08:08,830][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:08:09,327][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9864 tokens. +[2026-03-26 07:08:10,149][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.04%, ΔTime: 00:00:32 +[2026-03-26 07:08:10,908][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:08:10,910][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:08:10,912][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:08:11,625][__main__][INFO] - Iteration 594 took 50s (29.20% Gen, 69.39% Train). Generation: 14s, Training: 35s. Estimated remaining time: 33h 3m 31s. Estimated total time: 42h 10m 54s. Time estimates for 10 more iterations: 8m 26s, 100 more iterations: 1h 24m 21s, 500 more iterations: 7h 1m 49s. +[2026-03-26 07:08:11,627][__main__][INFO] - Starting iteration 594. +[2026-03-26 07:08:12,027][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 59 and human policies 1. +[2026-03-26 07:08:12,027][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:08:27,155][__main__][INFO] - Number of regex retries in iteration 594: 0 +[2026-03-26 07:08:27,156][__main__][INFO] - agents played in iteration 594 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:08:27,938][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:08:27,962][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:08:27,986][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:08:28,009][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:08:28,010][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:08:28,011][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:08:28,830][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:08:29,290][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:08:29,796][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:08:30,297][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:08:30,797][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:08:31,300][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:08:31,803][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:08:32,306][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:08:32,808][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:08:33,308][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:08:33,802][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:08:34,298][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:08:34,796][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:08:35,291][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:08:35,790][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:08:36,287][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:08:36,781][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:08:37,281][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:08:37,779][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:08:38,276][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:08:38,777][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:08:39,279][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:08:39,779][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:08:40,280][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:08:40,777][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:08:41,279][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:08:41,777][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:08:42,273][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:08:42,768][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:08:43,270][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:08:43,764][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:08:44,258][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:08:44,755][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:08:45,252][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:08:45,749][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:08:46,248][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:08:46,747][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:08:47,246][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:08:47,745][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:08:48,243][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:08:48,737][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:08:49,233][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:08:49,730][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:08:50,228][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:08:50,726][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:08:51,226][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:08:51,723][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:08:52,221][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:08:52,719][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:08:53,216][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:08:53,712][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:08:54,209][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:08:54,707][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:08:55,207][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:08:55,703][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:08:56,200][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:08:56,697][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:08:57,194][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:08:57,694][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:08:58,191][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:08:58,688][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:08:59,186][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:08:59,682][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:09:00,178][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:09:00,674][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9853 tokens. +[2026-03-26 07:09:01,506][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 62.00%, ΔTime: 00:00:32 +[2026-03-26 07:09:02,272][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:09:02,275][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:09:02,277][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:09:02,997][__main__][INFO] - Iteration 595 took 50s (29.68% Gen, 68.90% Train). Generation: 15s, Training: 35s. Estimated remaining time: 33h 20m 17s. Estimated total time: 42h 28m 31s. Time estimates for 10 more iterations: 8m 29s, 100 more iterations: 1h 24m 57s, 500 more iterations: 7h 4m 45s. +[2026-03-26 07:09:02,999][__main__][INFO] - Starting iteration 595. +[2026-03-26 07:09:03,398][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 59 and human policies 1. +[2026-03-26 07:09:03,399][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:09:18,545][__main__][INFO] - Number of regex retries in iteration 595: 0 +[2026-03-26 07:09:18,545][__main__][INFO] - agents played in iteration 595 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:09:19,329][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:09:19,353][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:09:19,376][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:09:19,399][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:09:19,400][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:09:19,401][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:09:20,221][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:09:20,672][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:09:21,171][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:09:21,666][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:09:22,166][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:09:22,663][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:09:23,159][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:09:23,660][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:09:24,155][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:09:24,650][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:09:25,152][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:09:25,654][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:09:26,154][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:09:26,652][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:09:27,150][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:09:27,648][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:09:28,149][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:09:28,644][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:09:29,140][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:09:29,637][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:09:30,135][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:09:30,632][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:09:31,128][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:09:31,628][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:09:32,125][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:09:32,621][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:09:33,120][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:09:33,618][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:09:34,117][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:09:34,621][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:09:35,121][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:09:35,623][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:09:36,121][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:09:36,621][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:09:37,119][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:09:37,619][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:09:38,116][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:09:38,616][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:09:39,113][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:09:39,611][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:09:40,133][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:09:40,631][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:09:41,128][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:09:41,625][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:09:42,120][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:09:42,616][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:09:43,112][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:09:43,609][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:09:44,105][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:09:44,600][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:09:45,097][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:09:45,594][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:09:46,091][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:09:46,589][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:09:47,086][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:09:47,584][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:09:48,080][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:09:48,574][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:09:49,070][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:09:49,567][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:09:50,065][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:09:50,562][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:09:51,058][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:09:51,551][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:09:52,048][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9857 tokens. +[2026-03-26 07:09:52,868][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.06%, ΔTime: 00:00:32 +[2026-03-26 07:09:53,660][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:09:53,663][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:09:53,665][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:09:54,406][__main__][INFO] - Iteration 596 took 51s (29.69% Gen, 68.85% Train). Generation: 15s, Training: 35s. Estimated remaining time: 33h 21m 18s. Estimated total time: 42h 30m 24s. Time estimates for 10 more iterations: 8m 30s, 100 more iterations: 1h 25m 0s, 500 more iterations: 7h 5m 4s. +[2026-03-26 07:09:54,408][__main__][INFO] - Starting iteration 596. +[2026-03-26 07:09:54,811][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 59 and human policies 1. +[2026-03-26 07:09:54,812][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:10:09,556][__main__][INFO] - Number of regex retries in iteration 596: 0 +[2026-03-26 07:10:09,557][__main__][INFO] - agents played in iteration 596 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:10:10,331][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:10:10,354][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:10:10,377][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:10:10,401][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:10:10,401][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:10:10,402][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:10:11,214][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:10:11,669][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:10:12,177][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:10:12,682][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:10:13,188][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:10:13,693][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:10:14,197][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:10:14,698][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:10:15,203][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:10:15,703][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:10:16,209][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:10:16,708][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:10:17,206][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:10:17,705][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:10:18,205][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:10:18,705][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:10:19,206][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:10:19,704][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:10:20,203][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:10:20,701][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:10:21,199][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:10:21,696][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:10:22,193][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:10:22,692][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:10:23,190][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:10:23,689][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:10:24,188][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:10:24,690][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:10:25,188][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:10:25,686][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:10:26,187][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:10:26,687][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:10:27,186][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:10:27,686][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:10:28,189][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:10:28,688][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:10:29,186][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:10:29,681][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:10:30,177][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:10:30,673][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:10:31,169][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:10:31,661][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:10:32,157][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:10:32,649][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:10:33,141][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:10:33,637][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:10:34,129][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:10:34,628][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:10:35,130][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:10:35,627][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:10:36,125][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:10:36,623][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:10:37,120][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:10:37,618][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:10:38,115][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:10:38,614][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:10:39,113][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:10:39,610][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:10:40,108][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:10:40,606][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:10:41,102][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:10:41,598][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:10:42,095][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:10:42,592][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:10:43,089][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9954 tokens. +[2026-03-26 07:10:43,919][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.17%, ΔTime: 00:00:32 +[2026-03-26 07:10:44,678][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:10:44,680][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:10:44,682][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:10:45,462][__main__][INFO] - Iteration 597 took 50s (29.11% Gen, 69.35% Train). Generation: 14s, Training: 35s. Estimated remaining time: 33h 2m 38s. Estimated total time: 42h 12m 35s. Time estimates for 10 more iterations: 8m 26s, 100 more iterations: 1h 24m 25s, 500 more iterations: 7h 2m 5s. +[2026-03-26 07:10:45,464][__main__][INFO] - Starting iteration 597. +[2026-03-26 07:10:45,873][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 59 and human policies 1. +[2026-03-26 07:10:45,873][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:11:00,060][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 20 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 07:11:00,986][__main__][INFO] - Number of regex retries in iteration 597: 1 +[2026-03-26 07:11:00,986][__main__][INFO] - agents played in iteration 597 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:11:01,789][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:11:01,812][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:11:01,835][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:11:01,859][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:11:01,859][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:11:01,860][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:11:02,683][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:11:03,135][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:11:03,640][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:11:04,138][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:11:04,639][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:11:05,137][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:11:05,639][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:11:06,138][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:11:06,640][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:11:07,137][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:11:07,635][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:11:08,129][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:11:08,626][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:11:09,124][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:11:09,621][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:11:10,119][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:11:10,616][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:11:11,115][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:11:11,614][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:11:12,113][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:11:12,615][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:11:13,114][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:11:13,612][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:11:14,111][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:11:14,610][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:11:15,104][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:11:15,598][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:11:16,096][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:11:16,595][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:11:17,097][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:11:17,596][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:11:18,095][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:11:18,596][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:11:19,093][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:11:19,589][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:11:20,085][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:11:20,580][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:11:21,075][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:11:21,570][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:11:22,065][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:11:22,558][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:11:23,055][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:11:23,553][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:11:24,052][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:11:24,548][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:11:25,045][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:11:25,543][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:11:26,039][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:11:26,537][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:11:27,034][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:11:27,531][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:11:28,028][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:11:28,525][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:11:29,021][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:11:29,545][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:11:30,046][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:11:30,543][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:11:31,040][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:11:31,537][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:11:32,034][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:11:32,531][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:11:33,027][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:11:33,524][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:11:34,020][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:11:34,516][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9857 tokens. +[2026-03-26 07:11:35,341][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.00%, ΔTime: 00:00:32 +[2026-03-26 07:11:36,015][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:11:36,018][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:11:36,020][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:11:36,740][__main__][INFO] - Iteration 598 took 50s (29.71% Gen, 68.87% Train). Generation: 15s, Training: 35s. Estimated remaining time: 33h 12m 35s. Estimated total time: 42h 23m 23s. Time estimates for 10 more iterations: 8m 28s, 100 more iterations: 1h 24m 46s, 500 more iterations: 7h 3m 53s. +[2026-03-26 07:11:36,742][__main__][INFO] - Starting iteration 598. +[2026-03-26 07:11:37,140][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 59 and human policies 1. +[2026-03-26 07:11:37,140][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:12:09,609][__main__][INFO] - Number of regex retries in iteration 598: 0 +[2026-03-26 07:12:09,610][__main__][INFO] - agents played in iteration 598 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:12:10,370][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:12:10,393][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:12:10,417][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:12:10,440][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:12:10,441][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:12:10,441][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:12:11,234][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:12:11,686][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:12:12,190][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:12:12,686][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:12:13,185][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:12:13,684][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:12:14,181][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:12:14,680][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:12:15,178][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:12:15,676][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:12:16,174][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:12:16,667][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:12:17,165][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:12:17,660][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:12:18,155][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:12:18,650][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:12:19,153][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:12:19,650][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:12:20,149][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:12:20,651][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:12:21,147][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:12:21,645][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:12:22,144][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:12:22,641][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:12:23,136][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:12:23,632][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:12:24,127][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:12:24,622][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:12:25,117][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:12:25,613][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:12:26,109][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:12:26,605][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:12:27,100][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:12:27,594][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:12:28,088][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:12:28,582][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:12:29,077][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:12:29,574][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:12:30,076][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:12:30,571][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:12:31,064][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:12:31,561][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:12:32,057][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:12:32,555][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:12:33,053][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:12:33,548][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:12:34,046][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:12:34,543][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:12:35,039][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:12:35,536][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:12:36,031][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:12:36,531][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:12:37,030][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:12:37,529][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:12:38,027][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:12:38,525][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:12:39,022][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:12:39,519][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:12:40,016][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:12:40,513][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:12:41,011][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:12:41,508][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:12:42,006][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:12:42,503][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:12:43,000][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9830 tokens. +[2026-03-26 07:12:43,816][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.05%, ΔTime: 00:00:32 +[2026-03-26 07:12:44,555][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:12:44,558][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:12:44,559][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:12:45,297][__main__][INFO] - Iteration 599 took 1m 8s (47.64% Gen, 51.28% Train). Generation: 32s, Training: 34s. Estimated remaining time: 47h 35m 56s. Estimated total time: 56h 47m 53s. Time estimates for 10 more iterations: 11m 21s, 100 more iterations: 1h 53m 35s, 500 more iterations: 9h 27m 58s. +[2026-03-26 07:12:45,299][__main__][INFO] - Starting iteration 599. +[2026-03-26 07:12:45,723][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 59 and human policies 1. +[2026-03-26 07:12:45,723][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:12:51,395][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 07:13:00,064][__main__][INFO] - Number of regex retries in iteration 599: 1 +[2026-03-26 07:13:00,064][__main__][INFO] - agents played in iteration 599 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:13:00,830][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:13:00,854][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:13:00,877][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:13:00,900][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:13:00,900][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:13:00,901][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:13:01,695][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:13:02,147][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:13:02,652][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:13:03,152][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:13:03,657][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:13:04,157][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:13:04,658][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:13:05,160][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:13:05,661][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:13:06,163][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:13:06,668][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:13:07,169][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:13:07,671][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:13:08,173][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:13:08,670][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:13:09,166][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:13:09,663][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:13:10,161][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:13:10,659][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:13:11,156][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:13:11,653][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:13:12,151][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:13:12,649][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:13:13,148][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:13:13,648][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:13:14,149][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:13:14,649][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:13:15,149][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:13:15,649][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:13:16,149][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:13:16,650][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:13:17,147][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:13:17,648][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:13:18,150][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:13:18,647][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:13:19,145][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:13:19,646][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:13:20,146][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:13:20,648][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:13:21,147][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:13:21,644][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:13:22,139][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:13:22,634][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:13:23,130][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:13:23,625][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:13:24,124][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:13:24,619][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:13:25,115][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:13:25,611][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:13:26,110][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:13:26,608][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:13:27,107][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:13:27,606][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:13:28,104][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:13:28,601][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:13:29,098][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:13:29,596][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:13:30,117][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:13:30,617][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:13:31,114][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:13:31,612][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:13:32,110][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:13:32,609][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:13:33,110][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:13:33,608][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9979 tokens. +[2026-03-26 07:13:34,430][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.12%, ΔTime: 00:00:32 +[2026-03-26 07:13:35,289][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:13:35,291][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:13:35,293][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:13:35,947][__main__][INFO] - Iteration 600 took 50s (28.55% Gen, 70.14% Train). Generation: 14s, Training: 35s. Estimated remaining time: 32h 38m 29s. Estimated total time: 41h 51m 16s. Time estimates for 10 more iterations: 8m 22s, 100 more iterations: 1h 23m 42s, 500 more iterations: 6h 58m 32s. +[2026-03-26 07:13:35,949][__main__][INFO] - Starting iteration 600. +[2026-03-26 07:13:36,371][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 59 and human policies 1. +[2026-03-26 07:13:36,372][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:13:57,689][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 07:14:08,702][__main__][INFO] - Number of regex retries in iteration 600: 1 +[2026-03-26 07:14:08,703][__main__][INFO] - agents played in iteration 600 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:14:09,470][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:14:09,493][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:14:09,516][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:14:09,539][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:14:09,540][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:14:09,540][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:14:10,342][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:14:10,799][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:14:11,306][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:14:11,808][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:14:12,308][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:14:12,808][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:14:13,309][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:14:13,807][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:14:14,304][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:14:14,800][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:14:15,297][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:14:15,794][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:14:16,289][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:14:16,789][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:14:17,286][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:14:17,784][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:14:18,288][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:14:18,788][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:14:19,284][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:14:19,780][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:14:20,287][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:14:20,787][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:14:21,288][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:14:21,784][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:14:22,285][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:14:22,778][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:14:23,270][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:14:23,765][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:14:24,257][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:14:24,754][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:14:25,248][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:14:25,746][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:14:26,246][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:14:26,746][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:14:27,243][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:14:27,739][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:14:28,235][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:14:28,732][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:14:29,226][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:14:29,721][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:14:30,215][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:14:30,707][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:14:31,206][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:14:31,706][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:14:32,204][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:14:32,703][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:14:33,200][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:14:33,697][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:14:34,194][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:14:34,691][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:14:35,189][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:14:35,687][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:14:36,186][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:14:36,684][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:14:37,182][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:14:37,679][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:14:38,174][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:14:38,673][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:14:39,171][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:14:39,672][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:14:40,168][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:14:40,666][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:14:41,164][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:14:41,662][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:14:42,159][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9864 tokens. +[2026-03-26 07:14:43,004][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 61.93%, ΔTime: 00:00:32 +[2026-03-26 07:14:43,802][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:14:43,804][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:14:43,806][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:14:45,265][__main__][INFO] - Iteration 601 took 1m 8s (46.93% Gen, 50.95% Train). Generation: 32s, Training: 35s. Estimated remaining time: 48h 10m 48s. Estimated total time: 57h 24m 45s. Time estimates for 10 more iterations: 11m 28s, 100 more iterations: 1h 54m 49s, 500 more iterations: 9h 34m 7s. +[2026-03-26 07:14:45,269][__main__][INFO] - Starting iteration 601. +[2026-03-26 07:14:45,685][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 60 and human policies 1. +[2026-03-26 07:14:45,686][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:15:09,682][__main__][INFO] - Number of regex retries in iteration 601: 0 +[2026-03-26 07:15:09,682][__main__][INFO] - agents played in iteration 601 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:15:10,446][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:15:10,469][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:15:10,492][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:15:10,516][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:15:10,516][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:15:10,517][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:15:11,301][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:15:11,752][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:15:12,256][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:15:12,757][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:15:13,248][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:15:13,746][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:15:14,241][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:15:14,739][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:15:15,236][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:15:15,734][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:15:16,226][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:15:16,723][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:15:17,220][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:15:17,722][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:15:18,219][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:15:18,717][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:15:19,215][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:15:19,712][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:15:20,212][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:15:20,709][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:15:21,207][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:15:21,705][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:15:22,201][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:15:22,697][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:15:23,193][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:15:23,691][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:15:24,190][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:15:24,687][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:15:25,189][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:15:25,687][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:15:26,186][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:15:26,683][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:15:27,180][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:15:27,675][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:15:28,169][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:15:28,665][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:15:29,161][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:15:29,657][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:15:30,152][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:15:30,646][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:15:31,143][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:15:31,641][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:15:32,138][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:15:32,636][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:15:33,134][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:15:33,632][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:15:34,130][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:15:34,627][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:15:35,126][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:15:35,629][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:15:36,131][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:15:36,635][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:15:37,135][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:15:37,646][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:15:38,148][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:15:38,650][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:15:39,152][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:15:39,654][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:15:40,157][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:15:40,661][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:15:41,165][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:15:41,660][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:15:42,161][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:15:42,657][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:15:43,153][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9878 tokens. +[2026-03-26 07:15:44,007][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.05%, Current % of VRAM taken: 60.48%, Block Peak % of device VRAM: 62.10%, ΔTime: 00:00:32 +[2026-03-26 07:15:44,860][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:15:44,862][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:15:44,864][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:15:45,653][__main__][INFO] - Iteration 602 took 59s (40.02% Gen, 58.67% Train). Generation: 23s, Training: 35s. Estimated remaining time: 40h 43m 28s. Estimated total time: 49h 58m 25s. Time estimates for 10 more iterations: 9m 59s, 100 more iterations: 1h 39m 56s, 500 more iterations: 8h 19m 44s. +[2026-03-26 07:15:45,655][__main__][INFO] - Starting iteration 602. +[2026-03-26 07:15:46,056][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 60 and human policies 1. +[2026-03-26 07:15:46,056][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:16:01,966][__main__][INFO] - Number of regex retries in iteration 602: 0 +[2026-03-26 07:16:01,967][__main__][INFO] - agents played in iteration 602 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:16:02,731][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:16:02,754][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:16:02,777][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:16:02,800][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:16:02,801][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:16:02,802][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:16:03,605][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:16:04,056][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:16:04,557][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:16:05,053][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:16:05,552][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:16:06,050][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:16:06,554][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:16:07,054][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:16:07,550][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:16:08,047][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:16:08,546][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:16:09,039][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:16:09,534][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:16:10,030][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:16:10,525][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:16:11,020][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:16:11,517][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:16:12,016][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:16:12,516][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:16:13,013][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:16:13,508][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:16:14,005][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:16:14,511][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:16:15,013][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:16:15,513][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:16:16,016][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:16:16,520][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:16:17,027][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:16:17,531][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:16:18,035][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:16:18,535][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:16:19,034][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:16:19,533][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:16:20,036][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:16:20,538][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:16:21,040][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:16:21,543][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:16:22,046][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:16:22,552][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:16:23,054][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:16:23,558][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:16:24,057][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:16:24,562][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:16:25,063][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:16:25,564][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:16:26,064][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:16:26,565][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:16:27,069][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:16:27,570][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:16:28,074][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:16:28,573][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:16:29,070][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:16:29,568][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:16:30,065][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:16:30,565][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:16:31,066][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:16:31,565][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:16:32,061][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:16:32,557][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:16:33,053][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:16:33,549][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:16:34,047][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:16:34,547][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:16:35,044][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:16:35,543][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9904 tokens. +[2026-03-26 07:16:36,393][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.29%, ΔTime: 00:00:32 +[2026-03-26 07:16:37,260][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:16:37,262][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:16:37,264][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:16:38,280][__main__][INFO] - Iteration 603 took 52s (30.46% Gen, 67.59% Train). Generation: 15s, Training: 35s. Estimated remaining time: 34h 15m 25s. Estimated total time: 43h 31m 15s. Time estimates for 10 more iterations: 8m 42s, 100 more iterations: 1h 27m 2s, 500 more iterations: 7h 15m 12s. +[2026-03-26 07:16:38,282][__main__][INFO] - Starting iteration 603. +[2026-03-26 07:16:38,697][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 60 and human policies 1. +[2026-03-26 07:16:38,698][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:16:55,528][__main__][INFO] - Number of regex retries in iteration 603: 0 +[2026-03-26 07:16:55,529][__main__][INFO] - agents played in iteration 603 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:16:56,303][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:16:56,327][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:16:56,350][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:16:56,373][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:16:56,374][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:16:56,374][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:16:57,188][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:16:57,641][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:16:58,144][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:16:58,644][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:16:59,138][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:16:59,636][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:17:00,132][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:17:00,632][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:17:01,135][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:17:01,634][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:17:02,138][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:17:02,637][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:17:03,135][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:17:03,632][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:17:04,128][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:17:04,627][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:17:05,121][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:17:05,615][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:17:06,113][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:17:06,610][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:17:07,107][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:17:07,605][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:17:08,102][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:17:08,597][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:17:09,091][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:17:09,591][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:17:10,091][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:17:10,590][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:17:11,089][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:17:11,589][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:17:12,089][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:17:12,584][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:17:13,081][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:17:13,578][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:17:14,076][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:17:14,574][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:17:15,076][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:17:15,574][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:17:16,072][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:17:16,570][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:17:17,069][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:17:17,564][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:17:18,059][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:17:18,557][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:17:19,057][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:17:19,551][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:17:20,048][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:17:20,547][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:17:21,041][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:17:21,538][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:17:22,034][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:17:22,531][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:17:23,025][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:17:23,519][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:17:24,010][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:17:24,503][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:17:24,997][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:17:25,496][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:17:25,994][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:17:26,494][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:17:26,992][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:17:27,491][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:17:27,990][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:17:28,486][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:17:28,984][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9870 tokens. +[2026-03-26 07:17:29,825][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.06%, ΔTime: 00:00:32 +[2026-03-26 07:17:30,585][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:17:30,587][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:17:30,589][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:17:31,455][__main__][INFO] - Iteration 604 took 52s (31.90% Gen, 66.45% Train). Generation: 16s, Training: 35s. Estimated remaining time: 34h 41m 12s. Estimated total time: 43h 57m 55s. Time estimates for 10 more iterations: 8m 47s, 100 more iterations: 1h 27m 55s, 500 more iterations: 7h 19m 39s. +[2026-03-26 07:17:31,457][__main__][INFO] - Starting iteration 604. +[2026-03-26 07:17:31,897][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 60 and human policies 1. +[2026-03-26 07:17:31,897][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:17:47,806][__main__][INFO] - Number of regex retries in iteration 604: 0 +[2026-03-26 07:17:47,807][__main__][INFO] - agents played in iteration 604 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:17:48,576][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:17:48,599][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:17:48,623][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:17:48,646][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:17:48,646][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:17:48,647][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:17:49,483][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:17:49,937][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:17:50,438][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:17:50,935][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:17:51,436][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:17:51,935][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:17:52,432][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:17:52,927][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:17:53,421][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:17:53,916][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:17:54,414][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:17:54,911][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:17:55,407][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:17:55,905][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:17:56,401][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:17:56,897][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:17:57,393][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:17:57,893][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:17:58,389][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:17:58,921][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:17:59,422][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:17:59,921][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:18:00,418][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:18:00,916][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:18:01,413][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:18:01,911][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:18:02,409][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:18:02,907][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:18:03,403][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:18:03,899][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:18:04,396][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:18:04,893][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:18:05,391][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:18:05,889][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:18:06,388][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:18:06,886][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:18:07,382][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:18:07,877][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:18:08,375][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:18:08,872][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:18:09,366][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:18:09,862][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:18:10,358][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:18:10,854][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:18:11,350][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:18:11,848][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:18:12,344][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:18:12,840][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:18:13,337][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:18:13,832][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:18:14,337][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:18:14,831][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:18:15,326][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:18:15,818][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:18:16,313][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:18:16,809][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:18:17,309][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:18:17,805][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:18:18,300][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:18:18,794][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:18:19,290][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:18:19,786][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:18:20,280][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:18:20,773][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:18:21,269][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9787 tokens. +[2026-03-26 07:18:22,122][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.01%, ΔTime: 00:00:32 +[2026-03-26 07:18:22,881][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:18:22,884][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:18:22,885][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:18:23,657][__main__][INFO] - Iteration 605 took 51s (30.74% Gen, 67.77% Train). Generation: 15s, Training: 35s. Estimated remaining time: 33h 50m 29s. Estimated total time: 43h 8m 4s. Time estimates for 10 more iterations: 8m 37s, 100 more iterations: 1h 26m 16s, 500 more iterations: 7h 11m 20s. +[2026-03-26 07:18:23,659][__main__][INFO] - Starting iteration 605. +[2026-03-26 07:18:24,102][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 60 and human policies 1. +[2026-03-26 07:18:24,103][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:18:28,315][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 07:18:38,935][__main__][INFO] - Number of regex retries in iteration 605: 1 +[2026-03-26 07:18:38,936][__main__][INFO] - agents played in iteration 605 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:18:39,701][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:18:39,725][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:18:39,748][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:18:39,771][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:18:39,771][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:18:39,772][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:18:40,595][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:18:41,049][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:18:41,549][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:18:42,047][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:18:42,544][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:18:43,041][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:18:43,540][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:18:44,037][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:18:44,531][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:18:45,028][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:18:45,524][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:18:46,020][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:18:46,517][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:18:47,012][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:18:47,507][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:18:48,002][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:18:48,497][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:18:48,993][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:18:49,485][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:18:49,980][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:18:50,473][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:18:50,962][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:18:51,457][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:18:51,958][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:18:52,454][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:18:52,947][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:18:53,441][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:18:53,934][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:18:54,430][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:18:54,927][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:18:55,420][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:18:55,916][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:18:56,412][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:18:56,911][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:18:57,409][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:18:57,907][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:18:58,406][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:18:58,903][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:18:59,400][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:18:59,898][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:19:00,395][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:19:00,888][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:19:01,379][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:19:01,870][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:19:02,360][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:19:02,852][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:19:03,348][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:19:03,840][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:19:04,331][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:19:04,824][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:19:05,315][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:19:05,812][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:19:06,308][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:19:06,808][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:19:07,303][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:19:07,797][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:19:08,292][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:19:08,787][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:19:09,281][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:19:09,777][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:19:10,272][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:19:10,766][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:19:11,259][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:19:11,753][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:19:12,248][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9727 tokens. +[2026-03-26 07:19:13,089][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 12.96%, Current % of VRAM taken: 59.39%, Block Peak % of device VRAM: 62.01%, ΔTime: 00:00:32 +[2026-03-26 07:19:13,849][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:19:13,850][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:19:13,852][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:19:14,579][__main__][INFO] - Iteration 606 took 50s (29.38% Gen, 69.17% Train). Generation: 14s, Training: 34s. Estimated remaining time: 32h 45m 27s. Estimated total time: 42h 3m 53s. Time estimates for 10 more iterations: 8m 24s, 100 more iterations: 1h 24m 7s, 500 more iterations: 7h 0m 38s. +[2026-03-26 07:19:14,582][__main__][INFO] - Starting iteration 606. +[2026-03-26 07:19:14,985][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 60 and human policies 1. +[2026-03-26 07:19:14,985][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:19:30,447][__main__][INFO] - Number of regex retries in iteration 606: 0 +[2026-03-26 07:19:30,448][__main__][INFO] - agents played in iteration 606 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:19:31,217][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:19:31,242][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:19:31,266][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:19:31,290][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:19:31,291][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:19:31,291][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:19:32,115][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:19:32,568][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:19:33,073][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:19:33,570][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:19:34,070][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:19:34,569][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:19:35,066][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:19:35,563][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:19:36,067][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:19:36,564][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:19:37,060][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:19:37,564][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:19:38,063][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:19:38,561][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:19:39,057][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:19:39,554][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:19:40,053][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:19:40,558][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:19:41,056][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:19:41,560][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:19:42,057][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:19:42,554][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:19:43,056][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:19:43,553][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:19:44,048][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:19:44,545][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:19:45,042][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:19:45,536][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:19:46,034][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:19:46,533][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:19:47,028][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:19:47,530][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:19:48,028][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:19:48,524][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:19:49,018][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:19:49,516][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:19:50,011][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:19:50,510][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:19:51,007][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:19:51,504][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:19:51,999][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:19:52,496][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:19:52,991][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:19:53,493][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:19:53,991][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:19:54,488][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:19:54,987][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:19:55,484][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:19:55,981][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:19:56,475][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:19:56,971][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:19:57,465][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:19:57,962][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:19:58,458][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:19:58,952][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:19:59,448][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:19:59,946][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:20:00,439][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:20:00,929][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:20:01,425][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:20:01,921][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:20:02,413][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:20:02,904][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:20:03,399][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:20:03,898][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9865 tokens. +[2026-03-26 07:20:04,748][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.20%, Current % of VRAM taken: 60.63%, Block Peak % of device VRAM: 62.13%, ΔTime: 00:00:32 +[2026-03-26 07:20:05,493][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:20:05,496][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:20:05,498][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:20:06,279][__main__][INFO] - Iteration 607 took 51s (30.14% Gen, 68.33% Train). Generation: 15s, Training: 35s. Estimated remaining time: 33h 25m 26s. Estimated total time: 42h 44m 44s. Time estimates for 10 more iterations: 8m 32s, 100 more iterations: 1h 25m 29s, 500 more iterations: 7h 7m 27s. +[2026-03-26 07:20:06,281][__main__][INFO] - Starting iteration 607. +[2026-03-26 07:20:06,684][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 60 and human policies 1. +[2026-03-26 07:20:06,685][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:20:22,470][__main__][INFO] - Number of regex retries in iteration 607: 0 +[2026-03-26 07:20:22,471][__main__][INFO] - agents played in iteration 607 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:20:23,244][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:20:23,267][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:20:23,291][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:20:23,314][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:20:23,314][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:20:23,315][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:20:24,132][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:20:24,584][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:20:25,082][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:20:25,574][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:20:26,075][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:20:26,571][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:20:27,073][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:20:27,572][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:20:28,067][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:20:28,561][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:20:29,060][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:20:29,558][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:20:30,058][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:20:30,556][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:20:31,053][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:20:31,552][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:20:32,052][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:20:32,553][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:20:33,052][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:20:33,548][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:20:34,045][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:20:34,542][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:20:35,039][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:20:35,535][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:20:36,031][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:20:36,529][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:20:37,024][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:20:37,520][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:20:38,015][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:20:38,510][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:20:39,006][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:20:39,500][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:20:40,002][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:20:40,494][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:20:40,986][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:20:41,482][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:20:41,975][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:20:42,472][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:20:42,968][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:20:43,462][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:20:43,959][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:20:44,451][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:20:44,946][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:20:45,439][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:20:45,929][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:20:46,423][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:20:46,918][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:20:47,415][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:20:47,914][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:20:48,415][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:20:48,913][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:20:49,412][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:20:49,911][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:20:50,411][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:20:50,910][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:20:51,408][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:20:51,907][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:20:52,404][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:20:52,897][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:20:53,391][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:20:53,886][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:20:54,379][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:20:54,873][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:20:55,373][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:20:55,867][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9730 tokens. +[2026-03-26 07:20:56,708][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.86%, Current % of VRAM taken: 60.28%, Block Peak % of device VRAM: 61.86%, ΔTime: 00:00:32 +[2026-03-26 07:20:57,480][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:20:57,483][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:20:57,485][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:20:58,313][__main__][INFO] - Iteration 608 took 51s (30.58% Gen, 67.82% Train). Generation: 15s, Training: 35s. Estimated remaining time: 33h 41m 18s. Estimated total time: 43h 1m 28s. Time estimates for 10 more iterations: 8m 36s, 100 more iterations: 1h 26m 2s, 500 more iterations: 7h 10m 14s. +[2026-03-26 07:20:58,315][__main__][INFO] - Starting iteration 608. +[2026-03-26 07:20:58,749][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 60 and human policies 1. +[2026-03-26 07:20:58,750][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:21:12,540][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 07:21:15,197][__main__][INFO] - Number of regex retries in iteration 608: 1 +[2026-03-26 07:21:15,198][__main__][INFO] - agents played in iteration 608 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:21:15,969][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.37%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:21:15,993][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.37%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:21:16,016][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.37%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:21:16,039][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.37%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:21:16,040][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:21:16,040][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:21:16,881][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:21:17,336][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:21:17,836][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:21:18,336][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:21:18,830][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:21:19,333][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:21:19,835][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:21:20,341][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:21:20,847][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:21:21,350][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:21:21,851][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:21:22,349][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:21:22,846][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:21:23,343][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:21:23,839][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:21:24,336][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:21:24,834][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:21:25,332][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:21:25,830][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:21:26,330][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:21:26,827][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:21:27,324][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:21:27,822][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:21:28,318][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:21:28,812][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:21:29,308][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:21:29,802][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:21:30,303][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:21:30,799][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:21:31,295][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:21:31,788][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:21:32,285][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:21:32,782][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:21:33,281][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:21:33,775][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:21:34,269][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:21:34,761][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:21:35,256][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:21:35,750][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:21:36,246][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:21:36,739][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:21:37,232][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:21:37,726][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:21:38,219][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:21:38,712][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:21:39,206][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:21:39,701][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:21:40,195][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:21:40,690][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:21:41,186][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:21:41,682][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:21:42,177][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:21:42,670][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:21:43,167][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:21:43,660][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:21:44,156][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:21:44,653][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:21:45,150][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:21:45,650][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:21:46,146][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:21:46,646][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:21:47,144][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:21:47,642][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:21:48,143][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:21:48,643][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9889 tokens. +[2026-03-26 07:21:49,499][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.21%, ΔTime: 00:00:32 +[2026-03-26 07:21:50,349][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:21:50,351][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:21:50,353][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:21:51,093][__main__][INFO] - Iteration 609 took 52s (31.42% Gen, 67.16% Train). Generation: 16s, Training: 35s. Estimated remaining time: 34h 16m 9s. Estimated total time: 43h 37m 11s. Time estimates for 10 more iterations: 8m 43s, 100 more iterations: 1h 27m 14s, 500 more iterations: 7h 16m 11s. +[2026-03-26 07:21:51,095][__main__][INFO] - Starting iteration 609. +[2026-03-26 07:21:51,517][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 60 and human policies 1. +[2026-03-26 07:21:51,518][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:22:06,500][__main__][INFO] - Number of regex retries in iteration 609: 0 +[2026-03-26 07:22:06,501][__main__][INFO] - agents played in iteration 609 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:22:07,264][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:22:07,288][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:22:07,311][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:22:07,334][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:22:07,335][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:22:07,335][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:22:08,131][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:22:08,587][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:22:09,089][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:22:09,588][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:22:10,086][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:22:10,583][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:22:11,082][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:22:11,586][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:22:12,087][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:22:12,583][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:22:13,081][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:22:13,579][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:22:14,076][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:22:14,573][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:22:15,071][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:22:15,567][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:22:16,064][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:22:16,561][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:22:17,063][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:22:17,560][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:22:18,059][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:22:18,559][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:22:19,056][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:22:19,553][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:22:20,051][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:22:20,549][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:22:21,049][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:22:21,548][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:22:22,047][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:22:22,544][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:22:23,045][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:22:23,542][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:22:24,039][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:22:24,536][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:22:25,030][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:22:25,527][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:22:26,021][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:22:26,514][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:22:27,010][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:22:27,506][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:22:28,000][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:22:28,502][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:22:28,998][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:22:29,498][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:22:29,993][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:22:30,491][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:22:30,988][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:22:31,484][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:22:31,981][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:22:32,481][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:22:32,979][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:22:33,478][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:22:33,976][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:22:34,473][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:22:34,969][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:22:35,468][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:22:35,968][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:22:36,467][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:22:36,965][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:22:37,463][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:22:37,960][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:22:38,457][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:22:38,955][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:22:39,457][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:22:39,956][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9894 tokens. +[2026-03-26 07:22:40,800][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 61.94%, ΔTime: 00:00:32 +[2026-03-26 07:22:41,592][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:22:41,594][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:22:41,597][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:22:42,401][__main__][INFO] - Iteration 610 took 50s (29.44% Gen, 68.97% Train). Generation: 14s, Training: 35s. Estimated remaining time: 33h 2m 20s. Estimated total time: 42h 24m 14s. Time estimates for 10 more iterations: 8m 28s, 100 more iterations: 1h 24m 48s, 500 more iterations: 7h 4m 2s. +[2026-03-26 07:22:42,404][__main__][INFO] - Starting iteration 610. +[2026-03-26 07:22:42,803][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 60 and human policies 1. +[2026-03-26 07:22:42,804][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:22:58,011][__main__][INFO] - Number of regex retries in iteration 610: 0 +[2026-03-26 07:22:58,012][__main__][INFO] - agents played in iteration 610 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:22:58,771][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:22:58,795][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:22:58,818][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:22:58,842][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:22:58,842][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:22:58,843][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:22:59,667][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:23:00,121][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:23:00,621][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:23:01,117][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:23:01,615][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:23:02,114][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:23:02,612][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:23:03,114][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:23:03,613][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:23:04,111][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:23:04,608][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:23:05,113][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:23:05,612][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:23:06,110][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:23:06,607][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:23:07,104][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:23:07,599][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:23:08,091][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:23:08,586][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:23:09,077][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:23:09,568][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:23:10,065][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:23:10,556][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:23:11,052][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:23:11,546][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:23:12,044][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:23:12,541][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:23:13,039][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:23:13,537][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:23:14,030][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:23:14,526][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:23:15,022][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:23:15,514][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:23:16,010][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:23:16,505][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:23:17,001][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:23:17,497][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:23:17,991][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:23:18,484][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:23:18,977][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:23:19,471][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:23:19,968][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:23:20,468][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:23:20,967][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:23:21,464][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:23:21,960][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:23:22,458][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:23:22,955][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:23:23,452][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:23:23,973][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:23:24,468][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:23:24,961][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:23:25,460][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:23:25,969][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:23:26,466][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:23:26,962][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:23:27,460][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:23:27,956][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:23:28,451][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:23:28,948][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:23:29,443][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:23:29,937][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:23:30,432][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:23:30,938][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:23:31,439][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9725 tokens. +[2026-03-26 07:23:32,277][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 10.93%, Current % of VRAM taken: 57.35%, Block Peak % of device VRAM: 62.04%, ΔTime: 00:00:32 +[2026-03-26 07:23:33,023][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:23:33,026][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:23:33,028][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:23:34,398][__main__][INFO] - Iteration 611 took 51s (29.48% Gen, 67.86% Train). Generation: 15s, Training: 35s. Estimated remaining time: 33h 37m 0s. Estimated total time: 42h 59m 46s. Time estimates for 10 more iterations: 8m 35s, 100 more iterations: 1h 25m 59s, 500 more iterations: 7h 9m 57s. +[2026-03-26 07:23:34,401][__main__][INFO] - Starting iteration 611. +[2026-03-26 07:23:34,802][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 61 and human policies 1. +[2026-03-26 07:23:34,803][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:23:51,396][__main__][INFO] - Number of regex retries in iteration 611: 0 +[2026-03-26 07:23:51,397][__main__][INFO] - agents played in iteration 611 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:23:52,160][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 50.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:23:52,184][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 50.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:23:52,207][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 50.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:23:52,230][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 50.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:23:52,231][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:23:52,232][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:23:53,061][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:23:53,515][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:23:54,017][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:23:54,519][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:23:55,015][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:23:55,514][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:23:56,014][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:23:56,513][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:23:57,008][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:23:57,513][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:23:58,010][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:23:58,510][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:23:59,007][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:23:59,503][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:24:00,000][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:24:00,497][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:24:00,994][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:24:01,493][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:24:02,015][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:24:02,518][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:24:03,014][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:24:03,515][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:24:04,011][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:24:04,510][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:24:05,007][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:24:05,505][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:24:06,006][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:24:06,503][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:24:07,002][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:24:07,499][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:24:07,997][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:24:08,495][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:24:08,997][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:24:09,493][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:24:09,988][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:24:10,485][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:24:10,979][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:24:11,473][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:24:11,968][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:24:12,464][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:24:12,958][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:24:13,455][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:24:13,953][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:24:14,453][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:24:14,951][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:24:15,450][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:24:15,947][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:24:16,444][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:24:16,941][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:24:17,440][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:24:17,936][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:24:18,435][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:24:18,932][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:24:19,431][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:24:19,930][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:24:20,432][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:24:20,931][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:24:21,426][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:24:21,920][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:24:22,418][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:24:22,913][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:24:23,410][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:24:23,906][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:24:24,400][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:24:24,896][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9844 tokens. +[2026-03-26 07:24:25,737][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.98%, Current % of VRAM taken: 60.41%, Block Peak % of device VRAM: 61.97%, ΔTime: 00:00:32 +[2026-03-26 07:24:26,534][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:24:26,537][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:24:26,538][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:24:27,399][__main__][INFO] - Iteration 612 took 52s (31.55% Gen, 66.81% Train). Generation: 16s, Training: 35s. Estimated remaining time: 34h 26m 15s. Estimated total time: 43h 49m 53s. Time estimates for 10 more iterations: 8m 45s, 100 more iterations: 1h 27m 39s, 500 more iterations: 7h 18m 18s. +[2026-03-26 07:24:27,402][__main__][INFO] - Starting iteration 612. +[2026-03-26 07:24:27,800][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 61 and human policies 1. +[2026-03-26 07:24:27,801][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:24:42,678][__main__][INFO] - Number of regex retries in iteration 612: 0 +[2026-03-26 07:24:42,679][__main__][INFO] - agents played in iteration 612 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:24:43,473][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:24:43,498][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:24:43,521][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:24:43,544][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:24:43,545][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:24:43,545][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:24:44,843][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:24:45,419][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:24:45,926][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:24:46,427][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:24:46,930][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:24:47,430][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:24:47,929][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:24:48,427][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:24:48,924][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:24:49,420][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:24:49,914][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:24:50,413][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:24:50,909][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:24:51,403][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:24:51,894][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:24:52,397][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:24:52,893][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:24:53,391][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:24:53,888][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:24:54,380][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:24:54,874][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:24:55,367][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:24:55,862][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:24:56,354][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:24:56,848][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:24:57,348][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:24:57,846][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:24:58,342][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:24:58,837][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:24:59,335][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:24:59,833][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:25:00,332][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:25:00,826][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:25:01,321][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:25:01,817][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:25:02,312][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:25:02,806][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:25:03,300][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:25:03,794][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:25:04,289][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:25:04,785][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:25:05,282][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:25:05,780][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:25:06,278][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:25:06,774][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:25:07,270][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:25:07,768][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:25:08,265][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:25:08,762][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:25:09,258][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:25:09,756][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:25:10,253][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:25:10,750][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:25:11,249][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:25:11,746][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:25:12,244][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:25:12,742][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:25:13,239][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:25:13,736][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:25:14,237][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:25:14,734][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:25:15,231][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:25:15,729][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:25:16,226][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:25:16,723][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9834 tokens. +[2026-03-26 07:25:17,630][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 61.90%, ΔTime: 00:00:33 +[2026-03-26 07:25:18,458][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:25:18,460][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:25:18,462][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:25:19,255][__main__][INFO] - Iteration 613 took 51s (28.91% Gen, 69.54% Train). Generation: 14s, Training: 35s. Estimated remaining time: 33h 28m 15s. Estimated total time: 42h 52m 46s. Time estimates for 10 more iterations: 8m 34s, 100 more iterations: 1h 25m 45s, 500 more iterations: 7h 8m 47s. +[2026-03-26 07:25:19,277][__main__][INFO] - Starting iteration 613. +[2026-03-26 07:25:19,817][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 61 and human policies 1. +[2026-03-26 07:25:19,818][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:25:43,356][__main__][INFO] - Number of regex retries in iteration 613: 0 +[2026-03-26 07:25:43,357][__main__][INFO] - agents played in iteration 613 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:25:44,132][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:25:44,156][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:25:44,179][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:25:44,201][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:25:44,202][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:25:44,202][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:25:45,010][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:25:45,463][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:25:45,964][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:25:46,461][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:25:46,958][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:25:47,455][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:25:47,952][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:25:48,454][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:25:48,952][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:25:49,448][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:25:49,946][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:25:50,448][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:25:50,948][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:25:51,447][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:25:51,943][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:25:52,440][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:25:52,938][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:25:53,437][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:25:53,933][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:25:54,431][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:25:54,934][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:25:55,432][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:25:55,930][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:25:56,428][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:25:56,930][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:25:57,425][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:25:57,924][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:25:58,420][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:25:58,916][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:25:59,412][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:25:59,907][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:26:00,402][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:26:00,908][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:26:01,403][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:26:01,897][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:26:02,391][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:26:02,888][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:26:03,382][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:26:03,877][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:26:04,372][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:26:04,867][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:26:05,357][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:26:05,850][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:26:06,347][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:26:06,844][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:26:07,341][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:26:07,834][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:26:08,327][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:26:08,823][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:26:09,320][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:26:09,817][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:26:10,316][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:26:10,814][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:26:11,312][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:26:11,812][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:26:12,315][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:26:12,814][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:26:13,308][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:26:13,800][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:26:14,294][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:26:14,787][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:26:15,279][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:26:15,776][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:26:16,268][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:26:16,766][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9793 tokens. +[2026-03-26 07:26:17,609][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.12%, ΔTime: 00:00:32 +[2026-03-26 07:26:18,425][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:26:18,427][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:26:18,429][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:26:19,235][__main__][INFO] - Iteration 614 took 59s (39.62% Gen, 59.02% Train). Generation: 23s, Training: 35s. Estimated remaining time: 40h 5m 25s. Estimated total time: 49h 30m 56s. Time estimates for 10 more iterations: 9m 54s, 100 more iterations: 1h 39m 1s, 500 more iterations: 8h 15m 9s. +[2026-03-26 07:26:19,240][__main__][INFO] - Starting iteration 614. +[2026-03-26 07:26:19,649][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 61 and human policies 1. +[2026-03-26 07:26:19,649][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:26:26,368][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 07:26:36,121][__main__][INFO] - Number of regex retries in iteration 614: 1 +[2026-03-26 07:26:36,122][__main__][INFO] - agents played in iteration 614 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:26:36,880][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:26:36,904][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:26:36,927][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:26:36,950][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:26:36,951][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:26:36,951][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:26:37,758][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:26:38,207][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:26:38,704][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:26:39,203][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:26:39,707][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:26:40,203][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:26:40,697][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:26:41,192][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:26:41,687][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:26:42,184][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:26:42,683][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:26:43,184][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:26:43,677][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:26:44,174][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:26:44,675][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:26:45,173][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:26:45,673][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:26:46,170][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:26:46,666][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:26:47,162][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:26:47,658][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:26:48,160][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:26:48,655][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:26:49,152][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:26:49,649][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:26:50,148][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:26:50,647][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:26:51,144][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:26:51,639][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:26:52,137][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:26:52,638][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:26:53,137][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:26:53,635][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:26:54,134][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:26:54,632][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:26:55,131][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:26:55,628][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:26:56,125][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:26:56,622][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:26:57,120][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:26:57,619][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:26:58,113][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:26:58,611][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:26:59,109][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:26:59,603][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:27:00,097][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:27:00,594][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:27:01,091][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:27:01,590][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:27:02,088][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:27:02,585][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:27:03,081][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:27:03,578][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:27:04,074][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:27:04,571][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:27:05,067][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:27:05,563][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:27:06,063][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:27:06,562][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:27:07,058][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:27:07,553][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:27:08,050][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:27:08,548][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:27:09,044][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:27:09,540][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9817 tokens. +[2026-03-26 07:27:10,377][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 12.98%, Current % of VRAM taken: 59.41%, Block Peak % of device VRAM: 61.98%, ΔTime: 00:00:32 +[2026-03-26 07:27:11,090][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:27:11,093][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:27:11,094][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:27:12,180][__main__][INFO] - Iteration 615 took 52s (31.36% Gen, 66.57% Train). Generation: 16s, Training: 34s. Estimated remaining time: 34h 20m 14s. Estimated total time: 43h 46m 37s. Time estimates for 10 more iterations: 8m 45s, 100 more iterations: 1h 27m 33s, 500 more iterations: 7h 17m 46s. +[2026-03-26 07:27:12,183][__main__][INFO] - Starting iteration 615. +[2026-03-26 07:27:12,595][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 61 and human policies 1. +[2026-03-26 07:27:12,596][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:27:29,624][__main__][INFO] - Number of regex retries in iteration 615: 0 +[2026-03-26 07:27:29,625][__main__][INFO] - agents played in iteration 615 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:27:30,397][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:27:30,421][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:27:30,444][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:27:30,467][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:27:30,468][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:27:30,468][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:27:31,279][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:27:31,731][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:27:32,236][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:27:32,735][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:27:33,234][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:27:33,733][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:27:34,233][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:27:34,732][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:27:35,237][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:27:35,737][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:27:36,236][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:27:36,735][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:27:37,234][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:27:37,729][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:27:38,225][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:27:38,721][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:27:39,220][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:27:39,717][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:27:40,212][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:27:40,713][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:27:41,209][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:27:41,705][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:27:42,197][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:27:42,693][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:27:43,188][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:27:43,680][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:27:44,179][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:27:44,673][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:27:45,172][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:27:45,672][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:27:46,163][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:27:46,663][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:27:47,166][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:27:47,660][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:27:48,156][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:27:48,653][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:27:49,148][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:27:49,644][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:27:50,141][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:27:50,637][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:27:51,132][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:27:51,628][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:27:52,124][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:27:52,621][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:27:53,121][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:27:53,618][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:27:54,116][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:27:54,617][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:27:55,119][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:27:55,619][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:27:56,124][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:27:56,626][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:27:57,122][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:27:57,618][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:27:58,119][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:27:58,616][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:27:59,109][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:27:59,603][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:28:00,097][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:28:00,594][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:28:01,091][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:28:01,587][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:28:02,082][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:28:02,577][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:28:03,072][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9772 tokens. +[2026-03-26 07:28:03,910][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.99%, Current % of VRAM taken: 60.42%, Block Peak % of device VRAM: 62.20%, ΔTime: 00:00:32 +[2026-03-26 07:28:04,677][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:28:04,680][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:28:04,681][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:28:05,408][__main__][INFO] - Iteration 616 took 52s (32.24% Gen, 66.38% Train). Generation: 17s, Training: 35s. Estimated remaining time: 34h 33m 24s. Estimated total time: 44h 0m 41s. Time estimates for 10 more iterations: 8m 48s, 100 more iterations: 1h 28m 1s, 500 more iterations: 7h 20m 6s. +[2026-03-26 07:28:05,410][__main__][INFO] - Starting iteration 616. +[2026-03-26 07:28:05,813][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 61 and human policies 1. +[2026-03-26 07:28:05,814][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:28:20,913][__main__][INFO] - Number of regex retries in iteration 616: 0 +[2026-03-26 07:28:20,913][__main__][INFO] - agents played in iteration 616 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:28:21,696][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:28:21,719][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:28:21,743][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:28:21,767][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:28:21,767][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:28:21,768][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:28:22,567][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:28:23,024][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:28:23,530][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:28:24,032][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:28:24,533][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:28:25,035][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:28:25,536][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:28:26,037][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:28:26,538][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:28:27,034][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:28:27,531][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:28:28,026][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:28:28,525][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:28:29,020][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:28:29,519][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:28:30,016][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:28:30,516][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:28:31,019][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:28:31,523][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:28:32,025][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:28:32,524][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:28:33,024][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:28:33,528][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:28:34,032][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:28:34,532][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:28:35,030][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:28:35,527][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:28:36,023][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:28:36,521][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:28:37,019][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:28:37,517][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:28:38,014][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:28:38,518][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:28:39,015][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:28:39,511][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:28:40,007][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:28:40,504][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:28:41,000][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:28:41,498][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:28:41,996][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:28:42,494][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:28:42,992][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:28:43,488][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:28:43,984][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:28:44,480][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:28:44,979][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:28:45,475][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:28:45,973][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:28:46,473][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:28:46,976][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:28:47,476][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:28:47,974][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:28:48,471][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:28:48,970][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:28:49,466][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:28:49,963][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:28:50,462][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:28:50,959][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:28:51,454][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:28:51,950][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:28:52,446][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:28:52,942][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:28:53,438][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:28:53,933][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:28:54,429][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9946 tokens. +[2026-03-26 07:28:55,264][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.07%, ΔTime: 00:00:32 +[2026-03-26 07:28:55,981][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:28:55,983][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:28:55,985][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:28:56,651][__main__][INFO] - Iteration 617 took 50s (29.70% Gen, 68.99% Train). Generation: 15s, Training: 35s. Estimated remaining time: 32h 53m 49s. Estimated total time: 42h 21m 57s. Time estimates for 10 more iterations: 8m 28s, 100 more iterations: 1h 24m 43s, 500 more iterations: 7h 3m 39s. +[2026-03-26 07:28:56,655][__main__][INFO] - Starting iteration 617. +[2026-03-26 07:28:57,061][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 61 and human policies 1. +[2026-03-26 07:28:57,062][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:29:04,424][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 07:29:12,372][__main__][INFO] - Number of regex retries in iteration 617: 1 +[2026-03-26 07:29:12,373][__main__][INFO] - agents played in iteration 617 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:29:13,140][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:29:13,164][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:29:13,187][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:29:13,211][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:29:13,211][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:29:13,212][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:29:14,022][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:29:14,480][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:29:14,979][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:29:15,479][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:29:15,979][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:29:16,478][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:29:16,978][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:29:17,477][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:29:17,974][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:29:18,471][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:29:18,971][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:29:19,471][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:29:19,972][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:29:20,472][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:29:20,969][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:29:21,469][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:29:21,970][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:29:22,478][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:29:22,972][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:29:23,468][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:29:23,968][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:29:24,468][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:29:24,970][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:29:25,462][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:29:25,958][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:29:26,461][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:29:26,959][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:29:27,457][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:29:27,954][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:29:28,453][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:29:28,950][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:29:29,449][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:29:29,948][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:29:30,445][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:29:30,942][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:29:31,443][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:29:31,944][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:29:32,440][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:29:32,941][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:29:33,440][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:29:33,938][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:29:34,437][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:29:34,936][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:29:35,434][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:29:35,933][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:29:36,432][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:29:36,930][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:29:37,433][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:29:37,935][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:29:38,437][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:29:38,936][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:29:39,431][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:29:39,926][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:29:40,425][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:29:40,921][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:29:41,414][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:29:41,909][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:29:42,402][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:29:42,899][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:29:43,397][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:29:43,896][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:29:44,393][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:29:44,892][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:29:45,391][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:29:45,892][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9874 tokens. +[2026-03-26 07:29:46,734][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 61.97%, ΔTime: 00:00:32 +[2026-03-26 07:29:47,509][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:29:47,511][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:29:47,514][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:29:48,251][__main__][INFO] - Iteration 618 took 51s (29.91% Gen, 68.65% Train). Generation: 15s, Training: 35s. Estimated remaining time: 33h 10m 31s. Estimated total time: 42h 39m 31s. Time estimates for 10 more iterations: 8m 31s, 100 more iterations: 1h 25m 19s, 500 more iterations: 7h 6m 35s. +[2026-03-26 07:29:48,253][__main__][INFO] - Starting iteration 618. +[2026-03-26 07:29:48,652][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 61 and human policies 1. +[2026-03-26 07:29:48,653][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:30:04,837][__main__][INFO] - Number of regex retries in iteration 618: 0 +[2026-03-26 07:30:04,838][__main__][INFO] - agents played in iteration 618 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:30:05,615][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:30:05,639][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:30:05,662][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:30:05,686][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:30:05,687][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:30:05,687][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:30:06,492][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:30:06,945][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:30:07,454][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:30:07,952][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:30:08,451][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:30:08,951][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:30:09,450][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:30:09,947][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:30:10,444][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:30:10,942][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:30:11,438][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:30:11,932][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:30:12,428][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:30:12,928][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:30:13,424][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:30:13,923][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:30:14,420][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:30:14,922][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:30:15,429][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:30:15,933][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:30:16,439][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:30:16,942][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:30:17,447][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:30:17,948][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:30:18,453][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:30:18,949][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:30:19,445][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:30:19,943][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:30:20,443][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:30:20,941][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:30:21,439][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:30:21,938][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:30:22,437][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:30:22,935][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:30:23,434][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:30:23,933][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:30:24,431][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:30:24,929][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:30:25,426][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:30:25,924][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:30:26,422][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:30:26,919][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:30:27,423][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:30:27,925][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:30:28,431][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:30:28,930][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:30:29,428][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:30:29,925][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:30:30,422][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:30:30,923][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:30:31,423][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:30:31,922][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:30:32,422][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:30:32,920][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:30:33,420][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:30:33,920][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:30:34,419][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:30:34,918][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:30:35,410][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:30:35,910][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:30:36,410][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:30:36,907][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:30:37,400][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:30:37,894][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:30:38,390][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10000 tokens. +[2026-03-26 07:30:39,224][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.48%, Block Peak % of device VRAM: 62.47%, ΔTime: 00:00:32 +[2026-03-26 07:30:39,960][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:30:39,962][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:30:39,964][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:30:40,690][__main__][INFO] - Iteration 619 took 52s (31.10% Gen, 67.50% Train). Generation: 16s, Training: 35s. Estimated remaining time: 33h 52m 2s. Estimated total time: 43h 21m 54s. Time estimates for 10 more iterations: 8m 40s, 100 more iterations: 1h 26m 43s, 500 more iterations: 7h 13m 39s. +[2026-03-26 07:30:40,692][__main__][INFO] - Starting iteration 619. +[2026-03-26 07:30:41,094][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 61 and human policies 1. +[2026-03-26 07:30:41,094][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:30:57,098][__main__][INFO] - Number of regex retries in iteration 619: 0 +[2026-03-26 07:30:57,099][__main__][INFO] - agents played in iteration 619 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:30:57,867][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:30:57,891][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:30:57,914][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:30:57,937][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:30:57,938][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:30:57,938][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:30:58,748][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:30:59,201][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:30:59,703][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:31:00,200][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:31:00,698][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:31:01,195][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:31:01,692][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:31:02,189][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:31:02,695][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:31:03,195][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:31:03,695][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:31:04,191][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:31:04,689][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:31:05,194][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:31:05,690][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:31:06,188][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:31:06,689][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:31:07,194][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:31:07,695][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:31:08,201][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:31:08,702][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:31:09,201][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:31:09,701][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:31:10,201][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:31:10,700][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:31:11,197][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:31:11,694][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:31:12,191][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:31:12,684][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:31:13,180][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:31:13,678][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:31:14,181][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:31:14,681][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:31:15,177][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:31:15,673][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:31:16,171][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:31:16,669][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:31:17,161][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:31:17,660][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:31:18,158][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:31:18,664][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:31:19,165][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:31:19,665][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:31:20,164][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:31:20,660][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:31:21,161][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:31:21,666][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:31:22,168][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:31:22,665][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:31:23,163][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:31:23,662][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:31:24,160][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:31:24,658][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:31:25,155][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:31:25,654][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:31:26,152][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:31:26,650][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:31:27,148][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:31:27,645][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:31:28,142][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:31:28,639][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:31:29,137][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:31:29,630][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:31:30,128][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:31:30,627][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9853 tokens. +[2026-03-26 07:31:31,460][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 62.08%, ΔTime: 00:00:32 +[2026-03-26 07:31:32,199][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:31:32,201][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:31:32,203][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:31:33,041][__main__][INFO] - Iteration 620 took 51s (30.81% Gen, 67.57% Train). Generation: 16s, Training: 35s. Estimated remaining time: 33h 46m 39s. Estimated total time: 43h 17m 24s. Time estimates for 10 more iterations: 8m 39s, 100 more iterations: 1h 26m 34s, 500 more iterations: 7h 12m 54s. +[2026-03-26 07:31:33,043][__main__][INFO] - Starting iteration 620. +[2026-03-26 07:31:33,447][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 61 and human policies 1. +[2026-03-26 07:31:33,448][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:31:47,381][__main__][INFO] - Number of regex retries in iteration 620: 0 +[2026-03-26 07:31:47,382][__main__][INFO] - agents played in iteration 620 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:31:48,162][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:31:48,186][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:31:48,209][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:31:48,233][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:31:48,233][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:31:48,234][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:31:49,049][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:31:49,505][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:31:50,007][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:31:50,506][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:31:51,004][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:31:51,501][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:31:51,999][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:31:52,497][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:31:52,994][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:31:53,491][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:31:53,993][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:31:54,489][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:31:54,987][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:31:55,483][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:31:55,981][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:31:56,480][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:31:56,975][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:31:57,475][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:31:57,976][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:31:58,478][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:31:58,982][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:31:59,482][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:31:59,981][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:32:00,479][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:32:00,976][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:32:01,473][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:32:01,970][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:32:02,467][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:32:02,963][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:32:03,458][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:32:03,956][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:32:04,457][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:32:04,953][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:32:05,453][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:32:05,970][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:32:06,469][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:32:06,969][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:32:07,467][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:32:07,971][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:32:08,476][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:32:08,978][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:32:09,476][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:32:09,978][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:32:10,475][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:32:10,976][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:32:11,472][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:32:11,968][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:32:12,465][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:32:12,965][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:32:13,456][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:32:13,953][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:32:14,451][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:32:14,947][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:32:15,447][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:32:15,938][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:32:16,433][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:32:16,927][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:32:17,424][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:32:17,921][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:32:18,418][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:32:18,917][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:32:19,414][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:32:19,913][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:32:20,412][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:32:20,910][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9880 tokens. +[2026-03-26 07:32:21,757][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.04%, ΔTime: 00:00:32 +[2026-03-26 07:32:22,512][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:32:22,514][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:32:22,516][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:32:23,911][__main__][INFO] - Iteration 621 took 50s (27.61% Gen, 69.62% Train). Generation: 13s, Training: 35s. Estimated remaining time: 32h 31m 42s. Estimated total time: 42h 3m 17s. Time estimates for 10 more iterations: 8m 24s, 100 more iterations: 1h 24m 6s, 500 more iterations: 7h 0m 32s. +[2026-03-26 07:32:23,913][__main__][INFO] - Starting iteration 621. +[2026-03-26 07:32:24,320][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 62 and human policies 1. +[2026-03-26 07:32:24,321][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:32:40,515][__main__][INFO] - Number of regex retries in iteration 621: 0 +[2026-03-26 07:32:40,516][__main__][INFO] - agents played in iteration 621 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:32:41,291][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:32:41,315][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:32:41,338][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:32:41,361][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:32:41,362][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:32:41,362][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:32:42,170][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:32:42,621][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:32:43,121][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:32:43,617][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:32:44,111][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:32:44,611][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:32:45,108][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:32:45,606][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:32:46,105][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:32:46,600][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:32:47,103][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:32:47,604][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:32:48,103][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:32:48,606][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:32:49,101][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:32:49,601][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:32:50,101][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:32:50,598][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:32:51,094][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:32:51,591][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:32:52,089][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:32:52,585][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:32:53,081][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:32:53,577][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:32:54,073][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:32:54,570][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:32:55,067][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:32:55,564][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:32:56,062][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:32:56,563][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:32:57,059][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:32:57,560][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:32:58,062][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:32:58,564][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:32:59,062][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:32:59,558][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:33:00,056][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:33:00,552][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:33:01,048][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:33:01,544][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:33:02,039][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:33:02,534][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:33:03,028][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:33:03,522][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:33:04,017][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:33:04,513][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:33:05,008][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:33:05,504][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:33:05,999][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:33:06,495][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:33:06,987][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:33:07,482][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:33:07,981][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:33:08,477][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:33:08,973][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:33:09,467][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:33:09,961][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:33:10,458][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:33:10,972][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:33:11,473][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:33:11,971][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:33:12,469][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:33:12,967][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:33:13,473][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:33:13,970][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9812 tokens. +[2026-03-26 07:33:14,796][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.09%, ΔTime: 00:00:32 +[2026-03-26 07:33:15,543][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:33:15,545][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:33:15,547][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:33:16,380][__main__][INFO] - Iteration 622 took 52s (31.11% Gen, 67.29% Train). Generation: 16s, Training: 35s. Estimated remaining time: 33h 50m 34s. Estimated total time: 43h 23m 1s. Time estimates for 10 more iterations: 8m 40s, 100 more iterations: 1h 26m 46s, 500 more iterations: 7h 13m 50s. +[2026-03-26 07:33:16,382][__main__][INFO] - Starting iteration 622. +[2026-03-26 07:33:16,782][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 62 and human policies 1. +[2026-03-26 07:33:16,783][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:33:31,664][__main__][INFO] - Number of regex retries in iteration 622: 0 +[2026-03-26 07:33:31,664][__main__][INFO] - agents played in iteration 622 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:33:32,427][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:33:32,450][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:33:32,473][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:33:32,496][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:33:32,497][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:33:32,498][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:33:33,305][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:33:33,753][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:33:34,252][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:33:34,745][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:33:35,239][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:33:35,737][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:33:36,236][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:33:36,731][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:33:37,226][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:33:37,718][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:33:38,213][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:33:38,716][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:33:39,216][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:33:39,708][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:33:40,204][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:33:40,699][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:33:41,193][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:33:41,694][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:33:42,189][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:33:42,687][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:33:43,186][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:33:43,679][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:33:44,178][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:33:44,672][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:33:45,165][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:33:45,667][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:33:46,165][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:33:46,663][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:33:47,161][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:33:47,664][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:33:48,169][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:33:48,689][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:33:49,199][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:33:49,697][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:33:50,197][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:33:50,695][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:33:51,192][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:33:51,685][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:33:52,182][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:33:52,679][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:33:53,176][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:33:53,672][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:33:54,168][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:33:54,669][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:33:55,162][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:33:55,657][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:33:56,154][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:33:56,650][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:33:57,147][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:33:57,644][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:33:58,139][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:33:58,636][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:33:59,133][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:33:59,632][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:34:00,131][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:34:00,626][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:34:01,123][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:34:01,619][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:34:02,119][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:34:02,614][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:34:03,110][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:34:03,605][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:34:04,101][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:34:04,596][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:34:05,093][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9756 tokens. +[2026-03-26 07:34:05,925][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.03%, Current % of VRAM taken: 60.46%, Block Peak % of device VRAM: 61.95%, ΔTime: 00:00:32 +[2026-03-26 07:34:06,687][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:34:06,689][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:34:06,691][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:34:07,586][__main__][INFO] - Iteration 623 took 50s (29.29% Gen, 68.94% Train). Generation: 14s, Training: 35s. Estimated remaining time: 32h 46m 56s. Estimated total time: 42h 20m 15s. Time estimates for 10 more iterations: 8m 28s, 100 more iterations: 1h 24m 40s, 500 more iterations: 7h 3m 22s. +[2026-03-26 07:34:07,589][__main__][INFO] - Starting iteration 623. +[2026-03-26 07:34:07,989][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 62 and human policies 1. +[2026-03-26 07:34:07,989][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:34:28,280][__main__][INFO] - Number of regex retries in iteration 623: 0 +[2026-03-26 07:34:28,280][__main__][INFO] - agents played in iteration 623 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:34:29,068][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:34:29,091][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:34:29,115][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:34:29,139][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:34:29,140][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:34:29,140][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:34:29,963][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:34:30,418][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:34:30,919][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:34:31,417][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:34:31,918][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:34:32,417][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:34:32,915][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:34:33,411][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:34:33,907][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:34:34,405][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:34:34,902][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:34:35,398][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:34:35,895][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:34:36,392][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:34:36,889][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:34:37,382][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:34:37,879][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:34:38,375][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:34:38,873][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:34:39,373][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:34:39,874][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:34:40,382][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:34:40,877][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:34:41,374][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:34:41,870][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:34:42,367][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:34:42,864][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:34:43,361][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:34:43,857][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:34:44,353][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:34:44,850][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:34:45,346][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:34:45,843][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:34:46,337][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:34:46,833][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:34:47,326][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:34:47,822][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:34:48,320][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:34:48,817][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:34:49,314][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:34:49,812][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:34:50,313][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:34:50,811][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:34:51,310][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:34:51,806][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:34:52,301][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:34:52,797][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:34:53,292][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:34:53,790][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:34:54,285][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:34:54,781][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:34:55,277][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:34:55,772][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:34:56,267][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:34:56,764][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:34:57,260][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:34:57,755][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:34:58,253][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:34:58,755][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:34:59,253][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:34:59,750][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:35:00,248][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:35:00,749][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:35:01,248][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:35:01,750][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9888 tokens. +[2026-03-26 07:35:02,592][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.23%, Current % of VRAM taken: 60.65%, Block Peak % of device VRAM: 62.18%, ΔTime: 00:00:32 +[2026-03-26 07:35:03,302][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:35:03,304][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:35:03,306][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:35:04,178][__main__][INFO] - Iteration 624 took 56s (36.11% Gen, 62.33% Train). Generation: 20s, Training: 35s. Estimated remaining time: 37h 15m 13s. Estimated total time: 46h 49m 29s. Time estimates for 10 more iterations: 9m 21s, 100 more iterations: 1h 33m 38s, 500 more iterations: 7h 48m 14s. +[2026-03-26 07:35:04,180][__main__][INFO] - Starting iteration 624. +[2026-03-26 07:35:04,581][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 62 and human policies 1. +[2026-03-26 07:35:04,582][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:35:20,520][__main__][INFO] - Number of regex retries in iteration 624: 0 +[2026-03-26 07:35:20,521][__main__][INFO] - agents played in iteration 624 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:35:21,293][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:35:21,317][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:35:21,340][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:35:21,363][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:35:21,364][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:35:21,364][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:35:22,167][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:35:22,620][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:35:23,122][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:35:23,620][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:35:24,115][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:35:24,609][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:35:25,102][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:35:25,598][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:35:26,094][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:35:26,593][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:35:27,091][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:35:27,588][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:35:28,086][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:35:28,582][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:35:29,080][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:35:29,581][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:35:30,084][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:35:30,580][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:35:31,082][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:35:31,580][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:35:32,071][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:35:32,568][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:35:33,066][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:35:33,558][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:35:34,055][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:35:34,556][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:35:35,055][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:35:35,553][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:35:36,052][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:35:36,551][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:35:37,049][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:35:37,547][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:35:38,067][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:35:38,565][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:35:39,056][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:35:39,558][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:35:40,052][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:35:40,546][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:35:41,037][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:35:41,530][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:35:42,024][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:35:42,521][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:35:43,018][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:35:43,518][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:35:44,015][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:35:44,513][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:35:45,010][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:35:45,511][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:35:46,007][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:35:46,504][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:35:47,002][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:35:47,498][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:35:47,996][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:35:48,495][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:35:48,994][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:35:49,490][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:35:49,987][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:35:50,487][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:35:50,979][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:35:51,476][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:35:51,973][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:35:52,473][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:35:52,970][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:35:53,470][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:35:53,969][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9867 tokens. +[2026-03-26 07:35:54,803][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.16%, Current % of VRAM taken: 60.58%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 07:35:55,543][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:35:55,545][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:35:55,547][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:35:56,545][__main__][INFO] - Iteration 625 took 51s (30.67% Gen, 67.40% Train). Generation: 15s, Training: 35s. Estimated remaining time: 33h 43m 4s. Estimated total time: 43h 18m 12s. Time estimates for 10 more iterations: 8m 39s, 100 more iterations: 1h 26m 36s, 500 more iterations: 7h 13m 2s. +[2026-03-26 07:35:56,548][__main__][INFO] - Starting iteration 625. +[2026-03-26 07:35:56,989][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 62 and human policies 1. +[2026-03-26 07:35:56,989][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:36:11,691][__main__][INFO] - Number of regex retries in iteration 625: 0 +[2026-03-26 07:36:11,692][__main__][INFO] - agents played in iteration 625 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:36:12,459][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:36:12,482][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:36:12,505][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:36:12,528][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:36:12,529][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:36:12,529][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:36:13,333][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:36:13,783][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:36:14,288][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:36:14,782][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:36:15,280][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:36:15,777][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:36:16,269][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:36:16,769][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:36:17,261][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:36:17,757][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:36:18,253][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:36:18,750][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:36:19,247][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:36:19,749][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:36:20,252][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:36:20,752][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:36:21,249][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:36:21,748][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:36:22,247][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:36:22,746][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:36:23,245][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:36:23,743][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:36:24,241][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:36:24,738][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:36:25,237][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:36:25,737][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:36:26,237][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:36:26,736][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:36:27,235][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:36:27,735][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:36:28,234][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:36:28,734][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:36:29,237][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:36:29,734][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:36:30,231][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:36:30,727][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:36:31,225][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:36:31,722][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:36:32,219][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:36:32,715][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:36:33,212][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:36:33,710][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:36:34,204][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:36:34,699][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:36:35,194][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:36:35,687][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:36:36,179][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:36:36,676][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:36:37,172][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:36:37,668][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:36:38,165][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:36:38,663][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:36:39,197][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:36:39,688][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:36:40,184][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:36:40,679][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:36:41,176][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:36:41,676][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:36:42,171][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:36:42,668][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:36:43,165][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:36:43,660][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:36:44,156][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:36:44,653][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:36:45,149][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9854 tokens. +[2026-03-26 07:36:45,981][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.04%, Current % of VRAM taken: 60.47%, Block Peak % of device VRAM: 61.95%, ΔTime: 00:00:32 +[2026-03-26 07:36:46,728][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:36:46,730][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:36:46,732][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:36:47,415][__main__][INFO] - Iteration 626 took 50s (29.16% Gen, 69.49% Train). Generation: 14s, Training: 35s. Estimated remaining time: 32h 25m 21s. Estimated total time: 42h 1m 20s. Time estimates for 10 more iterations: 8m 24s, 100 more iterations: 1h 24m 2s, 500 more iterations: 7h 0m 13s. +[2026-03-26 07:36:47,417][__main__][INFO] - Starting iteration 626. +[2026-03-26 07:36:47,845][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 62 and human policies 1. +[2026-03-26 07:36:47,846][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:37:03,729][__main__][INFO] - Number of regex retries in iteration 626: 0 +[2026-03-26 07:37:03,730][__main__][INFO] - agents played in iteration 626 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:37:04,540][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:37:04,564][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:37:04,588][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:37:04,612][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:37:04,612][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:37:04,613][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:37:05,432][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:37:05,885][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:37:06,389][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:37:06,883][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:37:07,382][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:37:07,883][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:37:08,380][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:37:08,880][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:37:09,382][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:37:09,882][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:37:10,389][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:37:10,889][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:37:11,388][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:37:11,888][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:37:12,387][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:37:12,885][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:37:13,383][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:37:13,886][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:37:14,382][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:37:14,880][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:37:15,378][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:37:15,883][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:37:16,380][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:37:16,879][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:37:17,377][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:37:17,876][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:37:18,373][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:37:18,871][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:37:19,369][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:37:19,863][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:37:20,362][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:37:20,859][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:37:21,353][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:37:21,851][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:37:22,349][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:37:22,850][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:37:23,350][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:37:23,848][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:37:24,348][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:37:24,848][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:37:25,345][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:37:25,843][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:37:26,340][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:37:26,838][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:37:27,336][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:37:27,834][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:37:28,335][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:37:28,832][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:37:29,335][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:37:29,831][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:37:30,325][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:37:30,818][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:37:31,312][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:37:31,810][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:37:32,306][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:37:32,801][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:37:33,296][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:37:33,792][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:37:34,290][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:37:34,786][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:37:35,282][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:37:35,782][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:37:36,278][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:37:36,775][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:37:37,272][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9930 tokens. +[2026-03-26 07:37:38,116][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 07:37:38,867][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:37:38,870][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:37:38,872][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:37:39,621][__main__][INFO] - Iteration 627 took 51s (30.68% Gen, 67.87% Train). Generation: 15s, Training: 35s. Estimated remaining time: 33h 32m 0s. Estimated total time: 43h 8m 50s. Time estimates for 10 more iterations: 8m 37s, 100 more iterations: 1h 26m 17s, 500 more iterations: 7h 11m 28s. +[2026-03-26 07:37:39,624][__main__][INFO] - Starting iteration 627. +[2026-03-26 07:37:40,026][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 62 and human policies 1. +[2026-03-26 07:37:40,026][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:37:54,164][__main__][INFO] - Number of regex retries in iteration 627: 0 +[2026-03-26 07:37:54,165][__main__][INFO] - agents played in iteration 627 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:37:54,928][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:37:54,952][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:37:54,975][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:37:54,998][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:37:54,999][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:37:54,999][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:37:55,812][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:37:56,267][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:37:56,770][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:37:57,266][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:37:57,762][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:37:58,260][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:37:58,756][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:37:59,254][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:37:59,757][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:38:00,256][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:38:00,756][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:38:01,257][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:38:01,757][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:38:02,254][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:38:02,752][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:38:03,250][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:38:03,748][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:38:04,245][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:38:04,742][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:38:05,243][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:38:05,745][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:38:06,246][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:38:06,745][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:38:07,243][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:38:07,740][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:38:08,261][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:38:08,757][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:38:09,253][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:38:09,751][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:38:10,252][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:38:10,748][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:38:11,242][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:38:11,736][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:38:12,233][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:38:12,729][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:38:13,226][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:38:13,721][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:38:14,220][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:38:14,721][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:38:15,217][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:38:15,715][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:38:16,212][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:38:16,716][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:38:17,214][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:38:17,715][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:38:18,213][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:38:18,710][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:38:19,211][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:38:19,710][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:38:20,210][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:38:20,710][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:38:21,208][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:38:21,705][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:38:22,207][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:38:22,705][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:38:23,226][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:38:23,724][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:38:24,222][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:38:24,719][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:38:25,212][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:38:25,714][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:38:26,212][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:38:26,710][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:38:27,209][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:38:27,712][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9892 tokens. +[2026-03-26 07:38:28,543][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.15%, ΔTime: 00:00:32 +[2026-03-26 07:38:29,293][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:38:29,295][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:38:29,297][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:38:30,026][__main__][INFO] - Iteration 628 took 50s (28.28% Gen, 70.26% Train). Generation: 14s, Training: 35s. Estimated remaining time: 32h 2m 23s. Estimated total time: 41h 40m 4s. Time estimates for 10 more iterations: 8m 20s, 100 more iterations: 1h 23m 20s, 500 more iterations: 6h 56m 40s. +[2026-03-26 07:38:30,029][__main__][INFO] - Starting iteration 628. +[2026-03-26 07:38:30,426][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 62 and human policies 1. +[2026-03-26 07:38:30,427][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:38:44,107][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 07:38:45,176][__main__][INFO] - Number of regex retries in iteration 628: 1 +[2026-03-26 07:38:45,177][__main__][INFO] - agents played in iteration 628 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:38:45,960][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:38:45,984][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:38:46,008][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:38:46,031][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:38:46,032][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:38:46,032][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:38:46,875][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:38:47,335][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:38:47,839][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:38:48,343][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:38:48,843][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:38:49,346][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:38:49,844][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:38:50,343][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:38:50,846][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:38:51,341][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:38:51,839][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:38:52,332][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:38:52,824][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:38:53,319][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:38:53,811][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:38:54,311][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:38:54,808][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:38:55,305][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:38:55,801][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:38:56,297][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:38:56,794][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:38:57,291][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:38:57,789][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:38:58,290][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:38:58,789][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:38:59,292][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:38:59,794][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:39:00,297][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:39:00,800][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:39:01,302][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:39:01,799][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:39:02,298][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:39:02,795][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:39:03,290][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:39:03,790][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:39:04,291][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:39:04,791][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:39:05,292][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:39:05,789][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:39:06,291][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:39:06,792][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:39:07,288][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:39:07,781][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:39:08,278][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:39:08,773][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:39:09,268][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:39:09,763][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:39:10,260][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:39:10,757][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:39:11,248][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:39:11,741][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:39:12,237][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:39:12,733][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:39:13,230][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:39:13,726][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:39:14,222][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:39:14,715][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:39:15,212][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:39:15,709][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:39:16,206][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:39:16,707][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:39:17,203][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:39:17,699][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:39:18,195][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:39:18,693][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9823 tokens. +[2026-03-26 07:39:19,527][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.02%, Current % of VRAM taken: 60.44%, Block Peak % of device VRAM: 62.13%, ΔTime: 00:00:32 +[2026-03-26 07:39:20,276][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:39:20,279][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:39:20,280][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:39:21,014][__main__][INFO] - Iteration 629 took 50s (29.16% Gen, 69.39% Train). Generation: 14s, Training: 35s. Estimated remaining time: 32h 30m 52s. Estimated total time: 42h 9m 24s. Time estimates for 10 more iterations: 8m 25s, 100 more iterations: 1h 24m 18s, 500 more iterations: 7h 1m 34s. +[2026-03-26 07:39:21,016][__main__][INFO] - Starting iteration 629. +[2026-03-26 07:39:21,419][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 62 and human policies 1. +[2026-03-26 07:39:21,419][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:39:38,507][__main__][INFO] - Number of regex retries in iteration 629: 0 +[2026-03-26 07:39:38,508][__main__][INFO] - agents played in iteration 629 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:39:39,276][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:39:39,299][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:39:39,323][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:39:39,347][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:39:39,347][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:39:39,348][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:39:40,170][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:39:40,622][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:39:41,121][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:39:41,618][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:39:42,113][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:39:42,608][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:39:43,103][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:39:43,605][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:39:44,100][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:39:44,602][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:39:45,100][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:39:45,596][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:39:46,093][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:39:46,591][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:39:47,088][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:39:47,584][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:39:48,081][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:39:48,583][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:39:49,082][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:39:49,584][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:39:50,085][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:39:50,586][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:39:51,086][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:39:51,583][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:39:52,088][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:39:52,584][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:39:53,081][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:39:53,576][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:39:54,074][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:39:54,577][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:39:55,071][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:39:55,566][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:39:56,063][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:39:56,563][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:39:57,063][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:39:57,564][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:39:58,065][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:39:58,564][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:39:59,064][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:39:59,564][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:40:00,082][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:40:00,579][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:40:01,073][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:40:01,571][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:40:02,068][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:40:02,563][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:40:03,057][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:40:03,554][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:40:04,048][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:40:04,545][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:40:05,041][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:40:05,538][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:40:06,034][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:40:06,532][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:40:07,030][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:40:07,525][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:40:08,021][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:40:08,518][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:40:09,015][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:40:09,513][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:40:10,010][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:40:10,502][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:40:10,997][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:40:11,494][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:40:11,991][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9832 tokens. +[2026-03-26 07:40:12,831][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.06%, ΔTime: 00:00:32 +[2026-03-26 07:40:13,551][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:40:13,553][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:40:13,555][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:40:14,567][__main__][INFO] - Iteration 630 took 53s (32.15% Gen, 65.94% Train). Generation: 17s, Training: 35s. Estimated remaining time: 34h 38m 1s. Estimated total time: 44h 17m 27s. Time estimates for 10 more iterations: 8m 51s, 100 more iterations: 1h 28m 34s, 500 more iterations: 7h 22m 54s. +[2026-03-26 07:40:14,569][__main__][INFO] - Starting iteration 630. +[2026-03-26 07:40:14,971][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 62 and human policies 1. +[2026-03-26 07:40:14,972][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:40:28,881][__main__][INFO] - Number of regex retries in iteration 630: 0 +[2026-03-26 07:40:28,882][__main__][INFO] - agents played in iteration 630 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:40:29,651][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:40:29,675][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:40:29,699][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:40:29,723][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:40:29,724][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:40:29,724][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:40:30,532][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:40:30,984][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:40:31,482][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:40:31,979][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:40:32,476][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:40:32,975][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:40:33,475][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:40:33,972][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:40:34,468][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:40:34,970][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:40:35,466][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:40:35,962][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:40:36,458][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:40:36,954][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:40:37,452][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:40:37,949][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:40:38,449][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:40:38,946][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:40:39,443][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:40:39,942][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:40:40,442][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:40:40,940][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:40:41,437][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:40:41,936][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:40:42,437][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:40:42,937][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:40:43,436][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:40:43,939][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:40:44,439][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:40:44,937][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:40:45,434][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:40:45,931][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:40:46,429][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:40:46,927][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:40:47,425][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:40:47,919][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:40:48,412][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:40:48,917][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:40:49,415][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:40:49,913][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:40:50,414][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:40:50,912][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:40:51,406][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:40:51,900][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:40:52,394][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:40:52,888][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:40:53,384][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:40:53,878][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:40:54,372][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:40:54,868][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:40:55,363][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:40:55,860][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:40:56,355][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:40:56,850][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:40:57,348][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:40:57,844][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:40:58,340][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:40:58,836][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:40:59,333][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:40:59,829][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:41:00,328][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:41:00,823][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:41:01,319][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:41:01,816][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:41:02,313][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9829 tokens. +[2026-03-26 07:41:03,151][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.99%, Current % of VRAM taken: 60.42%, Block Peak % of device VRAM: 62.00%, ΔTime: 00:00:32 +[2026-03-26 07:41:03,909][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:41:03,912][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:41:03,913][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:41:05,204][__main__][INFO] - Iteration 631 took 50s (27.69% Gen, 69.74% Train). Generation: 13s, Training: 35s. Estimated remaining time: 32h 11m 22s. Estimated total time: 41h 51m 38s. Time estimates for 10 more iterations: 8m 22s, 100 more iterations: 1h 23m 43s, 500 more iterations: 6h 58m 36s. +[2026-03-26 07:41:05,206][__main__][INFO] - Starting iteration 631. +[2026-03-26 07:41:05,607][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 63 and human policies 1. +[2026-03-26 07:41:05,608][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:41:21,899][__main__][INFO] - Number of regex retries in iteration 631: 0 +[2026-03-26 07:41:21,900][__main__][INFO] - agents played in iteration 631 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:41:22,668][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:41:22,692][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:41:22,716][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:41:22,742][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:41:22,743][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:41:22,744][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:41:23,565][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:41:24,020][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:41:24,521][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:41:25,019][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:41:25,514][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:41:26,013][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:41:26,511][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:41:27,009][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:41:27,505][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:41:28,008][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:41:28,504][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:41:29,001][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:41:29,498][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:41:29,995][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:41:30,492][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:41:30,984][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:41:31,475][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:41:31,974][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:41:32,473][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:41:32,970][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:41:33,469][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:41:33,966][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:41:34,467][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:41:34,963][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:41:35,463][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:41:35,960][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:41:36,458][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:41:36,955][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:41:37,458][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:41:37,961][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:41:38,461][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:41:38,953][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:41:39,449][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:41:39,948][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:41:40,445][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:41:40,942][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:41:41,441][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:41:41,936][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:41:42,434][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:41:42,933][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:41:43,432][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:41:43,952][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:41:44,451][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:41:44,951][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:41:45,449][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:41:45,949][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:41:46,447][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:41:46,944][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:41:47,442][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:41:47,939][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:41:48,438][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:41:48,937][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:41:49,435][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:41:49,936][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:41:50,440][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:41:50,938][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:41:51,436][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:41:51,935][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:41:52,433][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:41:52,929][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:41:53,427][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:41:53,926][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:41:54,429][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:41:54,924][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:41:55,424][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9878 tokens. +[2026-03-26 07:41:56,268][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.09%, ΔTime: 00:00:32 +[2026-03-26 07:41:57,013][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:41:57,015][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:41:57,016][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:41:57,785][__main__][INFO] - Iteration 632 took 52s (31.22% Gen, 67.30% Train). Generation: 16s, Training: 35s. Estimated remaining time: 33h 47m 47s. Estimated total time: 43h 28m 56s. Time estimates for 10 more iterations: 8m 41s, 100 more iterations: 1h 26m 57s, 500 more iterations: 7h 14m 49s. +[2026-03-26 07:41:57,788][__main__][INFO] - Starting iteration 632. +[2026-03-26 07:41:58,194][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 63 and human policies 1. +[2026-03-26 07:41:58,195][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:42:13,683][__main__][INFO] - Number of regex retries in iteration 632: 0 +[2026-03-26 07:42:13,684][__main__][INFO] - agents played in iteration 632 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:42:14,451][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:42:14,474][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:42:14,497][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:42:14,520][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:42:14,521][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:42:14,521][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:42:15,333][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:42:15,782][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:42:16,285][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:42:16,782][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:42:17,274][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:42:17,768][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:42:18,260][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:42:18,758][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:42:19,256][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:42:19,758][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:42:20,258][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:42:20,752][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:42:21,249][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:42:21,746][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:42:22,244][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:42:22,743][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:42:23,240][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:42:23,739][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:42:24,236][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:42:24,734][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:42:25,233][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:42:25,733][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:42:26,233][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:42:26,731][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:42:27,231][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:42:27,769][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:42:28,275][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:42:28,777][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:42:29,277][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:42:29,781][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:42:30,281][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:42:30,781][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:42:31,280][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:42:31,774][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:42:32,271][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:42:32,771][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:42:33,270][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:42:33,765][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:42:34,262][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:42:34,757][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:42:35,256][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:42:35,754][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:42:36,253][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:42:36,751][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:42:37,249][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:42:37,748][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:42:38,245][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:42:38,742][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:42:39,243][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:42:39,741][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:42:40,239][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:42:40,737][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:42:41,236][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:42:41,734][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:42:42,232][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:42:42,730][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:42:43,228][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:42:43,725][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:42:44,220][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:42:44,713][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:42:45,211][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:42:45,710][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:42:46,204][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:42:46,698][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:42:47,197][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9792 tokens. +[2026-03-26 07:42:48,043][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.11%, ΔTime: 00:00:32 +[2026-03-26 07:42:48,795][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:42:48,797][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:42:48,799][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:42:49,788][__main__][INFO] - Iteration 633 took 51s (30.02% Gen, 68.06% Train). Generation: 15s, Training: 35s. Estimated remaining time: 33h 17m 40s. Estimated total time: 42h 59m 42s. Time estimates for 10 more iterations: 8m 35s, 100 more iterations: 1h 25m 59s, 500 more iterations: 7h 9m 57s. +[2026-03-26 07:42:49,790][__main__][INFO] - Starting iteration 633. +[2026-03-26 07:42:50,192][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 63 and human policies 1. +[2026-03-26 07:42:50,193][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:43:29,675][__main__][INFO] - Number of regex retries in iteration 633: 0 +[2026-03-26 07:43:29,676][__main__][INFO] - agents played in iteration 633 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:43:30,449][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:43:30,472][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:43:30,496][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:43:30,519][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:43:30,520][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:43:30,520][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:43:31,327][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:43:31,779][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:43:32,283][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:43:32,785][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:43:33,282][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:43:33,779][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:43:34,277][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:43:34,780][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:43:35,279][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:43:35,777][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:43:36,274][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:43:36,772][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:43:37,271][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:43:37,769][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:43:38,267][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:43:38,765][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:43:39,263][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:43:39,760][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:43:40,260][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:43:40,756][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:43:41,252][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:43:41,746][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:43:42,241][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:43:42,737][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:43:43,234][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:43:43,732][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:43:44,230][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:43:44,728][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:43:45,226][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:43:45,727][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:43:46,226][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:43:46,722][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:43:47,219][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:43:47,721][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:43:48,230][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:43:48,727][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:43:49,223][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:43:49,721][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:43:50,220][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:43:50,721][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:43:51,225][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:43:51,726][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:43:52,225][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:43:52,723][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:43:53,223][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:43:53,720][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:43:54,216][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:43:54,713][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:43:55,212][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:43:55,711][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:43:56,211][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:43:56,713][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:43:57,215][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:43:57,716][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:43:58,218][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:43:58,718][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:43:59,220][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:43:59,718][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:44:00,216][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:44:00,712][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:44:01,211][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:44:01,709][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:44:02,204][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:44:02,704][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:44:03,200][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9863 tokens. +[2026-03-26 07:44:04,042][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.01%, Current % of VRAM taken: 60.44%, Block Peak % of device VRAM: 61.90%, ΔTime: 00:00:32 +[2026-03-26 07:44:04,794][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:44:04,797][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:44:04,799][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:44:05,570][__main__][INFO] - Iteration 634 took 1m 15s (52.38% Gen, 46.60% Train). Generation: 39s, Training: 35s. Estimated remaining time: 53h 5m 38s. Estimated total time: 62h 48m 55s. Time estimates for 10 more iterations: 12m 33s, 100 more iterations: 2h 5m 37s, 500 more iterations: 10h 28m 9s. +[2026-03-26 07:44:05,572][__main__][INFO] - Starting iteration 634. +[2026-03-26 07:44:05,972][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 63 and human policies 1. +[2026-03-26 07:44:05,973][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:44:45,918][__main__][INFO] - Number of regex retries in iteration 634: 0 +[2026-03-26 07:44:45,919][__main__][INFO] - agents played in iteration 634 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:44:46,701][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:44:46,724][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:44:46,748][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:44:46,771][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:44:46,772][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:44:46,773][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:44:47,623][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:44:48,081][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:44:48,590][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:44:49,094][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:44:49,593][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:44:50,092][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:44:50,593][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:44:51,090][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:44:51,591][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:44:52,091][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:44:52,592][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:44:53,089][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:44:53,587][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:44:54,084][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:44:54,583][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:44:55,081][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:44:55,584][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:44:56,083][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:44:56,581][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:44:57,077][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:44:57,596][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:44:58,100][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:44:58,598][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:44:59,097][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:44:59,598][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:45:00,096][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:45:00,595][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:45:01,090][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:45:01,587][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:45:02,085][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:45:02,579][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:45:03,073][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:45:03,570][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:45:04,077][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:45:04,575][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:45:05,074][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:45:05,573][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:45:06,071][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:45:06,569][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:45:07,067][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:45:07,607][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:45:08,107][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:45:08,604][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:45:09,101][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:45:09,597][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:45:10,094][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:45:10,591][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:45:11,090][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:45:11,587][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:45:12,087][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:45:12,586][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:45:13,086][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:45:13,584][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:45:14,083][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:45:14,580][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:45:15,078][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:45:15,577][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:45:16,076][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:45:16,574][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:45:17,076][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:45:17,578][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:45:18,076][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:45:18,577][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:45:19,075][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:45:19,576][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9890 tokens. +[2026-03-26 07:45:20,433][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.00%, ΔTime: 00:00:32 +[2026-03-26 07:45:21,207][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:45:21,209][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:45:21,211][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:45:21,941][__main__][INFO] - Iteration 635 took 1m 15s (52.58% Gen, 46.46% Train). Generation: 39s, Training: 35s. Estimated remaining time: 53h 33m 56s. Estimated total time: 63h 18m 29s. Time estimates for 10 more iterations: 12m 39s, 100 more iterations: 2h 6m 36s, 500 more iterations: 10h 33m 4s. +[2026-03-26 07:45:21,944][__main__][INFO] - Starting iteration 635. +[2026-03-26 07:45:22,343][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 63 and human policies 1. +[2026-03-26 07:45:22,344][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:45:40,907][__main__][INFO] - Number of regex retries in iteration 635: 0 +[2026-03-26 07:45:40,908][__main__][INFO] - agents played in iteration 635 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:45:41,683][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:45:41,706][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:45:41,730][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:45:41,753][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:45:41,754][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:45:41,754][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:45:42,568][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:45:43,022][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:45:43,526][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:45:44,023][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:45:44,518][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:45:45,013][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:45:45,507][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:45:46,003][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:45:46,501][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:45:47,008][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:45:47,513][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:45:48,013][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:45:48,513][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:45:49,013][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:45:49,513][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:45:50,013][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:45:50,514][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:45:51,012][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:45:51,508][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:45:52,005][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:45:52,502][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:45:52,996][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:45:53,494][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:45:53,991][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:45:54,490][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:45:54,988][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:45:55,483][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:45:55,980][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:45:56,477][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:45:56,976][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:45:57,481][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:45:57,978][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:45:58,476][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:45:58,974][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:45:59,472][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:45:59,969][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:46:00,463][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:46:00,961][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:46:01,455][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:46:01,951][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:46:02,444][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:46:02,941][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:46:03,444][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:46:03,940][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:46:04,436][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:46:04,937][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:46:05,435][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:46:05,932][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:46:06,428][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:46:06,948][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:46:07,451][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:46:07,949][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:46:08,449][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:46:08,943][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:46:09,441][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:46:09,935][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:46:10,434][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:46:10,932][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:46:11,431][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:46:11,930][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:46:12,429][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:46:12,929][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:46:13,427][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:46:13,926][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:46:14,423][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9850 tokens. +[2026-03-26 07:46:15,265][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.02%, ΔTime: 00:00:32 +[2026-03-26 07:46:16,026][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:46:16,029][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:46:16,030][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:46:16,783][__main__][INFO] - Iteration 636 took 54s (34.10% Gen, 64.52% Train). Generation: 18s, Training: 35s. Estimated remaining time: 35h 36m 33s. Estimated total time: 45h 22m 1s. Time estimates for 10 more iterations: 9m 4s, 100 more iterations: 1h 30m 44s, 500 more iterations: 7h 33m 40s. +[2026-03-26 07:46:16,785][__main__][INFO] - Starting iteration 636. +[2026-03-26 07:46:17,183][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 63 and human policies 1. +[2026-03-26 07:46:17,183][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:46:41,287][__main__][INFO] - Number of regex retries in iteration 636: 0 +[2026-03-26 07:46:41,289][__main__][INFO] - agents played in iteration 636 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:46:42,057][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:46:42,080][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:46:42,104][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:46:42,127][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:46:42,128][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:46:42,129][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:46:42,941][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:46:43,389][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:46:43,886][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:46:44,387][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:46:44,889][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:46:45,381][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:46:45,876][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:46:46,371][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:46:46,872][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:46:47,380][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:46:47,880][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:46:48,378][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:46:48,877][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:46:49,378][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:46:49,880][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:46:50,382][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:46:50,880][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:46:51,382][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:46:51,876][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:46:52,377][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:46:52,877][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:46:53,377][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:46:53,873][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:46:54,369][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:46:54,863][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:46:55,361][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:46:55,859][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:46:56,359][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:46:56,878][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:46:57,381][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:46:57,879][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:46:58,376][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:46:58,874][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:46:59,369][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:46:59,862][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:47:00,363][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:47:00,855][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:47:01,351][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:47:01,846][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:47:02,344][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:47:02,841][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:47:03,337][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:47:03,831][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:47:04,328][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:47:04,820][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:47:05,316][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:47:05,814][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:47:06,311][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:47:06,809][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:47:07,308][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:47:07,809][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:47:08,308][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:47:08,804][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:47:09,300][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:47:09,799][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:47:10,299][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:47:10,793][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:47:11,290][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:47:11,788][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:47:12,285][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:47:12,781][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:47:13,278][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:47:13,773][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:47:14,268][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:47:14,765][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9811 tokens. +[2026-03-26 07:47:15,594][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.15%, ΔTime: 00:00:32 +[2026-03-26 07:47:16,490][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:47:16,492][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:47:16,494][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:47:17,458][__main__][INFO] - Iteration 637 took 1m 0s (39.99% Gen, 58.41% Train). Generation: 24s, Training: 35s. Estimated remaining time: 40h 27m 18s. Estimated total time: 50h 13m 47s. Time estimates for 10 more iterations: 10m 2s, 100 more iterations: 1h 40m 27s, 500 more iterations: 8h 22m 17s. +[2026-03-26 07:47:17,460][__main__][INFO] - Starting iteration 637. +[2026-03-26 07:47:17,868][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 63 and human policies 1. +[2026-03-26 07:47:17,868][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:47:50,369][__main__][INFO] - Number of regex retries in iteration 637: 0 +[2026-03-26 07:47:50,370][__main__][INFO] - agents played in iteration 637 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:47:51,170][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:47:51,194][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:47:51,218][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:47:51,241][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:47:51,242][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:47:51,242][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:47:52,051][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:47:52,505][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:47:53,036][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:47:53,535][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:47:54,034][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:47:54,532][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:47:55,032][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:47:55,530][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:47:56,032][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:47:56,532][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:47:57,034][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:47:57,535][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:47:58,037][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:47:58,537][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:47:59,037][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:47:59,533][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:48:00,033][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:48:00,530][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:48:01,024][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:48:01,524][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:48:02,022][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:48:02,520][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:48:03,019][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:48:03,518][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:48:04,017][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:48:04,518][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:48:05,018][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:48:05,518][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:48:06,016][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:48:06,516][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:48:07,015][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:48:07,512][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:48:08,012][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:48:08,509][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:48:09,004][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:48:09,499][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:48:09,995][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:48:10,495][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:48:10,992][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:48:11,490][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:48:11,985][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:48:12,480][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:48:12,975][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:48:13,475][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:48:13,972][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:48:14,469][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:48:14,977][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:48:15,477][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:48:15,977][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:48:16,475][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:48:16,976][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:48:17,477][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:48:17,973][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:48:18,471][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:48:18,970][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:48:19,471][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:48:19,977][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:48:20,478][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:48:20,976][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:48:21,477][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:48:21,978][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:48:22,477][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:48:22,976][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:48:23,475][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:48:23,972][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9939 tokens. +[2026-03-26 07:48:24,841][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.05%, Current % of VRAM taken: 60.48%, Block Peak % of device VRAM: 62.06%, ΔTime: 00:00:32 +[2026-03-26 07:48:25,632][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:48:25,636][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:48:25,639][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:48:26,385][__main__][INFO] - Iteration 638 took 1m 8s (47.44% Gen, 51.47% Train). Generation: 32s, Training: 35s. Estimated remaining time: 47h 18m 15s. Estimated total time: 57h 5m 53s. Time estimates for 10 more iterations: 11m 25s, 100 more iterations: 1h 54m 11s, 500 more iterations: 9h 30m 58s. +[2026-03-26 07:48:26,387][__main__][INFO] - Starting iteration 638. +[2026-03-26 07:48:26,799][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 63 and human policies 1. +[2026-03-26 07:48:26,800][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:48:38,555][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 07:48:45,943][__main__][INFO] - Number of regex retries in iteration 638: 1 +[2026-03-26 07:48:45,944][__main__][INFO] - agents played in iteration 638 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:48:47,673][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:48:47,697][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:48:47,720][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:48:47,743][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:48:47,743][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:48:47,744][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:48:48,564][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:48:49,020][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:48:49,524][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:48:50,025][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:48:50,525][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:48:51,023][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:48:51,527][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:48:52,026][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:48:52,530][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:48:53,028][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:48:53,526][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:48:54,023][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:48:54,519][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:48:55,015][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:48:55,510][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:48:56,009][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:48:56,506][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:48:57,003][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:48:57,501][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:48:57,996][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:48:58,487][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:48:58,979][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:48:59,474][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:48:59,966][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:49:00,458][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:49:00,954][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:49:01,453][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:49:01,951][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:49:02,449][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:49:02,947][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:49:03,445][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:49:03,942][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:49:04,440][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:49:04,936][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:49:05,431][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:49:05,930][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:49:06,430][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:49:06,927][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:49:07,423][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:49:07,919][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:49:08,415][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:49:08,914][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:49:09,412][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:49:09,911][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:49:10,410][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:49:10,913][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:49:11,413][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:49:11,923][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:49:12,426][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:49:12,928][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:49:13,431][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:49:13,933][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:49:14,434][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:49:14,935][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:49:15,429][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:49:15,926][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:49:16,425][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:49:16,925][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:49:17,428][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:49:17,930][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:49:18,429][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:49:18,925][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:49:19,425][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:49:19,922][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:49:20,421][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9791 tokens. +[2026-03-26 07:49:21,299][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.04%, ΔTime: 00:00:32 +[2026-03-26 07:49:22,081][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:49:22,083][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:49:22,084][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:49:22,785][__main__][INFO] - Iteration 639 took 55s (34.19% Gen, 64.55% Train). Generation: 19s, Training: 36s. Estimated remaining time: 36h 50m 44s. Estimated total time: 46h 39m 18s. Time estimates for 10 more iterations: 9m 19s, 100 more iterations: 1h 33m 18s, 500 more iterations: 7h 46m 33s. +[2026-03-26 07:49:22,787][__main__][INFO] - Starting iteration 639. +[2026-03-26 07:49:23,209][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 63 and human policies 1. +[2026-03-26 07:49:23,210][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:50:08,414][__main__][INFO] - Number of regex retries in iteration 639: 0 +[2026-03-26 07:50:08,415][__main__][INFO] - agents played in iteration 639 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:50:09,177][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:50:09,200][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:50:09,224][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:50:09,247][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:50:09,248][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:50:09,248][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:50:10,068][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:50:10,520][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:50:11,027][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:50:11,526][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:50:12,025][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:50:12,523][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:50:13,021][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:50:13,516][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:50:14,015][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:50:14,512][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:50:15,009][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:50:15,505][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:50:15,998][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:50:16,490][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:50:16,981][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:50:17,483][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:50:17,981][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:50:18,484][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:50:18,984][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:50:19,484][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:50:19,983][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:50:20,487][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:50:20,991][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:50:21,492][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:50:21,988][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:50:22,483][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:50:22,974][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:50:23,474][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:50:23,975][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:50:24,480][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:50:24,978][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:50:25,476][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:50:25,975][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:50:26,474][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:50:26,967][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:50:27,457][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:50:27,948][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:50:28,444][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:50:28,942][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:50:29,434][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:50:29,925][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:50:30,421][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:50:30,921][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:50:31,419][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:50:31,920][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:50:32,421][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:50:32,919][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:50:33,422][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:50:33,922][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:50:34,418][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:50:34,919][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:50:35,422][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:50:35,915][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:50:36,408][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:50:36,903][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:50:37,395][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:50:37,892][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:50:38,391][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:50:38,889][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:50:39,388][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:50:39,886][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:50:40,384][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:50:40,881][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:50:41,378][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:50:41,880][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9742 tokens. +[2026-03-26 07:50:42,738][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.04%, Current % of VRAM taken: 60.47%, Block Peak % of device VRAM: 62.03%, ΔTime: 00:00:32 +[2026-03-26 07:50:43,468][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:50:43,471][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:50:43,472][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:50:44,199][__main__][INFO] - Iteration 640 took 1m 20s (55.82% Gen, 43.29% Train). Generation: 45s, Training: 35s. Estimated remaining time: 57h 39m 36s. Estimated total time: 67h 29m 32s. Time estimates for 10 more iterations: 13m 29s, 100 more iterations: 2h 14m 59s, 500 more iterations: 11h 14m 55s. +[2026-03-26 07:50:44,201][__main__][INFO] - Starting iteration 640. +[2026-03-26 07:50:44,604][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 63 and human policies 1. +[2026-03-26 07:50:44,605][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:50:59,665][__main__][INFO] - Number of regex retries in iteration 640: 0 +[2026-03-26 07:50:59,666][__main__][INFO] - agents played in iteration 640 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:51:00,425][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:51:00,448][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:51:00,471][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:51:00,494][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:51:00,495][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:51:00,495][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:51:01,328][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:51:01,785][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:51:02,290][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:51:02,789][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:51:03,290][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:51:03,789][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:51:04,290][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:51:04,790][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:51:05,292][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:51:05,796][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:51:06,296][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:51:06,795][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:51:07,294][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:51:07,798][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:51:08,296][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:51:08,798][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:51:09,297][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:51:09,795][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:51:10,292][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:51:10,789][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:51:11,286][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:51:11,788][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:51:12,286][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:51:12,786][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:51:13,284][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:51:13,777][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:51:14,275][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:51:14,797][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:51:15,297][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:51:15,794][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:51:16,291][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:51:16,789][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:51:17,286][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:51:17,782][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:51:18,279][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:51:18,777][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:51:19,276][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:51:19,775][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:51:20,276][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:51:20,773][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:51:21,269][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:51:21,760][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:51:22,251][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:51:22,742][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:51:23,238][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:51:23,730][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:51:24,221][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:51:24,716][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:51:25,208][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:51:25,701][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:51:26,200][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:51:26,695][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:51:27,191][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:51:27,688][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:51:28,184][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:51:28,680][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:51:29,174][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:51:29,672][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:51:30,174][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:51:30,671][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:51:31,169][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:51:31,666][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:51:32,164][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:51:32,662][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:51:33,160][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9813 tokens. +[2026-03-26 07:51:34,008][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.15%, Current % of VRAM taken: 60.58%, Block Peak % of device VRAM: 62.12%, ΔTime: 00:00:32 +[2026-03-26 07:51:34,909][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:51:34,911][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:51:34,914][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:51:36,332][__main__][INFO] - Iteration 641 took 51s (29.11% Gen, 68.14% Train). Generation: 15s, Training: 35s. Estimated remaining time: 33h 15m 38s. Estimated total time: 43h 6m 25s. Time estimates for 10 more iterations: 8m 37s, 100 more iterations: 1h 26m 12s, 500 more iterations: 7h 11m 4s. +[2026-03-26 07:51:36,335][__main__][INFO] - Starting iteration 641. +[2026-03-26 07:51:36,733][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 64 and human policies 1. +[2026-03-26 07:51:36,734][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:51:51,629][__main__][INFO] - Number of regex retries in iteration 641: 0 +[2026-03-26 07:51:51,630][__main__][INFO] - agents played in iteration 641 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:51:52,382][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:51:52,406][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:51:52,429][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:51:52,452][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:51:52,453][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:51:52,453][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:51:53,267][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:51:53,718][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:51:54,224][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:51:54,727][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:51:55,230][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:51:55,728][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:51:56,230][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:51:56,728][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:51:57,232][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:51:57,733][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:51:58,238][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:51:58,740][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:51:59,241][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:51:59,738][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:52:00,234][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:52:00,762][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:52:01,265][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:52:01,768][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:52:02,269][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:52:02,765][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:52:03,260][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:52:03,763][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:52:04,261][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:52:04,758][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:52:05,256][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:52:05,754][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:52:06,256][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:52:06,755][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:52:07,257][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:52:07,756][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:52:08,256][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:52:08,755][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:52:09,252][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:52:09,750][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:52:10,247][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:52:10,746][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:52:11,241][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:52:11,737][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:52:12,236][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:52:12,732][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:52:13,229][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:52:13,727][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:52:14,223][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:52:14,720][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:52:15,217][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:52:15,717][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:52:16,214][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:52:16,711][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:52:17,207][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:52:17,699][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:52:18,191][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:52:18,688][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:52:19,179][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:52:19,676][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:52:20,174][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:52:20,666][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:52:21,156][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:52:21,652][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:52:22,151][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:52:22,648][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:52:23,142][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:52:23,639][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:52:24,137][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:52:24,634][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:52:25,134][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9825 tokens. +[2026-03-26 07:52:25,983][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.21%, Current % of VRAM taken: 60.64%, Block Peak % of device VRAM: 62.13%, ΔTime: 00:00:32 +[2026-03-26 07:52:26,893][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:52:26,895][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:52:26,896][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:52:27,574][__main__][INFO] - Iteration 642 took 50s (29.30% Gen, 69.37% Train). Generation: 14s, Training: 35s. Estimated remaining time: 32h 30m 26s. Estimated total time: 42h 22m 5s. Time estimates for 10 more iterations: 8m 28s, 100 more iterations: 1h 24m 44s, 500 more iterations: 7h 3m 40s. +[2026-03-26 07:52:27,577][__main__][INFO] - Starting iteration 642. +[2026-03-26 07:52:27,989][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 64 and human policies 1. +[2026-03-26 07:52:27,990][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:52:55,373][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 07:52:55,375][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 07:53:06,202][__main__][INFO] - Number of regex retries in iteration 642: 2 +[2026-03-26 07:53:06,203][__main__][INFO] - agents played in iteration 642 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:53:07,013][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:53:07,037][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:53:07,061][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:53:07,085][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:53:07,086][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:53:07,086][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:53:07,904][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:53:08,356][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:53:08,857][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:53:09,356][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:53:09,854][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:53:10,349][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:53:10,845][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:53:11,346][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:53:11,843][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:53:12,341][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:53:12,865][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:53:13,367][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:53:13,865][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:53:14,363][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:53:14,861][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:53:15,359][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:53:15,860][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:53:16,357][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:53:16,860][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:53:17,362][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:53:17,858][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:53:18,359][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:53:18,858][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:53:19,355][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:53:19,854][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:53:20,348][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:53:20,849][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:53:21,343][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:53:21,837][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:53:22,331][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:53:22,827][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:53:23,322][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:53:23,816][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:53:24,314][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:53:24,812][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:53:25,312][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:53:25,810][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:53:26,312][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:53:26,811][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:53:27,311][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:53:27,811][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:53:28,313][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:53:28,811][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:53:29,310][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:53:29,808][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:53:30,306][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:53:30,804][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:53:31,302][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:53:31,800][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:53:32,296][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:53:32,792][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:53:33,289][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:53:33,787][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:53:34,285][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:53:34,783][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:53:35,279][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:53:35,776][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:53:36,270][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:53:36,764][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:53:37,258][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:53:37,752][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:53:38,250][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:53:38,744][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:53:39,240][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:53:39,735][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9847 tokens. +[2026-03-26 07:53:40,581][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 12.96%, Current % of VRAM taken: 59.38%, Block Peak % of device VRAM: 62.15%, ΔTime: 00:00:32 +[2026-03-26 07:53:41,354][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:53:41,357][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:53:41,358][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:53:42,072][__main__][INFO] - Iteration 643 took 1m 14s (51.58% Gen, 47.45% Train). Generation: 38s, Training: 35s. Estimated remaining time: 51h 51m 17s. Estimated total time: 61h 44m 10s. Time estimates for 10 more iterations: 12m 20s, 100 more iterations: 2h 3m 28s, 500 more iterations: 10h 17m 21s. +[2026-03-26 07:53:42,074][__main__][INFO] - Starting iteration 643. +[2026-03-26 07:53:42,519][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 64 and human policies 1. +[2026-03-26 07:53:42,520][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:53:56,888][__main__][INFO] - Number of regex retries in iteration 643: 0 +[2026-03-26 07:53:56,889][__main__][INFO] - agents played in iteration 643 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:53:57,648][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:53:57,673][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:53:57,697][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:53:57,724][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:53:57,725][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:53:57,726][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:53:58,532][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:53:58,983][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:53:59,481][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:53:59,975][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:54:00,470][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:54:00,964][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:54:01,459][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:54:01,953][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:54:02,450][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:54:02,950][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:54:03,446][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:54:03,941][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:54:04,438][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:54:04,935][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:54:05,430][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:54:05,925][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:54:06,422][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:54:06,924][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:54:07,426][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:54:07,925][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:54:08,429][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:54:08,928][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:54:09,447][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:54:09,947][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:54:10,445][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:54:10,940][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:54:11,435][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:54:11,930][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:54:12,425][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:54:12,919][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:54:13,413][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:54:13,908][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:54:14,404][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:54:14,897][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:54:15,392][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:54:15,885][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:54:16,378][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:54:16,872][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:54:17,370][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:54:17,864][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:54:18,358][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:54:18,859][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:54:19,361][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:54:19,855][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:54:20,350][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:54:20,845][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:54:21,341][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:54:21,837][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:54:22,331][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:54:22,830][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:54:23,325][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:54:23,824][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:54:24,322][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:54:24,817][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:54:25,312][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:54:25,807][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:54:26,300][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:54:26,798][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:54:27,296][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:54:27,793][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:54:28,290][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:54:28,786][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:54:29,282][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:54:29,779][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:54:30,275][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9721 tokens. +[2026-03-26 07:54:31,126][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.01%, Current % of VRAM taken: 60.43%, Block Peak % of device VRAM: 62.01%, ΔTime: 00:00:32 +[2026-03-26 07:54:31,927][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:54:31,929][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:54:31,930][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:54:32,660][__main__][INFO] - Iteration 644 took 50s (28.66% Gen, 69.89% Train). Generation: 14s, Training: 35s. Estimated remaining time: 31h 53m 19s. Estimated total time: 41h 47m 3s. Time estimates for 10 more iterations: 8m 21s, 100 more iterations: 1h 23m 34s, 500 more iterations: 6h 57m 50s. +[2026-03-26 07:54:32,662][__main__][INFO] - Starting iteration 644. +[2026-03-26 07:54:33,072][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 64 and human policies 1. +[2026-03-26 07:54:33,073][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:54:49,854][__main__][INFO] - Number of regex retries in iteration 644: 0 +[2026-03-26 07:54:49,855][__main__][INFO] - agents played in iteration 644 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:54:50,621][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:54:50,645][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:54:50,668][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:54:50,692][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:54:50,693][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:54:50,693][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:54:51,506][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:54:51,960][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:54:52,463][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:54:52,959][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:54:53,455][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:54:53,949][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:54:54,446][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:54:54,941][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:54:55,438][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:54:55,944][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:54:56,446][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:54:56,944][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:54:57,443][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:54:57,943][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:54:58,447][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:54:58,944][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:54:59,446][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:54:59,940][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:55:00,439][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:55:00,938][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:55:01,436][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:55:01,934][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:55:02,434][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:55:02,932][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:55:03,430][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:55:03,927][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:55:04,428][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:55:04,926][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:55:05,424][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:55:05,921][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:55:06,420][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:55:06,917][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:55:07,414][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:55:07,914][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:55:08,411][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:55:08,908][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:55:09,405][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:55:09,901][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:55:10,399][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:55:10,891][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:55:11,386][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:55:11,883][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:55:12,380][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:55:12,878][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:55:13,376][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:55:13,874][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:55:14,372][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:55:14,870][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:55:15,365][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:55:15,861][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:55:16,358][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:55:16,856][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:55:17,354][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:55:17,856][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:55:18,356][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:55:18,854][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:55:19,352][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:55:19,847][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:55:20,345][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:55:20,842][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:55:21,339][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:55:21,836][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:55:22,332][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:55:22,831][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:55:23,328][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9892 tokens. +[2026-03-26 07:55:24,177][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.98%, Current % of VRAM taken: 60.41%, Block Peak % of device VRAM: 61.95%, ΔTime: 00:00:32 +[2026-03-26 07:55:24,911][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:55:24,914][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:55:24,915][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:55:25,651][__main__][INFO] - Iteration 645 took 52s (31.92% Gen, 66.68% Train). Generation: 16s, Training: 35s. Estimated remaining time: 33h 54m 22s. Estimated total time: 43h 48m 59s. Time estimates for 10 more iterations: 8m 45s, 100 more iterations: 1h 27m 37s, 500 more iterations: 7h 18m 9s. +[2026-03-26 07:55:25,654][__main__][INFO] - Starting iteration 645. +[2026-03-26 07:55:26,070][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 64 and human policies 1. +[2026-03-26 07:55:26,071][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:55:37,604][mllm.models.large_language_model_local][WARNING] - Response Proposal: x hats, 10 books, 10 balls + +Given Bob's per-item values, he values books and balls much more than hats. To maximize my points, I should try to get as many of the items I value highly (balls) as possible. Since Bob values balls very little, I can propose to take all 10 balls. Additionally, I propose to take all 10 books since they are also valued by Bob, and I value them. For hats, since I value them the least and Bob also values them only moderately, I propose to take 0 hats. This strategy aims to secure the items with the highest value asymmetry between us. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 07:55:41,845][__main__][INFO] - Number of regex retries in iteration 645: 1 +[2026-03-26 07:55:41,846][__main__][INFO] - agents played in iteration 645 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:55:42,616][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:55:42,641][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:55:42,666][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:55:42,690][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:55:42,691][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:55:42,691][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:55:43,499][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:55:43,948][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:55:44,447][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:55:44,939][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:55:45,439][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:55:45,942][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:55:46,434][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:55:46,926][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:55:47,418][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:55:47,916][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:55:48,415][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:55:48,919][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:55:49,419][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:55:49,916][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:55:50,416][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:55:50,914][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:55:51,413][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:55:51,912][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:55:52,410][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:55:52,906][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:55:53,402][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:55:53,899][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:55:54,397][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:55:54,898][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:55:55,400][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:55:55,905][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:55:56,406][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:55:56,909][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:55:57,412][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:55:57,915][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:55:58,415][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:55:58,913][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:55:59,412][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:55:59,911][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:56:00,411][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:56:00,913][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:56:01,413][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:56:01,912][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:56:02,405][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:56:02,907][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:56:03,402][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:56:03,899][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:56:04,396][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:56:04,892][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:56:05,389][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:56:05,884][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:56:06,380][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:56:06,875][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:56:07,375][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:56:07,872][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:56:08,368][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:56:08,863][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:56:09,357][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:56:09,852][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:56:10,351][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:56:10,851][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:56:11,346][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:56:11,847][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:56:12,343][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:56:12,836][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:56:13,328][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:56:13,820][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:56:14,320][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:56:14,815][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:56:15,337][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9769 tokens. +[2026-03-26 07:56:16,174][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.25%, Current % of VRAM taken: 60.68%, Block Peak % of device VRAM: 62.13%, ΔTime: 00:00:32 +[2026-03-26 07:56:16,945][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:56:16,947][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:56:16,949][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:56:17,639][__main__][INFO] - Iteration 646 took 51s (30.59% Gen, 68.07% Train). Generation: 15s, Training: 35s. Estimated remaining time: 33h 3m 2s. Estimated total time: 42h 58m 31s. Time estimates for 10 more iterations: 8m 35s, 100 more iterations: 1h 25m 57s, 500 more iterations: 7h 9m 45s. +[2026-03-26 07:56:17,641][__main__][INFO] - Starting iteration 646. +[2026-03-26 07:56:18,060][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 64 and human policies 1. +[2026-03-26 07:56:18,060][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:56:50,922][__main__][INFO] - Number of regex retries in iteration 646: 0 +[2026-03-26 07:56:50,923][__main__][INFO] - agents played in iteration 646 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:56:51,694][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:56:51,724][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:56:51,751][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:56:51,775][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:56:51,776][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:56:51,776][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:56:52,650][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:56:53,117][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:56:53,619][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:56:54,118][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:56:54,620][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:56:55,121][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:56:55,624][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:56:56,138][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:56:56,642][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:56:57,160][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:56:57,678][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:56:58,182][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:56:58,684][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:56:59,182][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:56:59,680][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:57:00,175][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:57:00,673][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:57:01,172][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:57:01,668][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:57:02,164][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:57:02,660][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:57:03,158][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:57:03,659][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:57:04,153][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:57:04,649][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:57:05,144][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:57:05,642][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:57:06,140][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:57:06,639][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:57:07,135][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:57:07,634][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:57:08,138][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:57:08,635][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:57:09,133][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:57:09,632][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:57:10,131][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:57:10,630][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:57:11,128][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:57:11,629][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:57:12,126][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:57:12,623][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:57:13,119][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:57:13,622][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:57:14,123][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:57:14,628][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:57:15,125][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:57:15,626][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:57:16,122][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:57:16,617][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:57:17,114][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:57:17,615][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:57:18,113][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:57:18,611][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:57:19,108][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:57:19,609][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:57:20,108][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:57:20,610][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:57:21,105][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:57:21,598][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:57:22,092][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:57:22,586][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:57:23,080][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:57:23,575][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:57:24,069][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:57:24,567][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9801 tokens. +[2026-03-26 07:57:25,390][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.19%, ΔTime: 00:00:32 +[2026-03-26 07:57:26,147][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:57:26,149][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:57:26,151][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:57:26,870][__main__][INFO] - Iteration 647 took 1m 8s (47.76% Gen, 51.20% Train). Generation: 32s, Training: 35s. Estimated remaining time: 47h 23m 54s. Estimated total time: 57h 20m 32s. Time estimates for 10 more iterations: 11m 28s, 100 more iterations: 1h 54m 41s, 500 more iterations: 9h 33m 25s. +[2026-03-26 07:57:26,872][__main__][INFO] - Starting iteration 647. +[2026-03-26 07:57:27,283][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 64 and human policies 1. +[2026-03-26 07:57:27,284][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:57:43,334][__main__][INFO] - Number of regex retries in iteration 647: 0 +[2026-03-26 07:57:43,335][__main__][INFO] - agents played in iteration 647 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:57:44,104][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:57:44,128][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:57:44,152][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:57:44,175][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:57:44,175][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:57:44,176][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:57:44,989][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:57:45,445][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:57:45,944][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:57:46,449][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:57:46,959][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:57:47,464][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:57:47,969][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:57:48,472][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:57:48,971][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:57:49,468][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:57:49,964][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:57:50,463][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:57:50,957][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:57:51,454][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:57:51,952][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:57:52,450][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:57:52,944][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:57:53,444][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:57:53,941][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:57:54,438][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:57:54,938][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:57:55,436][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:57:55,941][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:57:56,439][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:57:56,940][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:57:57,434][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:57:57,931][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:57:58,429][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:57:58,927][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:57:59,426][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:57:59,923][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:58:00,422][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:58:00,921][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:58:01,421][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:58:01,920][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:58:02,416][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:58:02,913][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:58:03,410][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:58:03,905][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:58:04,401][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:58:04,897][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:58:05,393][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:58:05,891][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:58:06,389][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:58:06,886][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:58:07,383][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:58:07,879][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:58:08,376][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:58:08,874][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:58:09,369][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:58:09,863][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:58:10,360][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:58:10,856][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:58:11,350][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:58:11,846][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:58:12,343][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:58:12,836][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:58:13,333][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:58:13,832][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:58:14,331][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:58:14,828][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:58:15,324][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:58:15,822][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:58:16,318][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:58:16,815][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9837 tokens. +[2026-03-26 07:58:17,657][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.05%, Current % of VRAM taken: 60.48%, Block Peak % of device VRAM: 62.49%, ΔTime: 00:00:32 +[2026-03-26 07:58:18,433][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:58:18,436][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:58:18,437][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:58:19,299][__main__][INFO] - Iteration 648 took 52s (30.86% Gen, 67.48% Train). Generation: 16s, Training: 35s. Estimated remaining time: 33h 23m 19s. Estimated total time: 43h 20m 50s. Time estimates for 10 more iterations: 8m 40s, 100 more iterations: 1h 26m 41s, 500 more iterations: 7h 13m 28s. +[2026-03-26 07:58:19,301][__main__][INFO] - Starting iteration 648. +[2026-03-26 07:58:19,705][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 64 and human policies 1. +[2026-03-26 07:58:19,706][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:58:39,839][__main__][INFO] - Number of regex retries in iteration 648: 0 +[2026-03-26 07:58:39,840][__main__][INFO] - agents played in iteration 648 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:58:40,592][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:58:40,616][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:58:40,639][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:58:40,662][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:58:40,663][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:58:40,663][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:58:41,484][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:58:41,936][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:58:42,438][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:58:42,937][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:58:43,433][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:58:43,929][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:58:44,424][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:58:44,918][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:58:45,413][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:58:45,911][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:58:46,408][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:58:46,905][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:58:47,404][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:58:47,912][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:58:48,416][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:58:48,917][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:58:49,413][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:58:49,912][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:58:50,416][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:58:50,917][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:58:51,417][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:58:51,913][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:58:52,421][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:58:52,925][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:58:53,427][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:58:53,928][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:58:54,432][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 07:58:54,929][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 07:58:55,426][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 07:58:55,925][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 07:58:56,422][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 07:58:56,918][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 07:58:57,417][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 07:58:57,915][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 07:58:58,414][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 07:58:58,912][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 07:58:59,413][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 07:58:59,914][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 07:59:00,413][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 07:59:00,910][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 07:59:01,407][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 07:59:01,905][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 07:59:02,404][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 07:59:02,903][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 07:59:03,404][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 07:59:03,908][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 07:59:04,405][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 07:59:04,904][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 07:59:05,404][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 07:59:05,906][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 07:59:06,403][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 07:59:06,900][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 07:59:07,399][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 07:59:07,894][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 07:59:08,390][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 07:59:08,883][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 07:59:09,378][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 07:59:09,873][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 07:59:10,367][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 07:59:10,861][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 07:59:11,356][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 07:59:11,852][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 07:59:12,348][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 07:59:12,842][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 07:59:13,338][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9870 tokens. +[2026-03-26 07:59:14,188][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.97%, Current % of VRAM taken: 60.40%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 07:59:14,934][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 07:59:14,936][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 07:59:14,937][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 07:59:15,784][__main__][INFO] - Iteration 649 took 56s (35.90% Gen, 62.59% Train). Generation: 20s, Training: 35s. Estimated remaining time: 36h 45m 33s. Estimated total time: 46h 44m 0s. Time estimates for 10 more iterations: 9m 20s, 100 more iterations: 1h 33m 28s, 500 more iterations: 7h 47m 20s. +[2026-03-26 07:59:15,786][__main__][INFO] - Starting iteration 649. +[2026-03-26 07:59:16,203][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 64 and human policies 1. +[2026-03-26 07:59:16,204][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 07:59:34,080][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 07:59:45,274][__main__][INFO] - Number of regex retries in iteration 649: 1 +[2026-03-26 07:59:45,275][__main__][INFO] - agents played in iteration 649 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 07:59:46,041][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:59:46,065][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:59:46,088][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:59:46,112][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 07:59:46,112][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 07:59:46,113][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 07:59:46,928][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 07:59:47,383][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 07:59:47,888][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 07:59:48,385][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 07:59:48,883][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 07:59:49,382][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 07:59:49,886][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 07:59:50,387][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 07:59:50,886][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 07:59:51,381][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 07:59:51,879][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 07:59:52,372][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 07:59:52,863][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 07:59:53,366][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 07:59:53,861][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 07:59:54,353][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 07:59:54,848][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 07:59:55,350][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 07:59:55,849][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 07:59:56,348][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 07:59:56,853][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 07:59:57,353][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 07:59:57,858][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 07:59:58,358][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 07:59:58,861][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 07:59:59,353][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 07:59:59,848][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:00:00,345][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:00:00,836][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:00:01,336][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:00:01,833][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:00:02,329][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:00:02,826][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:00:03,321][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:00:03,815][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:00:04,310][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:00:04,804][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:00:05,297][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:00:05,790][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:00:06,284][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:00:06,781][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:00:07,279][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:00:07,775][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:00:08,272][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:00:08,765][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:00:09,262][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:00:09,759][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:00:10,253][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:00:10,756][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:00:11,255][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:00:11,753][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:00:12,250][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:00:12,744][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:00:13,239][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:00:13,736][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:00:14,236][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:00:14,732][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:00:15,229][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:00:15,725][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:00:16,221][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:00:16,717][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:00:17,214][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:00:17,710][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:00:18,206][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:00:18,708][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9802 tokens. +[2026-03-26 08:00:19,554][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.03%, Current % of VRAM taken: 60.46%, Block Peak % of device VRAM: 61.96%, ΔTime: 00:00:32 +[2026-03-26 08:00:20,325][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:00:20,327][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:00:20,329][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:00:21,087][__main__][INFO] - Iteration 650 took 1m 4s (44.81% Gen, 54.02% Train). Generation: 29s, Training: 35s. Estimated remaining time: 44h 4m 40s. Estimated total time: 54h 4m 13s. Time estimates for 10 more iterations: 10m 48s, 100 more iterations: 1h 48m 8s, 500 more iterations: 9h 0m 42s. +[2026-03-26 08:00:21,089][__main__][INFO] - Starting iteration 650. +[2026-03-26 08:00:21,509][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 64 and human policies 1. +[2026-03-26 08:00:21,510][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:00:31,290][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:00:43,084][__main__][INFO] - Number of regex retries in iteration 650: 1 +[2026-03-26 08:00:43,085][__main__][INFO] - agents played in iteration 650 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:00:43,856][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:00:43,879][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:00:43,903][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:00:43,926][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:00:43,927][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:00:43,927][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:00:44,749][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:00:45,203][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:00:45,702][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:00:46,201][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:00:46,698][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:00:47,194][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:00:47,698][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:00:48,202][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:00:48,698][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:00:49,195][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:00:49,691][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:00:50,194][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:00:50,693][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:00:51,191][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:00:51,690][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:00:52,188][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:00:52,691][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:00:53,187][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:00:53,693][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:00:54,190][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:00:54,691][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:00:55,188][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:00:55,689][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:00:56,182][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:00:56,681][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:00:57,179][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:00:57,677][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:00:58,176][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:00:58,673][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:00:59,170][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:00:59,668][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:01:00,165][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:01:00,666][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:01:01,160][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:01:01,653][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:01:02,149][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:01:02,647][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:01:03,149][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:01:03,648][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:01:04,141][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:01:04,637][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:01:05,132][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:01:05,630][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:01:06,128][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:01:06,622][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:01:07,117][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:01:07,611][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:01:08,109][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:01:08,606][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:01:09,102][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:01:09,598][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:01:10,094][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:01:10,591][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:01:11,087][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:01:11,582][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:01:12,076][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:01:12,571][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:01:13,066][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:01:13,559][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:01:14,054][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:01:14,549][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:01:15,044][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:01:15,537][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:01:16,032][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:01:16,529][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9785 tokens. +[2026-03-26 08:01:17,389][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.97%, Current % of VRAM taken: 60.39%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 08:01:18,219][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:01:18,221][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:01:18,224][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:01:19,583][__main__][INFO] - Iteration 651 took 58s (37.15% Gen, 60.51% Train). Generation: 21s, Training: 35s. Estimated remaining time: 38h 23m 13s. Estimated total time: 48h 23m 44s. Time estimates for 10 more iterations: 9m 40s, 100 more iterations: 1h 36m 47s, 500 more iterations: 8h 3m 57s. +[2026-03-26 08:01:19,586][__main__][INFO] - Starting iteration 651. +[2026-03-26 08:01:19,998][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 65 and human policies 1. +[2026-03-26 08:01:19,998][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:01:25,085][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:01:35,310][__main__][INFO] - Number of regex retries in iteration 651: 1 +[2026-03-26 08:01:35,310][__main__][INFO] - agents played in iteration 651 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:01:36,083][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:01:36,106][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:01:36,130][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:01:36,153][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:01:36,154][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:01:36,155][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:01:36,972][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:01:37,428][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:01:37,930][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:01:38,426][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:01:38,929][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:01:39,425][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:01:39,922][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:01:40,423][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:01:40,920][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:01:41,416][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:01:41,913][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:01:42,408][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:01:42,904][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:01:43,398][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:01:43,895][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:01:44,391][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:01:44,894][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:01:45,390][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:01:45,885][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:01:46,382][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:01:46,879][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:01:47,380][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:01:47,875][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:01:48,371][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:01:48,870][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:01:49,367][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:01:49,868][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:01:50,366][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:01:50,863][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:01:51,358][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:01:51,855][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:01:52,352][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:01:52,848][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:01:53,342][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:01:53,847][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:01:54,342][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:01:54,843][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:01:55,352][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:01:55,852][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:01:56,355][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:01:56,856][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:01:57,352][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:01:57,849][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:01:58,345][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:01:58,840][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:01:59,335][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:01:59,829][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:02:00,329][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:02:00,822][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:02:01,316][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:02:01,810][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:02:02,309][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:02:02,804][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:02:03,298][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:02:03,792][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:02:04,286][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:02:04,780][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:02:05,276][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:02:05,773][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:02:06,269][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:02:06,766][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:02:07,262][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:02:07,758][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:02:08,256][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:02:08,753][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9801 tokens. +[2026-03-26 08:02:09,597][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.09%, ΔTime: 00:00:32 +[2026-03-26 08:02:10,382][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:02:10,385][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:02:10,386][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:02:11,148][__main__][INFO] - Iteration 652 took 51s (29.93% Gen, 68.57% Train). Generation: 15s, Training: 35s. Estimated remaining time: 32h 36m 12s. Estimated total time: 42h 37m 34s. Time estimates for 10 more iterations: 8m 31s, 100 more iterations: 1h 25m 15s, 500 more iterations: 7h 6m 15s. +[2026-03-26 08:02:11,150][__main__][INFO] - Starting iteration 652. +[2026-03-26 08:02:11,551][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 65 and human policies 1. +[2026-03-26 08:02:11,551][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:02:21,681][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:02:26,115][__main__][INFO] - Number of regex retries in iteration 652: 1 +[2026-03-26 08:02:26,116][__main__][INFO] - agents played in iteration 652 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:02:26,875][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:02:26,899][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:02:26,923][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:02:26,947][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:02:26,947][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:02:26,948][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:02:27,777][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:02:28,230][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:02:28,730][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:02:29,233][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:02:29,735][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:02:30,234][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:02:30,731][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:02:31,228][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:02:31,726][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:02:32,229][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:02:32,730][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:02:33,252][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:02:33,755][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:02:34,254][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:02:34,752][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:02:35,250][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:02:35,748][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:02:36,246][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:02:36,744][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:02:37,243][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:02:37,741][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:02:38,245][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:02:38,744][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:02:39,243][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:02:39,745][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:02:40,241][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:02:40,739][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:02:41,237][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:02:41,734][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:02:42,236][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:02:42,738][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:02:43,239][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:02:43,743][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:02:44,238][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:02:44,734][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:02:45,230][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:02:45,725][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:02:46,221][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:02:46,716][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:02:47,211][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:02:47,717][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:02:48,222][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:02:48,718][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:02:49,216][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:02:49,712][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:02:50,207][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:02:50,709][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:02:51,205][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:02:51,702][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:02:52,198][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:02:52,699][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:02:53,200][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:02:53,696][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:02:54,193][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:02:54,691][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:02:55,187][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:02:55,684][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:02:56,183][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:02:56,675][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:02:57,172][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:02:57,664][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:02:58,162][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:02:58,654][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:02:59,150][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:02:59,651][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9806 tokens. +[2026-03-26 08:03:00,505][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.20%, Current % of VRAM taken: 60.63%, Block Peak % of device VRAM: 62.08%, ΔTime: 00:00:32 +[2026-03-26 08:03:01,288][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:03:01,290][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:03:01,292][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:03:02,041][__main__][INFO] - Iteration 653 took 50s (28.85% Gen, 69.67% Train). Generation: 14s, Training: 35s. Estimated remaining time: 32h 2m 20s. Estimated total time: 42h 4m 33s. Time estimates for 10 more iterations: 8m 24s, 100 more iterations: 1h 24m 9s, 500 more iterations: 7h 0m 45s. +[2026-03-26 08:03:02,043][__main__][INFO] - Starting iteration 653. +[2026-03-26 08:03:02,444][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 65 and human policies 1. +[2026-03-26 08:03:02,444][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:03:17,915][__main__][INFO] - Number of regex retries in iteration 653: 0 +[2026-03-26 08:03:17,915][__main__][INFO] - agents played in iteration 653 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:03:18,681][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:03:18,705][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:03:18,728][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:03:18,752][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:03:18,752][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:03:18,753][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:03:19,579][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:03:20,028][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:03:20,531][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:03:21,027][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:03:21,525][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:03:22,020][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:03:22,518][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:03:23,015][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:03:23,518][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:03:24,014][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:03:24,511][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:03:25,007][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:03:25,504][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:03:25,999][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:03:26,500][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:03:26,998][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:03:27,494][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:03:27,990][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:03:28,483][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:03:28,980][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:03:29,475][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:03:29,970][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:03:30,467][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:03:30,964][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:03:31,455][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:03:31,949][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:03:32,443][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:03:32,938][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:03:33,433][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:03:33,927][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:03:34,430][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:03:34,924][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:03:35,419][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:03:35,915][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:03:36,411][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:03:36,910][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:03:37,403][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:03:37,899][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:03:38,394][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:03:38,885][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:03:39,378][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:03:39,880][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:03:40,376][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:03:40,907][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:03:41,411][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:03:41,908][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:03:42,404][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:03:42,904][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:03:43,404][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:03:43,904][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:03:44,399][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:03:44,903][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:03:45,401][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:03:45,898][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:03:46,394][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:03:46,892][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:03:47,389][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:03:47,886][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:03:48,384][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:03:48,886][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:03:49,383][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:03:49,882][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:03:50,382][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:03:50,883][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:03:51,380][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9725 tokens. +[2026-03-26 08:03:52,243][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.05%, Current % of VRAM taken: 60.48%, Block Peak % of device VRAM: 62.18%, ΔTime: 00:00:32 +[2026-03-26 08:03:52,989][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:03:52,991][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:03:52,993][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:03:53,829][__main__][INFO] - Iteration 654 took 51s (30.11% Gen, 68.26% Train). Generation: 15s, Training: 35s. Estimated remaining time: 32h 46m 15s. Estimated total time: 42h 49m 20s. Time estimates for 10 more iterations: 8m 33s, 100 more iterations: 1h 25m 38s, 500 more iterations: 7h 8m 13s. +[2026-03-26 08:03:53,832][__main__][INFO] - Starting iteration 654. +[2026-03-26 08:03:54,232][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 65 and human policies 1. +[2026-03-26 08:03:54,232][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:04:10,601][__main__][INFO] - Number of regex retries in iteration 654: 0 +[2026-03-26 08:04:10,602][__main__][INFO] - agents played in iteration 654 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:04:11,376][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:04:11,400][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:04:11,423][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:04:11,446][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:04:11,447][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:04:11,447][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:04:12,273][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:04:12,725][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:04:13,227][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:04:13,730][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:04:14,228][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:04:14,723][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:04:15,219][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:04:15,720][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:04:16,217][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:04:16,722][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:04:17,223][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:04:17,723][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:04:18,223][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:04:18,722][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:04:19,220][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:04:19,718][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:04:20,218][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:04:20,712][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:04:21,210][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:04:21,707][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:04:22,198][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:04:22,694][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:04:23,185][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:04:23,677][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:04:24,180][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:04:24,680][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:04:25,184][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:04:25,682][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:04:26,180][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:04:26,678][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:04:27,177][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:04:27,675][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:04:28,174][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:04:28,679][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:04:29,179][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:04:29,675][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:04:30,171][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:04:30,669][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:04:31,168][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:04:31,666][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:04:32,164][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:04:32,661][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:04:33,157][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:04:33,651][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:04:34,149][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:04:34,643][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:04:35,138][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:04:35,634][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:04:36,129][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:04:36,627][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:04:37,125][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:04:37,623][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:04:38,120][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:04:38,618][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:04:39,116][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:04:39,618][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:04:40,117][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:04:40,613][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:04:41,108][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:04:41,603][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:04:42,101][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:04:42,597][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:04:43,092][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:04:43,589][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:04:44,085][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9790 tokens. +[2026-03-26 08:04:44,951][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.01%, Current % of VRAM taken: 60.43%, Block Peak % of device VRAM: 61.97%, ΔTime: 00:00:32 +[2026-03-26 08:04:45,906][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:04:45,908][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:04:45,909][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:04:46,698][__main__][INFO] - Iteration 655 took 52s (31.20% Gen, 67.29% Train). Generation: 16s, Training: 35s. Estimated remaining time: 33h 39m 23s. Estimated total time: 43h 43m 21s. Time estimates for 10 more iterations: 8m 44s, 100 more iterations: 1h 27m 26s, 500 more iterations: 7h 17m 13s. +[2026-03-26 08:04:46,702][__main__][INFO] - Starting iteration 655. +[2026-03-26 08:04:47,113][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 65 and human policies 1. +[2026-03-26 08:04:47,114][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:05:01,743][__main__][INFO] - Number of regex retries in iteration 655: 0 +[2026-03-26 08:05:01,745][__main__][INFO] - agents played in iteration 655 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:05:02,510][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:05:02,534][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:05:02,558][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:05:02,582][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:05:02,582][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:05:02,583][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:05:03,421][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:05:03,875][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:05:04,376][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:05:04,887][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:05:05,387][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:05:05,885][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:05:06,386][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:05:06,886][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:05:07,386][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:05:07,880][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:05:08,372][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:05:08,862][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:05:09,359][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:05:09,854][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:05:10,348][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:05:10,842][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:05:11,335][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:05:11,837][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:05:12,333][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:05:12,830][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:05:13,325][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:05:13,820][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:05:14,315][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:05:14,814][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:05:15,312][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:05:15,807][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:05:16,302][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:05:16,812][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:05:17,306][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:05:17,806][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:05:18,302][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:05:18,799][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:05:19,296][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:05:19,791][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:05:20,294][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:05:20,793][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:05:21,289][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:05:21,784][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:05:22,282][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:05:22,781][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:05:23,276][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:05:23,772][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:05:24,271][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:05:24,770][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:05:25,265][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:05:25,762][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:05:26,259][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:05:26,756][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:05:27,253][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:05:27,749][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:05:28,246][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:05:28,742][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:05:29,237][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:05:29,733][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:05:30,230][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:05:30,725][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:05:31,219][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:05:31,712][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:05:32,206][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:05:32,702][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:05:33,196][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:05:33,692][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:05:34,185][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:05:34,680][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:05:35,175][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9736 tokens. +[2026-03-26 08:05:36,044][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.88%, Current % of VRAM taken: 60.31%, Block Peak % of device VRAM: 61.95%, ΔTime: 00:00:32 +[2026-03-26 08:05:36,799][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:05:36,801][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:05:36,803][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:05:37,538][__main__][INFO] - Iteration 656 took 50s (29.01% Gen, 69.52% Train). Generation: 14s, Training: 35s. Estimated remaining time: 31h 56m 28s. Estimated total time: 42h 1m 17s. Time estimates for 10 more iterations: 8m 24s, 100 more iterations: 1h 24m 2s, 500 more iterations: 7h 0m 12s. +[2026-03-26 08:05:37,540][__main__][INFO] - Starting iteration 656. +[2026-03-26 08:05:37,950][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 65 and human policies 1. +[2026-03-26 08:05:37,950][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:06:07,805][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:06:08,659][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:06:09,553][__main__][INFO] - Number of regex retries in iteration 656: 2 +[2026-03-26 08:06:09,554][__main__][INFO] - agents played in iteration 656 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:06:10,319][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.41%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:06:10,343][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.41%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:06:10,366][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.41%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:06:10,389][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.41%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:06:10,390][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:06:10,390][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:06:11,214][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:06:11,663][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:06:12,165][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:06:12,659][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:06:13,155][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:06:13,646][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:06:14,142][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:06:14,635][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:06:15,131][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:06:15,631][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:06:16,131][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:06:16,628][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:06:17,124][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:06:17,621][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:06:18,120][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:06:18,616][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:06:19,112][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:06:19,617][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:06:20,126][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:06:20,625][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:06:21,121][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:06:21,616][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:06:22,110][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:06:22,604][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:06:23,099][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:06:23,602][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:06:24,098][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:06:24,596][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:06:25,093][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:06:25,591][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:06:26,085][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:06:26,588][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:06:27,089][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:06:27,609][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:06:28,104][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:06:28,602][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:06:29,098][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:06:29,601][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:06:30,102][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:06:30,600][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:06:31,096][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:06:31,591][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:06:32,093][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:06:32,595][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:06:33,097][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:06:33,599][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:06:34,099][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:06:34,605][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:06:35,104][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:06:35,601][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:06:36,101][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:06:36,598][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:06:37,096][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:06:37,594][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:06:38,089][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:06:38,586][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:06:39,083][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:06:39,581][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:06:40,086][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:06:40,584][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:06:41,083][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:06:41,580][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:06:42,076][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:06:42,573][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:06:43,068][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9832 tokens. +[2026-03-26 08:06:43,928][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.04%, Current % of VRAM taken: 59.46%, Block Peak % of device VRAM: 62.25%, ΔTime: 00:00:32 +[2026-03-26 08:06:44,701][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:06:44,703][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:06:44,705][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:06:45,873][__main__][INFO] - Iteration 657 took 1m 7s (46.53% Gen, 51.75% Train). Generation: 31s, Training: 35s. Estimated remaining time: 46h 30m 14s. Estimated total time: 56h 36m 12s. Time estimates for 10 more iterations: 11m 19s, 100 more iterations: 1h 53m 12s, 500 more iterations: 9h 26m 2s. +[2026-03-26 08:06:45,875][__main__][INFO] - Starting iteration 657. +[2026-03-26 08:06:46,277][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 65 and human policies 1. +[2026-03-26 08:06:46,278][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:07:00,170][__main__][INFO] - Number of regex retries in iteration 657: 0 +[2026-03-26 08:07:00,171][__main__][INFO] - agents played in iteration 657 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:07:00,937][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:07:00,961][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:07:00,985][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:07:01,008][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:07:01,009][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:07:01,010][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:07:01,839][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:07:02,290][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:07:02,787][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:07:03,286][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:07:03,784][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:07:04,280][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:07:04,777][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:07:05,274][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:07:05,770][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:07:06,265][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:07:06,767][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:07:07,263][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:07:07,759][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:07:08,254][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:07:08,754][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:07:09,251][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:07:09,746][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:07:10,240][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:07:10,734][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:07:11,236][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:07:11,732][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:07:12,226][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:07:12,719][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:07:13,213][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:07:13,706][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:07:14,204][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:07:14,702][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:07:15,201][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:07:15,700][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:07:16,197][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:07:16,695][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:07:17,193][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:07:17,691][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:07:18,196][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:07:18,698][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:07:19,203][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:07:19,702][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:07:20,203][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:07:20,701][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:07:21,200][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:07:21,698][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:07:22,196][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:07:22,693][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:07:23,191][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:07:23,689][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:07:24,186][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:07:24,684][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:07:25,208][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:07:25,704][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:07:26,201][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:07:26,698][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:07:27,195][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:07:27,700][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:07:28,197][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:07:28,696][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:07:29,194][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:07:29,693][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:07:30,189][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:07:30,688][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:07:31,184][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:07:31,679][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:07:32,177][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:07:32,674][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:07:33,173][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:07:33,674][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9807 tokens. +[2026-03-26 08:07:34,536][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.01%, Current % of VRAM taken: 60.43%, Block Peak % of device VRAM: 61.93%, ΔTime: 00:00:32 +[2026-03-26 08:07:35,271][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:07:35,273][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:07:35,275][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:07:36,028][__main__][INFO] - Iteration 658 took 49s (27.93% Gen, 70.56% Train). Generation: 13s, Training: 35s. Estimated remaining time: 31h 20m 47s. Estimated total time: 41h 27m 35s. Time estimates for 10 more iterations: 8m 17s, 100 more iterations: 1h 22m 55s, 500 more iterations: 6h 54m 35s. +[2026-03-26 08:07:36,030][__main__][INFO] - Starting iteration 658. +[2026-03-26 08:07:36,442][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 65 and human policies 1. +[2026-03-26 08:07:36,442][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:07:40,844][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:07:42,608][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:07:51,754][__main__][INFO] - Number of regex retries in iteration 658: 2 +[2026-03-26 08:07:51,754][__main__][INFO] - agents played in iteration 658 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:07:52,530][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:07:52,554][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:07:52,577][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:07:52,601][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:07:52,602][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:07:52,602][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:07:53,429][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:07:53,880][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:07:54,378][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:07:54,871][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:07:55,368][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:07:55,863][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:07:56,356][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:07:56,853][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:07:57,356][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:07:57,855][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:07:58,353][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:07:58,851][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:07:59,347][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:07:59,851][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:08:00,351][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:08:00,850][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:08:01,348][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:08:01,843][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:08:02,341][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:08:02,839][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:08:03,337][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:08:03,835][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:08:04,332][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:08:04,825][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:08:05,323][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:08:05,820][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:08:06,316][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:08:06,812][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:08:07,308][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:08:07,803][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:08:08,299][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:08:08,795][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:08:09,298][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:08:09,798][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:08:10,297][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:08:10,819][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:08:11,318][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:08:11,813][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:08:12,312][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:08:12,811][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:08:13,309][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:08:13,809][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:08:14,306][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:08:14,802][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:08:15,297][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:08:15,794][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:08:16,291][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:08:16,787][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:08:17,282][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:08:17,777][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:08:18,272][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:08:18,771][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:08:19,272][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:08:19,766][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:08:20,265][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:08:20,761][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:08:21,255][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:08:21,749][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:08:22,243][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:08:22,737][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:08:23,235][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:08:23,730][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:08:24,226][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:08:24,727][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:08:25,230][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9829 tokens. +[2026-03-26 08:08:26,090][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.04%, Current % of VRAM taken: 59.46%, Block Peak % of device VRAM: 62.11%, ΔTime: 00:00:32 +[2026-03-26 08:08:26,853][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:08:26,855][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:08:26,857][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:08:27,648][__main__][INFO] - Iteration 659 took 51s (29.90% Gen, 68.55% Train). Generation: 15s, Training: 35s. Estimated remaining time: 32h 32m 42s. Estimated total time: 42h 40m 21s. Time estimates for 10 more iterations: 8m 32s, 100 more iterations: 1h 25m 20s, 500 more iterations: 7h 6m 43s. +[2026-03-26 08:08:27,650][__main__][INFO] - Starting iteration 659. +[2026-03-26 08:08:28,052][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 65 and human policies 1. +[2026-03-26 08:08:28,053][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:08:35,678][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:08:43,525][__main__][INFO] - Number of regex retries in iteration 659: 1 +[2026-03-26 08:08:43,526][__main__][INFO] - agents played in iteration 659 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:08:44,287][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:08:44,310][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:08:44,334][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:08:44,357][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:08:44,357][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:08:44,358][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:08:45,180][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:08:45,633][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:08:46,130][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:08:46,627][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:08:47,124][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:08:47,621][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:08:48,121][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:08:48,619][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:08:49,116][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:08:49,618][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:08:50,116][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:08:50,614][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:08:51,109][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:08:51,608][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:08:52,104][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:08:52,600][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:08:53,102][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:08:53,600][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:08:54,099][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:08:54,594][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:08:55,090][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:08:55,589][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:08:56,090][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:08:56,586][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:08:57,082][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:08:57,579][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:08:58,072][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:08:58,568][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:08:59,067][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:08:59,561][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:09:00,056][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:09:00,556][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:09:01,054][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:09:01,554][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:09:02,047][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:09:02,543][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:09:03,037][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:09:03,534][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:09:04,029][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:09:04,526][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:09:05,021][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:09:05,517][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:09:06,012][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:09:06,504][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:09:06,999][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:09:07,493][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:09:07,988][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:09:08,483][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:09:08,980][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:09:09,478][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:09:09,977][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:09:10,476][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:09:10,983][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:09:11,485][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:09:11,984][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:09:12,483][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:09:12,982][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:09:13,484][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:09:13,985][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:09:14,487][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:09:14,990][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:09:15,493][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:09:16,001][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:09:16,507][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:09:17,014][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9817 tokens. +[2026-03-26 08:09:17,919][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.22%, Current % of VRAM taken: 60.65%, Block Peak % of device VRAM: 62.15%, ΔTime: 00:00:32 +[2026-03-26 08:09:18,713][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:09:18,715][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:09:18,716][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:09:19,446][__main__][INFO] - Iteration 660 took 51s (30.11% Gen, 68.47% Train). Generation: 15s, Training: 35s. Estimated remaining time: 32h 41m 14s. Estimated total time: 42h 49m 45s. Time estimates for 10 more iterations: 8m 33s, 100 more iterations: 1h 25m 39s, 500 more iterations: 7h 8m 17s. +[2026-03-26 08:09:19,448][__main__][INFO] - Starting iteration 660. +[2026-03-26 08:09:19,853][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 65 and human policies 1. +[2026-03-26 08:09:19,853][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:09:35,602][__main__][INFO] - Number of regex retries in iteration 660: 0 +[2026-03-26 08:09:35,603][__main__][INFO] - agents played in iteration 660 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:09:36,377][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:09:36,401][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:09:36,424][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:09:36,447][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:09:36,448][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:09:36,448][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:09:37,264][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:09:37,715][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:09:38,215][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:09:38,714][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:09:39,211][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:09:39,714][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:09:40,213][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:09:40,709][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:09:41,205][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:09:41,704][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:09:42,199][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:09:42,696][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:09:43,194][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:09:43,692][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:09:44,189][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:09:44,685][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:09:45,182][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:09:45,677][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:09:46,173][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:09:46,667][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:09:47,160][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:09:47,660][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:09:48,156][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:09:48,653][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:09:49,149][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:09:49,645][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:09:50,144][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:09:50,641][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:09:51,138][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:09:51,635][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:09:52,134][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:09:52,634][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:09:53,137][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:09:53,639][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:09:54,138][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:09:54,640][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:09:55,139][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:09:55,639][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:09:56,137][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:09:56,637][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:09:57,135][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:09:57,638][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:09:58,133][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:09:58,626][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:09:59,126][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:09:59,625][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:10:00,127][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:10:00,621][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:10:01,122][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:10:01,614][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:10:02,114][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:10:02,608][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:10:03,105][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:10:03,598][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:10:04,096][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:10:04,594][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:10:05,091][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:10:05,590][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:10:06,087][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:10:06,584][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:10:07,082][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:10:07,583][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:10:08,087][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:10:08,588][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:10:09,091][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9879 tokens. +[2026-03-26 08:10:09,984][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 62.09%, ΔTime: 00:00:32 +[2026-03-26 08:10:10,734][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:10:10,738][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:10:10,740][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:10:12,074][__main__][INFO] - Iteration 661 took 52s (30.16% Gen, 67.28% Train). Generation: 15s, Training: 35s. Estimated remaining time: 33h 21m 41s. Estimated total time: 43h 31m 5s. Time estimates for 10 more iterations: 8m 42s, 100 more iterations: 1h 27m 2s, 500 more iterations: 7h 15m 10s. +[2026-03-26 08:10:12,076][__main__][INFO] - Starting iteration 661. +[2026-03-26 08:10:12,483][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 66 and human policies 1. +[2026-03-26 08:10:12,483][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:10:17,550][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:10:28,024][__main__][INFO] - Number of regex retries in iteration 661: 1 +[2026-03-26 08:10:28,024][__main__][INFO] - agents played in iteration 661 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:10:28,785][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:10:28,808][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:10:28,832][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:10:28,855][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:10:28,856][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:10:28,856][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:10:29,683][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:10:30,139][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:10:30,643][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:10:31,141][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:10:31,641][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:10:32,140][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:10:32,639][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:10:33,137][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:10:33,637][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:10:34,139][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:10:34,641][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:10:35,141][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:10:35,642][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:10:36,142][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:10:36,642][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:10:37,143][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:10:37,643][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:10:38,140][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:10:38,638][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:10:39,134][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:10:39,633][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:10:40,130][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:10:40,627][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:10:41,125][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:10:41,622][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:10:42,120][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:10:42,621][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:10:43,114][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:10:43,615][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:10:44,113][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:10:44,606][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:10:45,102][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:10:45,600][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:10:46,097][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:10:46,598][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:10:47,096][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:10:47,595][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:10:48,093][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:10:48,591][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:10:49,091][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:10:49,585][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:10:50,083][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:10:50,583][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:10:51,080][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:10:51,579][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:10:52,077][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:10:52,575][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:10:53,073][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:10:53,572][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:10:54,068][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:10:54,559][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:10:55,054][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:10:55,549][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:10:56,045][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:10:56,541][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:10:57,036][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:10:57,533][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:10:58,030][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:10:58,523][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:10:59,018][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:10:59,515][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:11:00,014][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:11:00,514][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:11:01,016][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:11:01,514][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9921 tokens. +[2026-03-26 08:11:02,400][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.28%, ΔTime: 00:00:32 +[2026-03-26 08:11:03,214][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:11:03,217][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:11:03,218][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:11:04,271][__main__][INFO] - Iteration 662 took 51s (30.01% Gen, 67.96% Train). Generation: 15s, Training: 35s. Estimated remaining time: 32h 59m 12s. Estimated total time: 43h 9m 27s. Time estimates for 10 more iterations: 8m 37s, 100 more iterations: 1h 26m 18s, 500 more iterations: 7h 11m 34s. +[2026-03-26 08:11:04,273][__main__][INFO] - Starting iteration 662. +[2026-03-26 08:11:04,678][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 66 and human policies 1. +[2026-03-26 08:11:04,679][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:11:27,681][__main__][INFO] - Number of regex retries in iteration 662: 0 +[2026-03-26 08:11:27,682][__main__][INFO] - agents played in iteration 662 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:11:28,448][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:11:28,472][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:11:28,495][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:11:28,519][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:11:28,519][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:11:28,520][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:11:29,354][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:11:29,813][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:11:30,317][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:11:30,816][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:11:31,316][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:11:31,815][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:11:32,313][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:11:32,815][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:11:33,314][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:11:33,811][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:11:34,304][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:11:34,800][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:11:35,294][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:11:35,792][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:11:36,287][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:11:36,785][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:11:37,288][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:11:37,790][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:11:38,287][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:11:38,784][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:11:39,283][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:11:39,777][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:11:40,276][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:11:40,773][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:11:41,272][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:11:41,772][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:11:42,271][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:11:42,770][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:11:43,269][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:11:43,766][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:11:44,266][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:11:44,765][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:11:45,264][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:11:45,763][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:11:46,261][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:11:46,757][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:11:47,252][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:11:47,754][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:11:48,255][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:11:48,755][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:11:49,253][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:11:49,750][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:11:50,253][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:11:50,747][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:11:51,242][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:11:51,737][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:11:52,231][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:11:52,727][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:11:53,228][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:11:53,723][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:11:54,221][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:11:54,717][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:11:55,216][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:11:55,721][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:11:56,222][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:11:56,723][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:11:57,219][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:11:57,715][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:11:58,214][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:11:58,713][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:11:59,214][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:11:59,711][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:12:00,206][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:12:00,706][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:12:01,202][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9864 tokens. +[2026-03-26 08:12:02,064][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.01%, Current % of VRAM taken: 60.44%, Block Peak % of device VRAM: 62.06%, ΔTime: 00:00:32 +[2026-03-26 08:12:02,818][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:12:02,821][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:12:02,822][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:12:03,559][__main__][INFO] - Iteration 663 took 58s (39.07% Gen, 59.68% Train). Generation: 23s, Training: 35s. Estimated remaining time: 38h 52m 51s. Estimated total time: 49h 4m 6s. Time estimates for 10 more iterations: 9m 48s, 100 more iterations: 1h 38m 8s, 500 more iterations: 8h 10m 41s. +[2026-03-26 08:12:03,561][__main__][INFO] - Starting iteration 663. +[2026-03-26 08:12:03,964][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 66 and human policies 1. +[2026-03-26 08:12:03,964][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:12:08,163][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:12:18,602][__main__][INFO] - Number of regex retries in iteration 663: 1 +[2026-03-26 08:12:18,603][__main__][INFO] - agents played in iteration 663 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:12:19,369][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:12:19,393][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:12:19,416][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:12:19,440][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:12:19,440][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:12:19,441][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:12:20,273][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:12:20,727][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:12:21,228][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:12:21,731][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:12:22,239][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:12:22,742][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:12:23,246][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:12:23,746][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:12:24,246][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:12:24,748][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:12:25,251][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:12:25,761][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:12:26,261][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:12:26,764][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:12:27,264][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:12:27,764][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:12:28,262][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:12:28,766][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:12:29,265][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:12:29,764][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:12:30,273][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:12:30,777][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:12:31,278][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:12:31,778][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:12:32,275][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:12:32,809][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:12:33,313][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:12:33,815][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:12:34,316][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:12:34,815][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:12:35,322][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:12:35,823][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:12:36,326][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:12:36,824][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:12:37,322][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:12:37,822][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:12:38,321][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:12:38,820][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:12:39,320][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:12:39,819][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:12:40,321][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:12:40,816][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:12:41,315][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:12:41,814][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:12:42,313][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:12:42,810][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:12:43,311][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:12:43,810][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:12:44,314][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:12:44,812][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:12:45,310][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:12:45,810][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:12:46,308][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:12:46,805][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:12:47,302][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:12:47,801][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:12:48,298][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:12:48,796][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:12:49,294][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:12:49,792][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:12:50,293][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:12:50,793][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:12:51,299][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:12:51,798][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:12:52,295][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9908 tokens. +[2026-03-26 08:12:53,178][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.29%, ΔTime: 00:00:32 +[2026-03-26 08:12:53,872][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:12:53,874][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:12:53,876][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:12:54,624][__main__][INFO] - Iteration 664 took 50s (28.89% Gen, 69.63% Train). Generation: 14s, Training: 35s. Estimated remaining time: 32h 0m 56s. Estimated total time: 42h 13m 2s. Time estimates for 10 more iterations: 8m 26s, 100 more iterations: 1h 24m 26s, 500 more iterations: 7h 2m 10s. +[2026-03-26 08:12:54,626][__main__][INFO] - Starting iteration 664. +[2026-03-26 08:12:55,032][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 66 and human policies 1. +[2026-03-26 08:12:55,033][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:13:10,488][__main__][INFO] - Number of regex retries in iteration 664: 0 +[2026-03-26 08:13:10,489][__main__][INFO] - agents played in iteration 664 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:13:11,253][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:13:11,277][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:13:11,300][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:13:11,323][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:13:11,324][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:13:11,324][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:13:12,164][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:13:12,622][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:13:13,133][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:13:13,636][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:13:14,138][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:13:14,639][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:13:15,142][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:13:15,646][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:13:16,149][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:13:16,655][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:13:17,161][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:13:17,665][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:13:18,170][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:13:18,674][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:13:19,177][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:13:19,677][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:13:20,178][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:13:20,674][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:13:21,170][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:13:21,664][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:13:22,160][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:13:22,655][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:13:23,151][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:13:23,649][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:13:24,145][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:13:24,642][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:13:25,146][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:13:25,650][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:13:26,147][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:13:26,643][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:13:27,138][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:13:27,632][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:13:28,127][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:13:28,625][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:13:29,119][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:13:29,615][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:13:30,111][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:13:30,611][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:13:31,109][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:13:31,605][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:13:32,104][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:13:32,604][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:13:33,099][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:13:33,592][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:13:34,086][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:13:34,581][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:13:35,075][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:13:35,571][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:13:36,072][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:13:36,570][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:13:37,065][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:13:37,559][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:13:38,053][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:13:38,551][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:13:39,044][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:13:39,536][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:13:40,028][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:13:40,523][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:13:41,025][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:13:41,523][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:13:42,015][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:13:42,514][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:13:43,020][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:13:43,521][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:13:44,020][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9837 tokens. +[2026-03-26 08:13:44,894][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.19%, ΔTime: 00:00:32 +[2026-03-26 08:13:45,741][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:13:45,743][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:13:45,745][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:13:46,761][__main__][INFO] - Iteration 665 took 51s (29.88% Gen, 68.15% Train). Generation: 15s, Training: 35s. Estimated remaining time: 32h 53m 31s. Estimated total time: 43h 6m 29s. Time estimates for 10 more iterations: 8m 37s, 100 more iterations: 1h 26m 12s, 500 more iterations: 7h 11m 4s. +[2026-03-26 08:13:46,763][__main__][INFO] - Starting iteration 665. +[2026-03-26 08:13:47,165][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 66 and human policies 1. +[2026-03-26 08:13:47,166][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:14:02,569][__main__][INFO] - Number of regex retries in iteration 665: 0 +[2026-03-26 08:14:02,570][__main__][INFO] - agents played in iteration 665 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:14:03,328][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:14:03,352][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:14:03,376][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:14:03,400][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:14:03,401][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:14:03,401][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:14:04,228][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:14:04,686][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:14:05,180][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:14:05,685][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:14:06,187][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:14:06,693][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:14:07,186][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:14:07,684][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:14:08,182][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:14:08,682][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:14:09,183][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:14:09,684][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:14:10,184][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:14:10,687][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:14:11,188][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:14:11,686][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:14:12,189][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:14:12,686][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:14:13,182][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:14:13,685][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:14:14,182][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:14:14,679][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:14:15,176][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:14:15,674][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:14:16,171][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:14:16,672][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:14:17,175][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:14:17,679][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:14:18,183][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:14:18,685][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:14:19,185][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:14:19,685][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:14:20,189][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:14:20,680][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:14:21,178][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:14:21,678][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:14:22,178][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:14:22,676][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:14:23,178][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:14:23,678][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:14:24,176][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:14:24,677][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:14:25,177][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:14:25,677][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:14:26,180][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:14:26,679][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:14:27,179][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:14:27,678][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:14:28,178][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:14:28,674][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:14:29,170][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:14:29,666][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:14:30,164][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:14:30,661][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:14:31,158][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:14:31,655][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:14:32,153][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:14:32,651][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:14:33,147][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:14:33,645][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:14:34,142][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:14:34,639][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:14:35,136][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:14:35,634][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:14:36,132][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9934 tokens. +[2026-03-26 08:14:37,001][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.97%, Current % of VRAM taken: 60.39%, Block Peak % of device VRAM: 62.04%, ΔTime: 00:00:32 +[2026-03-26 08:14:37,812][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:14:37,815][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:14:37,816][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:14:38,719][__main__][INFO] - Iteration 666 took 51s (29.88% Gen, 68.37% Train). Generation: 15s, Training: 35s. Estimated remaining time: 32h 43m 52s. Estimated total time: 42h 57m 42s. Time estimates for 10 more iterations: 8m 35s, 100 more iterations: 1h 25m 55s, 500 more iterations: 7h 9m 37s. +[2026-03-26 08:14:38,722][__main__][INFO] - Starting iteration 666. +[2026-03-26 08:14:39,127][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 66 and human policies 1. +[2026-03-26 08:14:39,128][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:14:54,671][__main__][INFO] - Number of regex retries in iteration 666: 0 +[2026-03-26 08:14:54,672][__main__][INFO] - agents played in iteration 666 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:14:55,431][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:14:55,454][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:14:55,478][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:14:55,501][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:14:55,501][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:14:55,502][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:14:56,338][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:14:56,795][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:14:57,306][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:14:57,810][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:14:58,311][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:14:58,817][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:14:59,320][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:14:59,824][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:15:00,327][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:15:00,822][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:15:01,324][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:15:01,826][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:15:02,322][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:15:02,817][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:15:03,309][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:15:03,800][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:15:04,295][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:15:04,792][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:15:05,287][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:15:05,782][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:15:06,277][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:15:06,768][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:15:07,264][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:15:07,756][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:15:08,248][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:15:08,746][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:15:09,249][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:15:09,751][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:15:10,251][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:15:10,749][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:15:11,247][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:15:11,745][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:15:12,242][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:15:12,738][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:15:13,236][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:15:13,732][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:15:14,236][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:15:14,735][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:15:15,233][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:15:15,729][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:15:16,224][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:15:16,716][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:15:17,213][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:15:17,704][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:15:18,200][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:15:18,691][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:15:19,181][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:15:19,676][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:15:20,174][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:15:20,679][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:15:21,180][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:15:21,701][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:15:22,204][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:15:22,704][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:15:23,206][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:15:23,705][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:15:24,205][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:15:24,701][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:15:25,199][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:15:25,697][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:15:26,196][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:15:26,694][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:15:27,192][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:15:27,690][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:15:28,182][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9780 tokens. +[2026-03-26 08:15:29,046][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.01%, Current % of VRAM taken: 60.44%, Block Peak % of device VRAM: 62.46%, ΔTime: 00:00:32 +[2026-03-26 08:15:29,755][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:15:29,757][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:15:29,759][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:15:30,469][__main__][INFO] - Iteration 667 took 51s (30.27% Gen, 68.34% Train). Generation: 15s, Training: 35s. Estimated remaining time: 32h 32m 28s. Estimated total time: 42h 47m 9s. Time estimates for 10 more iterations: 8m 33s, 100 more iterations: 1h 25m 34s, 500 more iterations: 7h 7m 51s. +[2026-03-26 08:15:30,472][__main__][INFO] - Starting iteration 667. +[2026-03-26 08:15:30,885][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 66 and human policies 1. +[2026-03-26 08:15:30,885][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:15:46,310][__main__][INFO] - Number of regex retries in iteration 667: 0 +[2026-03-26 08:15:46,311][__main__][INFO] - agents played in iteration 667 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:15:47,083][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:15:47,106][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:15:47,130][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:15:47,153][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:15:47,154][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:15:47,154][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:15:47,992][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:15:48,445][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:15:48,955][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:15:49,457][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:15:49,959][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:15:50,460][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:15:50,960][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:15:51,463][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:15:51,964][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:15:52,465][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:15:52,967][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:15:53,466][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:15:53,963][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:15:54,461][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:15:54,957][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:15:55,455][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:15:55,953][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:15:56,455][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:15:56,954][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:15:57,450][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:15:57,948][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:15:58,448][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:15:58,944][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:15:59,440][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:15:59,937][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:16:00,438][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:16:00,945][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:16:01,444][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:16:01,941][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:16:02,438][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:16:02,937][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:16:03,435][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:16:03,936][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:16:04,437][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:16:04,933][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:16:05,430][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:16:05,926][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:16:06,422][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:16:06,918][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:16:07,419][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:16:07,915][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:16:08,412][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:16:08,909][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:16:09,407][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:16:09,903][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:16:10,399][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:16:10,895][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:16:11,391][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:16:11,889][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:16:12,386][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:16:12,884][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:16:13,383][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:16:13,883][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:16:14,380][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:16:14,878][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:16:15,376][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:16:15,876][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:16:16,372][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:16:16,867][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:16:17,364][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:16:17,861][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:16:18,365][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:16:18,861][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:16:19,357][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:16:19,852][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9872 tokens. +[2026-03-26 08:16:20,719][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.95%, Current % of VRAM taken: 60.38%, Block Peak % of device VRAM: 61.98%, ΔTime: 00:00:32 +[2026-03-26 08:16:21,493][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:16:21,495][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:16:21,497][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:16:22,349][__main__][INFO] - Iteration 668 took 51s (29.97% Gen, 68.37% Train). Generation: 15s, Training: 35s. Estimated remaining time: 32h 37m 43s. Estimated total time: 42h 53m 16s. Time estimates for 10 more iterations: 8m 34s, 100 more iterations: 1h 25m 46s, 500 more iterations: 7h 8m 52s. +[2026-03-26 08:16:22,352][__main__][INFO] - Starting iteration 668. +[2026-03-26 08:16:22,762][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 66 and human policies 1. +[2026-03-26 08:16:22,763][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:16:37,099][__main__][INFO] - Number of regex retries in iteration 668: 0 +[2026-03-26 08:16:37,100][__main__][INFO] - agents played in iteration 668 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:16:37,915][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:16:37,938][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:16:37,962][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:16:37,985][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:16:37,986][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:16:37,986][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:16:38,849][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:16:39,308][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:16:39,815][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:16:40,319][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:16:40,823][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:16:41,327][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:16:41,830][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:16:42,334][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:16:42,838][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:16:43,340][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:16:43,841][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:16:44,340][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:16:44,838][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:16:45,338][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:16:45,837][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:16:46,335][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:16:46,834][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:16:47,332][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:16:47,829][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:16:48,324][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:16:48,821][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:16:49,318][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:16:49,815][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:16:50,312][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:16:50,805][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:16:51,298][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:16:51,799][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:16:52,299][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:16:52,798][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:16:53,295][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:16:53,792][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:16:54,292][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:16:54,795][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:16:55,293][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:16:55,790][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:16:56,284][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:16:56,779][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:16:57,276][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:16:57,771][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:16:58,265][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:16:58,762][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:16:59,256][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:16:59,750][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:17:00,245][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:17:00,743][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:17:01,240][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:17:01,736][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:17:02,231][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:17:02,738][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:17:03,237][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:17:03,739][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:17:04,236][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:17:04,733][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:17:05,231][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:17:05,728][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:17:06,226][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:17:06,723][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:17:07,220][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:17:07,713][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:17:08,209][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:17:08,704][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:17:09,198][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:17:09,691][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:17:10,185][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:17:10,680][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9828 tokens. +[2026-03-26 08:17:11,542][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.48%, Block Peak % of device VRAM: 61.91%, ΔTime: 00:00:32 +[2026-03-26 08:17:12,292][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:17:12,294][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:17:12,297][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:17:13,071][__main__][INFO] - Iteration 669 took 50s (28.50% Gen, 69.96% Train). Generation: 14s, Training: 35s. Estimated remaining time: 31h 39m 3s. Estimated total time: 41h 55m 27s. Time estimates for 10 more iterations: 8m 23s, 100 more iterations: 1h 23m 50s, 500 more iterations: 6h 59m 14s. +[2026-03-26 08:17:13,073][__main__][INFO] - Starting iteration 669. +[2026-03-26 08:17:13,485][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 66 and human policies 1. +[2026-03-26 08:17:13,486][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:17:28,387][__main__][INFO] - Number of regex retries in iteration 669: 0 +[2026-03-26 08:17:28,388][__main__][INFO] - agents played in iteration 669 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:17:29,190][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:17:29,213][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:17:29,237][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:17:29,260][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:17:29,261][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:17:29,261][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:17:30,136][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:17:30,595][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:17:31,104][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:17:31,615][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:17:32,120][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:17:32,625][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:17:33,129][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:17:33,633][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:17:34,135][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:17:34,632][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:17:35,123][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:17:35,614][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:17:36,106][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:17:36,607][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:17:37,103][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:17:37,601][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:17:38,098][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:17:38,597][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:17:39,097][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:17:39,596][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:17:40,095][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:17:40,595][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:17:41,094][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:17:41,593][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:17:42,097][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:17:42,601][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:17:43,105][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:17:43,605][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:17:44,106][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:17:44,603][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:17:45,099][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:17:45,602][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:17:46,106][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:17:46,604][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:17:47,102][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:17:47,600][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:17:48,101][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:17:48,600][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:17:49,097][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:17:49,594][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:17:50,094][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:17:50,595][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:17:51,093][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:17:51,603][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:17:52,101][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:17:52,599][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:17:53,097][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:17:53,599][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:17:54,097][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:17:54,599][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:17:55,092][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:17:55,584][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:17:56,079][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:17:56,575][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:17:57,077][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:17:57,571][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:17:58,066][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:17:58,564][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:17:59,065][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:17:59,562][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:18:00,060][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:18:00,558][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:18:01,057][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:18:01,554][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:18:02,051][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9992 tokens. +[2026-03-26 08:18:02,923][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.31%, ΔTime: 00:00:32 +[2026-03-26 08:18:03,650][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:18:03,652][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:18:03,654][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:18:04,409][__main__][INFO] - Iteration 670 took 50s (29.26% Gen, 69.25% Train). Generation: 14s, Training: 35s. Estimated remaining time: 32h 8m 56s. Estimated total time: 42h 26m 11s. Time estimates for 10 more iterations: 8m 29s, 100 more iterations: 1h 24m 52s, 500 more iterations: 7h 4m 21s. +[2026-03-26 08:18:04,411][__main__][INFO] - Starting iteration 670. +[2026-03-26 08:18:04,812][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 66 and human policies 1. +[2026-03-26 08:18:04,813][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:18:19,266][__main__][INFO] - Number of regex retries in iteration 670: 0 +[2026-03-26 08:18:19,267][__main__][INFO] - agents played in iteration 670 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:18:20,066][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:18:20,090][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:18:20,114][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:18:20,137][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:18:20,138][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:18:20,138][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:18:21,024][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:18:21,480][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:18:21,987][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:18:22,488][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:18:22,987][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:18:23,486][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:18:23,992][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:18:24,492][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:18:24,990][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:18:25,490][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:18:25,986][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:18:26,484][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:18:26,983][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:18:27,481][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:18:27,977][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:18:28,476][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:18:28,975][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:18:29,473][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:18:29,971][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:18:30,468][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:18:30,966][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:18:31,462][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:18:31,964][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:18:32,461][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:18:32,959][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:18:33,455][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:18:33,959][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:18:34,461][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:18:34,967][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:18:35,466][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:18:35,963][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:18:36,459][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:18:36,956][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:18:37,457][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:18:37,956][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:18:38,457][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:18:38,955][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:18:39,456][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:18:39,956][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:18:40,454][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:18:40,975][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:18:41,472][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:18:41,968][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:18:42,464][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:18:42,960][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:18:43,456][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:18:43,950][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:18:44,447][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:18:44,944][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:18:45,441][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:18:45,938][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:18:46,436][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:18:46,932][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:18:47,430][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:18:47,928][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:18:48,426][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:18:48,923][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:18:49,421][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:18:49,919][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:18:50,414][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:18:50,909][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:18:51,406][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:18:51,903][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:18:52,400][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:18:52,898][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9836 tokens. +[2026-03-26 08:18:53,747][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.09%, ΔTime: 00:00:32 +[2026-03-26 08:18:54,498][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:18:54,500][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:18:54,501][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:18:55,969][__main__][INFO] - Iteration 671 took 51s (28.25% Gen, 68.87% Train). Generation: 14s, Training: 35s. Estimated remaining time: 32h 19m 46s. Estimated total time: 42h 37m 53s. Time estimates for 10 more iterations: 8m 31s, 100 more iterations: 1h 25m 15s, 500 more iterations: 7h 6m 18s. +[2026-03-26 08:18:55,972][__main__][INFO] - Starting iteration 671. +[2026-03-26 08:18:56,380][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 67 and human policies 1. +[2026-03-26 08:18:56,381][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:19:10,769][__main__][INFO] - Number of regex retries in iteration 671: 0 +[2026-03-26 08:19:10,770][__main__][INFO] - agents played in iteration 671 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:19:11,560][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:19:11,584][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:19:11,608][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:19:11,632][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:19:11,633][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:19:11,633][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:19:12,635][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:19:13,110][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:19:13,627][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:19:14,126][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:19:14,625][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:19:15,124][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:19:15,621][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:19:16,119][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:19:16,617][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:19:17,120][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:19:17,622][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:19:18,124][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:19:18,630][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:19:19,131][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:19:19,634][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:19:20,138][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:19:20,647][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:19:21,150][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:19:21,693][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:19:22,215][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:19:22,717][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:19:23,224][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:19:23,723][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:19:24,221][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:19:24,721][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:19:25,219][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:19:25,719][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:19:26,228][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:19:26,735][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:19:27,237][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:19:27,733][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:19:28,234][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:19:28,733][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:19:29,240][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:19:29,737][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:19:30,236][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:19:30,735][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:19:31,234][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:19:31,728][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:19:32,226][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:19:32,723][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:19:33,221][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:19:33,719][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:19:34,223][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:19:34,721][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:19:35,217][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:19:35,715][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:19:36,215][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:19:36,712][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:19:37,209][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:19:37,704][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:19:38,199][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:19:38,694][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:19:39,189][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:19:39,682][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:19:40,176][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:19:40,673][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:19:41,170][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:19:41,661][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:19:42,157][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:19:42,661][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:19:43,163][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:19:43,663][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:19:44,165][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:19:44,667][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9824 tokens. +[2026-03-26 08:19:45,515][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.01%, Current % of VRAM taken: 60.44%, Block Peak % of device VRAM: 61.96%, ΔTime: 00:00:32 +[2026-03-26 08:19:46,276][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:19:46,278][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:19:46,280][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:19:47,113][__main__][INFO] - Iteration 672 took 50s (28.36% Gen, 69.99% Train). Generation: 14s, Training: 35s. Estimated remaining time: 31h 57m 42s. Estimated total time: 42h 16m 41s. Time estimates for 10 more iterations: 8m 27s, 100 more iterations: 1h 24m 33s, 500 more iterations: 7h 2m 46s. +[2026-03-26 08:19:47,116][__main__][INFO] - Starting iteration 672. +[2026-03-26 08:19:47,516][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 67 and human policies 1. +[2026-03-26 08:19:47,517][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:20:05,000][__main__][INFO] - Number of regex retries in iteration 672: 0 +[2026-03-26 08:20:05,001][__main__][INFO] - agents played in iteration 672 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:20:05,780][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:20:05,805][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:20:05,830][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:20:05,855][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:20:05,855][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:20:05,856][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:20:06,692][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:20:07,145][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:20:07,655][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:20:08,152][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:20:08,645][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:20:09,138][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:20:09,633][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:20:10,129][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:20:10,627][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:20:11,129][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:20:11,625][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:20:12,122][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:20:12,618][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:20:13,114][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:20:13,614][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:20:14,115][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:20:14,606][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:20:15,100][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:20:15,598][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:20:16,096][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:20:16,595][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:20:17,094][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:20:17,592][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:20:18,087][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:20:18,587][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:20:19,091][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:20:19,591][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:20:20,087][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:20:20,584][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:20:21,081][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:20:21,579][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:20:22,080][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:20:22,578][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:20:23,085][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:20:23,584][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:20:24,086][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:20:24,584][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:20:25,083][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:20:25,588][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:20:26,085][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:20:26,584][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:20:27,081][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:20:27,576][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:20:28,073][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:20:28,570][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:20:29,065][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:20:29,562][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:20:30,060][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:20:30,558][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:20:31,056][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:20:31,555][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:20:32,061][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:20:32,560][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:20:33,059][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:20:33,558][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:20:34,057][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:20:34,555][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:20:35,051][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:20:35,546][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:20:36,044][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:20:36,542][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:20:37,037][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:20:37,534][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:20:38,030][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:20:38,528][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9837 tokens. +[2026-03-26 08:20:39,389][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.00%, ΔTime: 00:00:32 +[2026-03-26 08:20:40,145][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:20:40,147][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:20:40,149][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:20:40,944][__main__][INFO] - Iteration 673 took 53s (32.72% Gen, 65.79% Train). Generation: 17s, Training: 35s. Estimated remaining time: 34h 11m 35s. Estimated total time: 44h 31m 27s. Time estimates for 10 more iterations: 8m 54s, 100 more iterations: 1h 29m 2s, 500 more iterations: 7h 25m 14s. +[2026-03-26 08:20:40,947][__main__][INFO] - Starting iteration 673. +[2026-03-26 08:20:41,347][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 67 and human policies 1. +[2026-03-26 08:20:41,348][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:20:53,434][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:20:55,396][__main__][INFO] - Number of regex retries in iteration 673: 1 +[2026-03-26 08:20:55,397][__main__][INFO] - agents played in iteration 673 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:20:56,174][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:20:56,198][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:20:56,222][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:20:56,247][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:20:56,248][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:20:56,248][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:20:57,108][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:20:57,573][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:20:58,075][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:20:58,581][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:20:59,082][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:20:59,582][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:21:00,076][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:21:00,576][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:21:01,077][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:21:01,575][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:21:02,072][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:21:02,571][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:21:03,070][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:21:03,569][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:21:04,071][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:21:04,571][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:21:05,078][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:21:05,576][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:21:06,074][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:21:06,570][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:21:07,066][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:21:07,561][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:21:08,059][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:21:08,560][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:21:09,060][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:21:09,559][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:21:10,054][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:21:10,551][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:21:11,051][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:21:11,554][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:21:12,063][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:21:12,567][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:21:13,067][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:21:13,567][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:21:14,067][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:21:14,566][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:21:15,064][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:21:15,561][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:21:16,059][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:21:16,557][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:21:17,053][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:21:17,552][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:21:18,050][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:21:18,548][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:21:19,046][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:21:19,544][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:21:20,049][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:21:20,550][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:21:21,049][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:21:21,550][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:21:22,047][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:21:22,549][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:21:23,046][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:21:23,546][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:21:24,046][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:21:24,544][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:21:25,048][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:21:25,547][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:21:26,049][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:21:26,549][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:21:27,049][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:21:27,548][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:21:28,047][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:21:28,551][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:21:29,051][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9794 tokens. +[2026-03-26 08:21:29,923][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 61.93%, ΔTime: 00:00:32 +[2026-03-26 08:21:30,705][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:21:30,708][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:21:30,709][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:21:31,451][__main__][INFO] - Iteration 674 took 50s (28.04% Gen, 70.48% Train). Generation: 14s, Training: 35s. Estimated remaining time: 31h 24m 30s. Estimated total time: 41h 45m 12s. Time estimates for 10 more iterations: 8m 21s, 100 more iterations: 1h 23m 30s, 500 more iterations: 6h 57m 32s. +[2026-03-26 08:21:31,453][__main__][INFO] - Starting iteration 674. +[2026-03-26 08:21:31,853][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 67 and human policies 1. +[2026-03-26 08:21:31,854][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:21:56,474][__main__][INFO] - Number of regex retries in iteration 674: 0 +[2026-03-26 08:21:56,475][__main__][INFO] - agents played in iteration 674 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:21:57,239][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:21:57,263][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:21:57,286][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:21:57,310][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:21:57,311][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:21:57,311][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:21:58,141][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:21:58,595][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:21:59,106][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:21:59,614][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:22:00,122][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:22:00,621][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:22:01,122][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:22:01,622][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:22:02,124][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:22:02,627][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:22:03,133][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:22:03,631][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:22:04,135][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:22:04,635][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:22:05,137][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:22:05,646][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:22:06,156][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:22:06,659][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:22:07,160][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:22:07,658][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:22:08,154][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:22:08,651][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:22:09,147][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:22:09,643][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:22:10,143][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:22:10,635][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:22:11,124][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:22:11,615][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:22:12,111][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:22:12,602][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:22:13,094][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:22:13,592][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:22:14,088][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:22:14,583][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:22:15,083][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:22:15,581][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:22:16,080][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:22:16,579][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:22:17,077][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:22:17,576][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:22:18,079][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:22:18,580][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:22:19,079][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:22:19,587][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:22:20,089][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:22:20,588][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:22:21,087][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:22:21,584][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:22:22,082][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:22:22,579][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:22:23,074][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:22:23,571][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:22:24,066][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:22:24,561][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:22:25,055][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:22:25,563][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:22:26,061][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:22:26,557][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:22:27,054][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:22:27,552][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:22:28,048][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:22:28,545][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:22:29,042][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:22:29,537][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:22:30,034][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9809 tokens. +[2026-03-26 08:22:30,900][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.99%, Current % of VRAM taken: 60.41%, Block Peak % of device VRAM: 62.02%, ΔTime: 00:00:32 +[2026-03-26 08:22:31,667][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:22:31,669][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:22:31,671][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:22:32,393][__main__][INFO] - Iteration 675 took 1m 0s (40.67% Gen, 58.14% Train). Generation: 24s, Training: 35s. Estimated remaining time: 40h 5m 17s. Estimated total time: 50h 27m 0s. Time estimates for 10 more iterations: 10m 5s, 100 more iterations: 1h 40m 54s, 500 more iterations: 8h 24m 30s. +[2026-03-26 08:22:32,395][__main__][INFO] - Starting iteration 675. +[2026-03-26 08:22:32,796][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 67 and human policies 1. +[2026-03-26 08:22:32,797][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:22:58,472][__main__][INFO] - Number of regex retries in iteration 675: 0 +[2026-03-26 08:22:58,473][__main__][INFO] - agents played in iteration 675 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:22:59,227][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:22:59,251][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:22:59,274][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:22:59,298][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:22:59,299][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:22:59,299][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:23:00,146][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:23:00,599][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:23:01,111][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:23:01,613][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:23:02,117][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:23:02,621][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:23:03,123][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:23:03,622][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:23:04,121][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:23:04,615][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:23:05,110][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:23:05,605][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:23:06,101][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:23:06,599][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:23:07,095][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:23:07,597][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:23:08,095][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:23:08,596][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:23:09,100][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:23:09,598][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:23:10,097][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:23:10,595][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:23:11,095][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:23:11,591][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:23:12,086][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:23:12,581][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:23:13,079][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:23:13,583][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:23:14,081][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:23:14,575][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:23:15,070][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:23:15,563][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:23:16,061][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:23:16,559][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:23:17,055][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:23:17,552][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:23:18,053][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:23:18,561][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:23:19,064][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:23:19,573][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:23:20,080][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:23:20,580][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:23:21,079][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:23:21,576][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:23:22,072][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:23:22,569][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:23:23,066][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:23:23,563][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:23:24,060][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:23:24,562][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:23:25,062][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:23:25,562][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:23:26,061][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:23:26,560][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:23:27,058][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:23:27,557][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:23:28,055][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:23:28,551][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:23:29,046][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:23:29,544][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:23:30,041][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:23:30,537][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:23:31,034][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:23:31,539][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:23:32,038][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9828 tokens. +[2026-03-26 08:23:32,902][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.02%, Current % of VRAM taken: 60.44%, Block Peak % of device VRAM: 62.12%, ΔTime: 00:00:32 +[2026-03-26 08:23:33,676][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:23:33,679][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:23:33,691][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:23:34,423][__main__][INFO] - Iteration 676 took 1m 1s (41.66% Gen, 57.15% Train). Generation: 25s, Training: 35s. Estimated remaining time: 40h 58m 36s. Estimated total time: 51h 21m 22s. Time estimates for 10 more iterations: 10m 16s, 100 more iterations: 1h 42m 42s, 500 more iterations: 8h 33m 33s. +[2026-03-26 08:23:34,425][__main__][INFO] - Starting iteration 676. +[2026-03-26 08:23:34,826][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 67 and human policies 1. +[2026-03-26 08:23:34,827][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:24:06,223][__main__][INFO] - Number of regex retries in iteration 676: 0 +[2026-03-26 08:24:06,224][__main__][INFO] - agents played in iteration 676 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:24:07,005][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:24:07,028][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:24:07,052][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:24:07,076][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:24:07,077][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:24:07,077][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:24:07,959][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:24:08,416][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:24:08,920][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:24:09,420][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:24:09,922][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:24:10,422][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:24:10,929][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:24:11,432][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:24:11,936][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:24:12,434][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:24:12,925][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:24:13,427][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:24:13,924][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:24:14,425][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:24:14,927][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:24:15,428][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:24:15,927][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:24:16,426][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:24:16,926][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:24:17,425][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:24:17,930][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:24:18,430][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:24:18,930][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:24:19,434][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:24:19,934][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:24:20,435][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:24:20,930][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:24:21,428][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:24:21,934][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:24:22,440][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:24:22,948][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:24:23,442][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:24:23,940][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:24:24,443][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:24:24,945][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:24:25,447][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:24:25,943][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:24:26,444][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:24:26,947][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:24:27,451][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:24:27,954][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:24:28,453][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:24:28,951][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:24:29,447][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:24:29,945][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:24:30,442][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:24:30,947][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:24:31,448][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:24:31,946][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:24:32,443][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:24:32,942][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:24:33,439][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:24:33,937][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:24:34,435][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:24:34,934][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:24:35,433][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:24:35,931][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:24:36,432][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:24:36,928][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:24:37,423][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:24:37,918][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:24:38,419][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:24:38,916][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:24:39,414][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:24:39,912][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9911 tokens. +[2026-03-26 08:24:40,786][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.18%, ΔTime: 00:00:32 +[2026-03-26 08:24:41,570][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:24:41,573][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:24:41,574][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:24:42,319][__main__][INFO] - Iteration 677 took 1m 7s (46.52% Gen, 52.37% Train). Generation: 31s, Training: 35s. Estimated remaining time: 45h 50m 50s. Estimated total time: 56h 14m 44s. Time estimates for 10 more iterations: 11m 14s, 100 more iterations: 1h 52m 29s, 500 more iterations: 9h 22m 27s. +[2026-03-26 08:24:42,322][__main__][INFO] - Starting iteration 677. +[2026-03-26 08:24:42,722][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 67 and human policies 1. +[2026-03-26 08:24:42,723][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:25:03,850][__main__][INFO] - Number of regex retries in iteration 677: 0 +[2026-03-26 08:25:03,851][__main__][INFO] - agents played in iteration 677 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:25:04,616][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:25:04,640][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:25:04,664][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:25:04,688][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:25:04,688][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:25:04,689][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:25:05,541][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:25:05,998][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:25:06,509][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:25:07,009][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:25:07,510][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:25:08,014][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:25:08,517][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:25:09,019][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:25:09,524][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:25:10,025][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:25:10,529][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:25:11,029][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:25:11,527][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:25:12,033][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:25:12,531][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:25:13,030][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:25:13,526][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:25:14,024][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:25:14,519][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:25:15,015][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:25:15,519][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:25:16,015][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:25:16,514][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:25:17,011][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:25:17,506][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:25:18,002][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:25:18,497][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:25:18,995][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:25:19,497][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:25:19,995][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:25:20,493][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:25:20,989][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:25:21,489][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:25:21,985][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:25:22,482][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:25:22,979][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:25:23,475][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:25:23,972][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:25:24,466][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:25:24,961][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:25:25,457][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:25:25,955][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:25:26,453][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:25:26,951][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:25:27,449][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:25:27,946][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:25:28,449][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:25:28,949][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:25:29,446][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:25:29,943][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:25:30,440][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:25:30,939][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:25:31,436][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:25:31,938][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:25:32,439][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:25:32,945][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:25:33,443][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:25:33,941][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:25:34,440][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:25:34,938][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:25:35,439][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:25:35,937][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:25:36,436][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:25:36,937][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:25:37,438][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9881 tokens. +[2026-03-26 08:25:38,296][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.21%, Current % of VRAM taken: 60.64%, Block Peak % of device VRAM: 62.13%, ΔTime: 00:00:32 +[2026-03-26 08:25:39,057][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:25:39,060][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:25:39,061][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:25:39,790][__main__][INFO] - Iteration 678 took 57s (37.02% Gen, 61.70% Train). Generation: 21s, Training: 35s. Estimated remaining time: 37h 8m 36s. Estimated total time: 47h 33m 27s. Time estimates for 10 more iterations: 9m 30s, 100 more iterations: 1h 35m 6s, 500 more iterations: 7h 55m 34s. +[2026-03-26 08:25:39,795][__main__][INFO] - Starting iteration 678. +[2026-03-26 08:25:40,194][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 67 and human policies 1. +[2026-03-26 08:25:40,195][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:25:59,900][__main__][INFO] - Number of regex retries in iteration 678: 0 +[2026-03-26 08:25:59,900][__main__][INFO] - agents played in iteration 678 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:26:00,695][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:26:00,719][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:26:00,743][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:26:00,767][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:26:00,767][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:26:00,768][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:26:01,615][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:26:02,069][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:26:02,572][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:26:03,077][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:26:03,578][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:26:04,078][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:26:04,577][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:26:05,074][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:26:05,576][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:26:06,073][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:26:06,577][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:26:07,076][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:26:07,573][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:26:08,070][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:26:08,567][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:26:09,063][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:26:09,560][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:26:10,060][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:26:10,555][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:26:11,055][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:26:11,549][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:26:12,051][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:26:12,549][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:26:13,044][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:26:13,552][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:26:14,056][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:26:14,555][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:26:15,054][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:26:15,550][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:26:16,046][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:26:16,543][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:26:17,042][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:26:17,540][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:26:18,037][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:26:18,539][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:26:19,042][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:26:19,537][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:26:20,036][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:26:20,534][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:26:21,031][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:26:21,529][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:26:22,026][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:26:22,524][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:26:23,022][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:26:23,519][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:26:24,016][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:26:24,513][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:26:25,010][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:26:25,507][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:26:26,002][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:26:26,499][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:26:26,998][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:26:27,495][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:26:27,993][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:26:28,489][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:26:28,986][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:26:29,482][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:26:29,984][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:26:30,484][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:26:30,984][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:26:31,483][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:26:31,983][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:26:32,482][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:26:32,982][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:26:33,482][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9908 tokens. +[2026-03-26 08:26:34,344][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.16%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.03%, ΔTime: 00:00:32 +[2026-03-26 08:26:35,105][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:26:35,107][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:26:35,109][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:26:35,903][__main__][INFO] - Iteration 679 took 55s (35.37% Gen, 63.20% Train). Generation: 19s, Training: 35s. Estimated remaining time: 35h 59m 42s. Estimated total time: 46h 25m 29s. Time estimates for 10 more iterations: 9m 17s, 100 more iterations: 1h 32m 50s, 500 more iterations: 7h 44m 14s. +[2026-03-26 08:26:35,905][__main__][INFO] - Starting iteration 679. +[2026-03-26 08:26:36,307][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 67 and human policies 1. +[2026-03-26 08:26:36,308][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:26:41,027][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:26:45,477][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:26:50,981][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:26:52,484][__main__][INFO] - Number of regex retries in iteration 679: 3 +[2026-03-26 08:26:52,485][__main__][INFO] - agents played in iteration 679 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:26:53,248][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:26:53,272][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:26:53,296][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:26:53,319][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:26:53,320][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:26:53,320][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:26:54,166][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:26:54,623][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:26:55,125][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:26:55,627][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:26:56,127][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:26:56,632][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:26:57,133][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:26:57,636][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:26:58,137][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:26:58,631][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:26:59,123][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:26:59,615][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:27:00,107][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:27:00,604][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:27:01,103][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:27:01,612][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:27:02,110][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:27:02,608][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:27:03,105][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:27:03,598][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:27:04,096][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:27:04,595][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:27:05,093][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:27:05,596][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:27:06,093][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:27:06,599][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:27:07,101][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:27:07,599][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:27:08,102][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:27:08,600][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:27:09,101][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:27:09,598][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:27:10,097][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:27:10,604][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:27:11,105][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:27:11,606][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:27:12,103][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:27:12,601][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:27:13,100][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:27:13,599][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:27:14,098][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:27:14,600][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:27:15,120][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:27:15,619][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:27:16,115][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:27:16,616][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:27:17,113][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:27:17,609][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:27:18,103][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:27:18,601][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:27:19,099][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:27:19,597][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:27:20,097][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:27:20,596][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:27:21,094][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:27:21,596][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:27:22,093][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:27:22,591][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:27:23,085][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:27:23,583][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:27:24,084][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:27:24,579][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:27:25,076][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:27:25,569][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:27:26,066][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9873 tokens. +[2026-03-26 08:27:26,929][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.17%, ΔTime: 00:00:32 +[2026-03-26 08:27:27,692][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:27:27,695][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:27:27,696][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:27:28,467][__main__][INFO] - Iteration 680 took 52s (31.01% Gen, 67.50% Train). Generation: 16s, Training: 35s. Estimated remaining time: 33h 1m 21s. Estimated total time: 43h 28m 0s. Time estimates for 10 more iterations: 8m 41s, 100 more iterations: 1h 26m 56s, 500 more iterations: 7h 14m 40s. +[2026-03-26 08:27:28,469][__main__][INFO] - Starting iteration 680. +[2026-03-26 08:27:28,869][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 67 and human policies 1. +[2026-03-26 08:27:28,869][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:27:40,882][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:27:44,024][__main__][INFO] - Number of regex retries in iteration 680: 1 +[2026-03-26 08:27:44,025][__main__][INFO] - agents played in iteration 680 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:27:44,908][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:27:44,932][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:27:44,956][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:27:44,980][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:27:44,981][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:27:44,982][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:27:45,854][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:27:46,315][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:27:46,818][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:27:47,317][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:27:47,811][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:27:48,308][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:27:48,811][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:27:49,311][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:27:49,810][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:27:50,315][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:27:50,814][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:27:51,313][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:27:51,809][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:27:52,315][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:27:52,815][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:27:53,315][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:27:53,811][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:27:54,310][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:27:54,833][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:27:55,335][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:27:55,837][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:27:56,346][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:27:56,846][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:27:57,346][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:27:57,844][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:27:58,337][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:27:58,835][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:27:59,336][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:27:59,829][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:28:00,329][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:28:00,826][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:28:01,322][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:28:01,816][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:28:02,315][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:28:02,816][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:28:03,315][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:28:03,818][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:28:04,314][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:28:04,813][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:28:05,315][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:28:05,811][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:28:06,308][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:28:06,804][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:28:07,299][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:28:07,795][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:28:08,293][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:28:08,791][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:28:09,292][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:28:09,791][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:28:10,283][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:28:10,781][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:28:11,275][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:28:11,770][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:28:12,267][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:28:12,763][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:28:13,261][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:28:13,752][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:28:14,250][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:28:14,748][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:28:15,246][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:28:15,741][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:28:16,238][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:28:16,736][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:28:17,233][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:28:17,735][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9856 tokens. +[2026-03-26 08:28:18,608][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.10%, ΔTime: 00:00:32 +[2026-03-26 08:28:19,370][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:28:19,373][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:28:19,376][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:28:20,922][__main__][INFO] - Iteration 681 took 52s (29.12% Gen, 67.91% Train). Generation: 15s, Training: 35s. Estimated remaining time: 32h 55m 9s. Estimated total time: 43h 22m 42s. Time estimates for 10 more iterations: 8m 40s, 100 more iterations: 1h 26m 45s, 500 more iterations: 7h 13m 47s. +[2026-03-26 08:28:20,925][__main__][INFO] - Starting iteration 681. +[2026-03-26 08:28:21,325][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 68 and human policies 1. +[2026-03-26 08:28:21,326][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:28:30,089][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:28:40,922][__main__][INFO] - Number of regex retries in iteration 681: 1 +[2026-03-26 08:28:40,923][__main__][INFO] - agents played in iteration 681 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:28:41,708][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:28:41,732][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:28:41,755][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:28:41,778][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:28:41,779][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:28:41,779][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:28:42,628][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:28:43,076][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:28:43,572][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:28:44,064][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:28:44,555][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:28:45,051][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:28:45,548][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:28:46,040][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:28:46,531][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:28:47,022][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:28:47,522][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:28:48,020][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:28:48,541][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:28:49,041][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:28:49,538][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:28:50,037][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:28:50,531][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:28:51,027][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:28:51,526][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:28:52,024][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:28:52,520][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:28:53,021][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:28:53,522][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:28:54,019][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:28:54,516][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:28:55,014][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:28:55,509][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:28:56,005][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:28:56,501][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:28:56,999][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:28:57,496][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:28:57,994][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:28:58,493][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:28:58,991][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:28:59,488][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:28:59,984][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:29:00,480][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:29:00,976][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:29:01,473][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:29:01,968][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:29:02,465][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:29:02,958][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:29:03,456][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:29:03,951][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:29:04,448][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:29:04,942][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:29:05,437][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:29:05,931][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:29:06,427][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:29:06,924][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:29:07,425][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:29:07,920][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:29:08,415][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:29:08,915][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:29:09,409][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:29:09,906][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:29:10,401][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:29:10,899][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:29:11,397][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:29:11,898][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:29:12,395][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:29:12,894][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:29:13,393][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:29:13,896][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:29:14,393][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9798 tokens. +[2026-03-26 08:29:15,277][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 62.10%, ΔTime: 00:00:32 +[2026-03-26 08:29:16,035][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:29:16,037][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:29:16,039][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:29:16,852][__main__][INFO] - Iteration 682 took 55s (35.29% Gen, 63.24% Train). Generation: 19s, Training: 35s. Estimated remaining time: 35h 47m 54s. Estimated total time: 46h 16m 22s. Time estimates for 10 more iterations: 9m 15s, 100 more iterations: 1h 32m 32s, 500 more iterations: 7h 42m 43s. +[2026-03-26 08:29:16,854][__main__][INFO] - Starting iteration 682. +[2026-03-26 08:29:17,281][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 68 and human policies 1. +[2026-03-26 08:29:17,282][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:29:37,518][__main__][INFO] - Number of regex retries in iteration 682: 0 +[2026-03-26 08:29:37,519][__main__][INFO] - agents played in iteration 682 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:29:38,312][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:29:38,335][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:29:38,359][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:29:38,382][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:29:38,383][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:29:38,383][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:29:39,223][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:29:39,677][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:29:40,181][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:29:40,680][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:29:41,179][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:29:41,675][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:29:42,175][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:29:42,677][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:29:43,177][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:29:43,675][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:29:44,179][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:29:44,683][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:29:45,316][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:29:45,817][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:29:46,315][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:29:46,814][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:29:47,314][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:29:47,814][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:29:48,314][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:29:48,812][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:29:49,314][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:29:49,813][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:29:50,312][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:29:50,810][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:29:51,307][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:29:51,801][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:29:52,295][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:29:52,795][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:29:53,291][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:29:53,785][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:29:54,280][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:29:54,772][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:29:55,266][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:29:55,762][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:29:56,259][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:29:56,757][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:29:57,255][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:29:57,759][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:29:58,260][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:29:58,758][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:29:59,259][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:29:59,761][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:30:00,260][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:30:00,757][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:30:01,257][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:30:01,754][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:30:02,252][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:30:02,757][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:30:03,252][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:30:03,749][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:30:04,275][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:30:04,777][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:30:05,277][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:30:05,776][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:30:06,275][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:30:06,774][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:30:07,274][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:30:07,770][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:30:08,272][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:30:08,768][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:30:09,263][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:30:09,758][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:30:10,256][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:30:10,752][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:30:11,248][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9875 tokens. +[2026-03-26 08:30:12,118][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.03%, Current % of VRAM taken: 60.46%, Block Peak % of device VRAM: 62.13%, ΔTime: 00:00:32 +[2026-03-26 08:30:12,921][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:30:12,923][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:30:12,925][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:30:13,647][__main__][INFO] - Iteration 683 took 56s (35.90% Gen, 62.81% Train). Generation: 20s, Training: 35s. Estimated remaining time: 36h 28m 54s. Estimated total time: 46h 58m 19s. Time estimates for 10 more iterations: 9m 23s, 100 more iterations: 1h 33m 56s, 500 more iterations: 7h 49m 43s. +[2026-03-26 08:30:13,649][__main__][INFO] - Starting iteration 683. +[2026-03-26 08:30:14,049][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 68 and human policies 1. +[2026-03-26 08:30:14,050][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:30:20,094][mllm.models.large_language_model_local][WARNING] - Response Proposition: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:30:20,255][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:30:26,190][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:30:30,812][__main__][INFO] - Number of regex retries in iteration 683: 3 +[2026-03-26 08:30:30,813][__main__][INFO] - agents played in iteration 683 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:30:31,627][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:30:31,652][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:30:31,677][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:30:31,701][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:30:31,701][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:30:31,702][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:30:32,535][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:30:32,989][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:30:33,493][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:30:33,990][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:30:34,490][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:30:34,987][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:30:35,484][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:30:35,983][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:30:36,481][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:30:36,982][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:30:37,487][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:30:37,985][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:30:38,485][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:30:38,986][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:30:39,486][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:30:39,984][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:30:40,483][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:30:40,980][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:30:41,478][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:30:41,980][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:30:42,482][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:30:42,980][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:30:43,479][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:30:43,977][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:30:44,475][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:30:44,975][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:30:45,472][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:30:45,968][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:30:46,465][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:30:46,961][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:30:47,456][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:30:47,954][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:30:48,452][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:30:48,949][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:30:49,442][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:30:49,937][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:30:50,436][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:30:50,934][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:30:51,433][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:30:51,927][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:30:52,424][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:30:52,916][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:30:53,408][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:30:53,899][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:30:54,391][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:30:54,881][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:30:55,373][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:30:55,863][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:30:56,358][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:30:56,858][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:30:57,355][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:30:57,851][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:30:58,347][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:30:58,845][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:30:59,343][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:30:59,842][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:31:00,340][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:31:00,837][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:31:01,334][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:31:01,830][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:31:02,327][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:31:02,823][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:31:03,320][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:31:03,817][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:31:04,314][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9826 tokens. +[2026-03-26 08:31:05,183][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.03%, Current % of VRAM taken: 60.45%, Block Peak % of device VRAM: 61.88%, ΔTime: 00:00:32 +[2026-03-26 08:31:05,951][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:31:05,954][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:31:05,955][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:31:06,679][__main__][INFO] - Iteration 684 took 52s (31.85% Gen, 66.77% Train). Generation: 16s, Training: 35s. Estimated remaining time: 33h 21m 12s. Estimated total time: 43h 51m 30s. Time estimates for 10 more iterations: 8m 46s, 100 more iterations: 1h 27m 43s, 500 more iterations: 7h 18m 35s. +[2026-03-26 08:31:06,681][__main__][INFO] - Starting iteration 684. +[2026-03-26 08:31:07,081][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 68 and human policies 1. +[2026-03-26 08:31:07,082][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:31:21,282][__main__][INFO] - Number of regex retries in iteration 684: 0 +[2026-03-26 08:31:21,283][__main__][INFO] - agents played in iteration 684 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:31:22,063][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:31:22,086][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:31:22,110][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:31:22,133][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:31:22,133][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:31:22,134][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:31:22,989][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:31:23,445][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:31:23,947][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:31:24,451][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:31:24,946][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:31:25,443][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:31:25,939][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:31:26,440][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:31:26,937][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:31:27,438][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:31:27,938][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:31:28,438][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:31:28,939][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:31:29,444][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:31:29,944][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:31:30,446][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:31:30,949][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:31:31,449][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:31:31,955][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:31:32,455][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:31:32,951][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:31:33,448][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:31:33,945][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:31:34,444][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:31:34,942][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:31:35,439][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:31:35,941][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:31:36,440][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:31:36,936][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:31:37,434][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:31:37,933][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:31:38,428][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:31:38,924][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:31:39,422][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:31:39,919][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:31:40,416][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:31:40,913][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:31:41,414][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:31:41,911][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:31:42,408][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:31:42,905][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:31:43,401][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:31:43,898][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:31:44,396][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:31:44,892][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:31:45,389][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:31:45,885][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:31:46,381][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:31:46,877][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:31:47,370][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:31:47,873][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:31:48,375][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:31:48,868][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:31:49,374][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:31:49,867][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:31:50,361][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:31:50,852][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:31:51,348][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:31:51,845][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:31:52,343][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:31:52,839][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:31:53,333][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:31:53,830][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:31:54,326][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:31:54,825][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9810 tokens. +[2026-03-26 08:31:55,720][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.09%, ΔTime: 00:00:32 +[2026-03-26 08:31:56,473][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:31:56,475][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:31:56,477][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:31:57,312][__main__][INFO] - Iteration 685 took 50s (28.27% Gen, 70.06% Train). Generation: 14s, Training: 35s. Estimated remaining time: 31h 20m 25s. Estimated total time: 41h 51m 34s. Time estimates for 10 more iterations: 8m 22s, 100 more iterations: 1h 23m 43s, 500 more iterations: 6h 58m 35s. +[2026-03-26 08:31:57,316][__main__][INFO] - Starting iteration 685. +[2026-03-26 08:31:57,718][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 68 and human policies 1. +[2026-03-26 08:31:57,719][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:32:18,237][__main__][INFO] - Number of regex retries in iteration 685: 0 +[2026-03-26 08:32:18,238][__main__][INFO] - agents played in iteration 685 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:32:19,056][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:32:19,079][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:32:19,103][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:32:19,126][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:32:19,127][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:32:19,127][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:32:19,996][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:32:20,452][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:32:20,960][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:32:21,456][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:32:21,958][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:32:22,459][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:32:22,961][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:32:23,461][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:32:23,957][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:32:24,454][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:32:24,952][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:32:25,450][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:32:25,947][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:32:26,444][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:32:26,942][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:32:27,440][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:32:27,940][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:32:28,438][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:32:28,937][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:32:29,434][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:32:29,932][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:32:30,430][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:32:30,928][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:32:31,436][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:32:31,935][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:32:32,435][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:32:32,934][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:32:33,432][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:32:33,931][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:32:34,430][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:32:34,929][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:32:35,428][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:32:35,926][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:32:36,424][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:32:36,923][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:32:37,421][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:32:37,924][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:32:38,422][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:32:38,920][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:32:39,419][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:32:39,918][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:32:40,417][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:32:40,916][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:32:41,414][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:32:41,912][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:32:42,408][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:32:42,905][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:32:43,401][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:32:43,898][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:32:44,396][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:32:44,897][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:32:45,396][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:32:45,894][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:32:46,393][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:32:46,891][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:32:47,395][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:32:47,896][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:32:48,396][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:32:48,892][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:32:49,386][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:32:49,883][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:32:50,379][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:32:50,875][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:32:51,371][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:32:51,864][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9973 tokens. +[2026-03-26 08:32:52,754][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.97%, Current % of VRAM taken: 60.40%, Block Peak % of device VRAM: 62.19%, ΔTime: 00:00:32 +[2026-03-26 08:32:53,522][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:32:53,524][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:32:53,526][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:32:54,411][__main__][INFO] - Iteration 686 took 56s (36.19% Gen, 62.24% Train). Generation: 20s, Training: 35s. Estimated remaining time: 36h 42m 36s. Estimated total time: 47h 14m 41s. Time estimates for 10 more iterations: 9m 26s, 100 more iterations: 1h 34m 29s, 500 more iterations: 7h 52m 26s. +[2026-03-26 08:32:54,414][__main__][INFO] - Starting iteration 686. +[2026-03-26 08:32:54,824][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 68 and human policies 1. +[2026-03-26 08:32:54,824][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:32:58,929][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:32:59,116][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:33:08,778][__main__][INFO] - Number of regex retries in iteration 686: 2 +[2026-03-26 08:33:08,779][__main__][INFO] - agents played in iteration 686 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:33:09,555][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:33:09,580][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:33:09,604][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:33:09,628][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:33:09,628][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:33:09,629][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:33:10,480][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:33:10,932][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:33:11,436][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:33:11,931][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:33:12,435][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:33:12,933][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:33:13,436][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:33:13,935][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:33:14,430][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:33:14,928][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:33:15,425][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:33:15,922][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:33:16,423][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:33:16,924][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:33:17,422][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:33:17,920][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:33:18,417][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:33:18,919][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:33:19,418][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:33:19,916][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:33:20,419][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:33:20,918][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:33:21,416][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:33:21,914][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:33:22,412][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:33:22,910][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:33:23,412][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:33:23,916][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:33:24,412][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:33:24,909][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:33:25,407][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:33:25,899][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:33:26,425][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:33:26,921][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:33:27,419][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:33:27,915][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:33:28,414][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:33:28,910][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:33:29,409][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:33:29,912][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:33:30,410][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:33:30,907][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:33:31,405][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:33:31,903][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:33:32,402][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:33:32,900][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:33:33,398][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:33:33,897][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:33:34,396][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:33:34,894][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:33:35,392][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:33:35,887][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:33:36,383][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:33:36,881][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:33:37,381][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:33:37,882][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:33:38,379][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:33:38,877][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:33:39,377][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:33:39,875][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:33:40,374][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:33:40,872][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:33:41,370][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:33:41,868][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:33:42,366][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9869 tokens. +[2026-03-26 08:33:43,242][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 61.83%, ΔTime: 00:00:32 +[2026-03-26 08:33:44,005][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:33:44,008][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:33:44,009][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:33:44,748][__main__][INFO] - Iteration 687 took 49s (27.95% Gen, 70.57% Train). Generation: 13s, Training: 35s. Estimated remaining time: 31h 3m 19s. Estimated total time: 41h 36m 15s. Time estimates for 10 more iterations: 8m 19s, 100 more iterations: 1h 23m 12s, 500 more iterations: 6h 56m 2s. +[2026-03-26 08:33:44,750][__main__][INFO] - Starting iteration 687. +[2026-03-26 08:33:45,150][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 68 and human policies 1. +[2026-03-26 08:33:45,151][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:34:03,200][__main__][INFO] - Number of regex retries in iteration 687: 0 +[2026-03-26 08:34:03,201][__main__][INFO] - agents played in iteration 687 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:34:03,968][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:34:03,992][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:34:04,016][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:34:04,040][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:34:04,040][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:34:04,041][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:34:04,907][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:34:05,362][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:34:05,868][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:34:06,374][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:34:06,877][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:34:07,378][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:34:07,879][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:34:08,382][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:34:08,878][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:34:09,377][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:34:09,880][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:34:10,387][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:34:10,884][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:34:11,382][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:34:11,880][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:34:12,377][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:34:12,873][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:34:13,369][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:34:13,867][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:34:14,363][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:34:14,858][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:34:15,356][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:34:15,852][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:34:16,350][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:34:16,847][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:34:17,346][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:34:17,850][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:34:18,349][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:34:18,852][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:34:19,347][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:34:19,842][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:34:20,341][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:34:20,837][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:34:21,338][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:34:21,837][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:34:22,338][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:34:22,839][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:34:23,340][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:34:23,840][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:34:24,343][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:34:24,844][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:34:25,342][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:34:25,839][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:34:26,333][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:34:26,831][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:34:27,331][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:34:27,830][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:34:28,323][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:34:28,822][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:34:29,319][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:34:29,821][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:34:30,317][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:34:30,819][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:34:31,320][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:34:31,826][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:34:32,325][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:34:32,846][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:34:33,346][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:34:33,840][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:34:34,340][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:34:34,839][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:34:35,334][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:34:35,833][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:34:36,332][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:34:36,831][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9818 tokens. +[2026-03-26 08:34:37,719][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.03%, ΔTime: 00:00:32 +[2026-03-26 08:34:38,492][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:34:38,494][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:34:38,496][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:34:39,252][__main__][INFO] - Iteration 688 took 54s (33.36% Gen, 65.24% Train). Generation: 18s, Training: 35s. Estimated remaining time: 34h 31m 19s. Estimated total time: 45h 5m 9s. Time estimates for 10 more iterations: 9m 1s, 100 more iterations: 1h 30m 10s, 500 more iterations: 7h 30m 51s. +[2026-03-26 08:34:39,255][__main__][INFO] - Starting iteration 688. +[2026-03-26 08:34:39,679][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 68 and human policies 1. +[2026-03-26 08:34:39,680][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:34:57,021][__main__][INFO] - Number of regex retries in iteration 688: 0 +[2026-03-26 08:34:57,022][__main__][INFO] - agents played in iteration 688 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:34:57,821][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:34:57,845][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:34:57,868][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:34:57,891][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:34:57,892][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:34:57,893][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:34:58,739][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:34:59,192][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:34:59,696][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:35:00,194][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:35:00,689][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:35:01,188][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:35:01,686][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:35:02,188][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:35:02,693][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:35:03,193][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:35:03,691][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:35:04,188][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:35:04,685][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:35:05,181][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:35:05,676][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:35:06,174][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:35:06,673][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:35:07,173][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:35:07,668][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:35:08,165][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:35:08,663][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:35:09,160][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:35:09,658][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:35:10,155][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:35:10,654][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:35:11,154][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:35:11,653][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:35:12,152][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:35:12,653][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:35:13,154][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:35:13,652][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:35:14,150][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:35:14,648][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:35:15,154][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:35:15,655][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:35:16,153][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:35:16,652][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:35:17,150][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:35:17,648][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:35:18,142][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:35:18,640][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:35:19,136][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:35:19,633][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:35:20,131][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:35:20,629][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:35:21,126][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:35:21,624][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:35:22,121][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:35:22,617][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:35:23,112][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:35:23,606][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:35:24,100][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:35:24,595][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:35:25,090][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:35:25,583][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:35:26,077][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:35:26,571][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:35:27,068][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:35:27,564][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:35:28,060][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:35:28,557][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:35:29,054][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:35:29,554][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:35:30,051][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:35:30,551][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9834 tokens. +[2026-03-26 08:35:31,434][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.03%, Current % of VRAM taken: 60.46%, Block Peak % of device VRAM: 62.14%, ΔTime: 00:00:32 +[2026-03-26 08:35:32,192][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:35:32,195][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:35:32,197][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:35:33,048][__main__][INFO] - Iteration 689 took 53s (32.49% Gen, 65.91% Train). Generation: 17s, Training: 35s. Estimated remaining time: 33h 53m 44s. Estimated total time: 44h 28m 29s. Time estimates for 10 more iterations: 8m 53s, 100 more iterations: 1h 28m 56s, 500 more iterations: 7h 24m 44s. +[2026-03-26 08:35:33,051][__main__][INFO] - Starting iteration 689. +[2026-03-26 08:35:33,456][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 68 and human policies 1. +[2026-03-26 08:35:33,457][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:35:38,049][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:35:48,627][__main__][INFO] - Number of regex retries in iteration 689: 1 +[2026-03-26 08:35:48,628][__main__][INFO] - agents played in iteration 689 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:35:49,394][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:35:49,418][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:35:49,444][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:35:49,468][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:35:49,469][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:35:49,469][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:35:50,456][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:35:50,912][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:35:51,419][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:35:51,919][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:35:52,421][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:35:52,923][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:35:53,422][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:35:53,923][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:35:54,422][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:35:54,926][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:35:55,423][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:35:55,925][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:35:56,422][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:35:56,921][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:35:57,416][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:35:57,916][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:35:58,411][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:35:58,913][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:35:59,410][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:35:59,916][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:36:00,415][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:36:00,917][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:36:01,417][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:36:01,917][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:36:02,422][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:36:02,918][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:36:03,414][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:36:03,915][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:36:04,419][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:36:04,914][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:36:05,415][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:36:05,911][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:36:06,410][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:36:06,908][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:36:07,407][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:36:07,906][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:36:08,404][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:36:08,904][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:36:09,403][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:36:09,902][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:36:10,400][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:36:10,897][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:36:11,396][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:36:11,894][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:36:12,389][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:36:12,883][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:36:13,381][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:36:13,878][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:36:14,371][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:36:14,865][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:36:15,360][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:36:15,855][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:36:16,353][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:36:16,851][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:36:17,347][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:36:17,841][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:36:18,339][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:36:18,837][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:36:19,335][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:36:19,833][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:36:20,332][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:36:20,830][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:36:21,327][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:36:21,823][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:36:22,320][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9879 tokens. +[2026-03-26 08:36:23,213][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.04%, Current % of VRAM taken: 60.47%, Block Peak % of device VRAM: 62.14%, ΔTime: 00:00:32 +[2026-03-26 08:36:23,985][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:36:23,987][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:36:23,989][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:36:24,791][__main__][INFO] - Iteration 690 took 51s (29.55% Gen, 68.88% Train). Generation: 15s, Training: 35s. Estimated remaining time: 32h 11m 9s. Estimated total time: 42h 46m 45s. Time estimates for 10 more iterations: 8m 33s, 100 more iterations: 1h 25m 33s, 500 more iterations: 7h 7m 47s. +[2026-03-26 08:36:24,793][__main__][INFO] - Starting iteration 690. +[2026-03-26 08:36:25,247][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 68 and human policies 1. +[2026-03-26 08:36:25,248][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:36:40,863][__main__][INFO] - Number of regex retries in iteration 690: 0 +[2026-03-26 08:36:40,864][__main__][INFO] - agents played in iteration 690 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:36:41,654][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:36:41,678][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:36:41,702][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:36:41,726][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:36:41,726][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:36:41,727][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:36:42,556][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:36:43,011][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:36:43,517][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:36:44,025][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:36:44,522][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:36:45,028][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:36:45,532][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:36:46,031][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:36:46,529][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:36:47,031][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:36:47,529][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:36:48,036][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:36:48,537][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:36:49,035][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:36:49,533][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:36:50,031][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:36:50,528][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:36:51,027][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:36:51,525][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:36:52,024][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:36:52,524][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:36:53,023][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:36:53,521][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:36:54,019][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:36:54,517][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:36:55,016][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:36:55,510][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:36:56,009][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:36:56,507][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:36:57,004][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:36:57,501][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:36:58,000][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:36:58,497][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:36:58,996][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:36:59,495][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:36:59,994][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:37:00,494][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:37:00,993][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:37:01,497][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:37:01,995][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:37:02,496][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:37:02,999][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:37:03,498][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:37:04,019][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:37:04,523][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:37:05,022][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:37:05,523][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:37:06,022][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:37:06,522][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:37:07,024][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:37:07,523][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:37:08,021][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:37:08,520][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:37:09,016][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:37:09,515][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:37:10,016][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:37:10,514][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:37:11,011][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:37:11,507][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:37:12,005][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:37:12,503][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:37:13,003][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:37:13,501][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:37:13,998][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:37:14,497][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9982 tokens. +[2026-03-26 08:37:15,393][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.19%, ΔTime: 00:00:32 +[2026-03-26 08:37:16,153][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:37:16,155][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:37:16,159][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:37:17,465][__main__][INFO] - Iteration 691 took 52s (29.90% Gen, 67.59% Train). Generation: 15s, Training: 35s. Estimated remaining time: 32h 54m 27s. Estimated total time: 43h 30m 55s. Time estimates for 10 more iterations: 8m 42s, 100 more iterations: 1h 27m 1s, 500 more iterations: 7h 15m 9s. +[2026-03-26 08:37:17,468][__main__][INFO] - Starting iteration 691. +[2026-03-26 08:37:17,870][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 69 and human policies 1. +[2026-03-26 08:37:17,871][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:37:31,987][__main__][INFO] - Number of regex retries in iteration 691: 0 +[2026-03-26 08:37:31,988][__main__][INFO] - agents played in iteration 691 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:37:32,748][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:37:32,772][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:37:32,796][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:37:32,820][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:37:32,820][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:37:32,821][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:37:33,675][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:37:34,131][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:37:34,644][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:37:35,140][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:37:35,642][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:37:36,143][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:37:36,643][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:37:37,139][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:37:37,636][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:37:38,134][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:37:38,633][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:37:39,137][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:37:39,636][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:37:40,136][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:37:40,635][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:37:41,134][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:37:41,633][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:37:42,131][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:37:42,628][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:37:43,127][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:37:43,628][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:37:44,131][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:37:44,628][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:37:45,125][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:37:45,623][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:37:46,120][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:37:46,616][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:37:47,112][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:37:47,608][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:37:48,103][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:37:48,598][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:37:49,095][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:37:49,594][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:37:50,098][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:37:50,596][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:37:51,094][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:37:51,593][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:37:52,088][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:37:52,584][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:37:53,081][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:37:53,578][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:37:54,071][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:37:54,562][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:37:55,059][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:37:55,551][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:37:56,047][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:37:56,545][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:37:57,045][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:37:57,540][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:37:58,038][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:37:58,537][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:37:59,037][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:37:59,538][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:38:00,040][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:38:00,538][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:38:01,036][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:38:01,534][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:38:02,033][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:38:02,530][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:38:03,028][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:38:03,526][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:38:04,022][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:38:04,519][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:38:05,017][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:38:05,516][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9816 tokens. +[2026-03-26 08:38:06,400][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.03%, Current % of VRAM taken: 60.46%, Block Peak % of device VRAM: 61.90%, ΔTime: 00:00:32 +[2026-03-26 08:38:07,160][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:38:07,163][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:38:07,165][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:38:07,934][__main__][INFO] - Iteration 692 took 50s (28.20% Gen, 70.26% Train). Generation: 14s, Training: 35s. Estimated remaining time: 31h 5m 54s. Estimated total time: 41h 43m 14s. Time estimates for 10 more iterations: 8m 20s, 100 more iterations: 1h 23m 26s, 500 more iterations: 6h 57m 12s. +[2026-03-26 08:38:07,936][__main__][INFO] - Starting iteration 692. +[2026-03-26 08:38:08,336][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 69 and human policies 1. +[2026-03-26 08:38:08,337][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:38:15,146][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:38:24,674][__main__][INFO] - Number of regex retries in iteration 692: 1 +[2026-03-26 08:38:24,675][__main__][INFO] - agents played in iteration 692 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:38:25,438][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:38:25,462][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:38:25,486][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:38:25,510][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:38:25,511][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:38:25,511][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:38:26,357][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:38:26,812][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:38:27,315][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:38:27,823][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:38:28,326][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:38:28,824][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:38:29,321][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:38:29,817][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:38:30,314][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:38:30,812][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:38:31,307][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:38:31,805][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:38:32,305][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:38:32,807][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:38:33,306][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:38:33,802][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:38:34,300][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:38:34,801][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:38:35,300][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:38:35,800][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:38:36,300][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:38:36,800][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:38:37,300][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:38:37,800][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:38:38,299][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:38:38,802][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:38:39,304][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:38:39,801][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:38:40,299][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:38:40,795][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:38:41,294][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:38:41,788][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:38:42,282][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:38:42,783][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:38:43,279][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:38:43,776][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:38:44,272][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:38:44,769][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:38:45,265][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:38:45,761][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:38:46,253][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:38:46,748][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:38:47,244][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:38:47,741][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:38:48,241][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:38:48,739][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:38:49,238][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:38:49,737][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:38:50,236][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:38:50,729][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:38:51,226][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:38:51,724][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:38:52,220][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:38:52,716][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:38:53,211][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:38:53,703][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:38:54,198][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:38:54,697][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:38:55,195][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:38:55,695][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:38:56,194][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:38:56,693][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:38:57,191][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:38:57,693][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:38:58,221][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9847 tokens. +[2026-03-26 08:38:59,110][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.26%, Current % of VRAM taken: 60.68%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 08:38:59,873][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:38:59,875][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:38:59,877][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:39:00,727][__main__][INFO] - Iteration 693 took 52s (31.18% Gen, 67.19% Train). Generation: 16s, Training: 35s. Estimated remaining time: 33h 1m 21s. Estimated total time: 43h 39m 33s. Time estimates for 10 more iterations: 8m 43s, 100 more iterations: 1h 27m 19s, 500 more iterations: 7h 16m 35s. +[2026-03-26 08:39:00,730][__main__][INFO] - Starting iteration 693. +[2026-03-26 08:39:01,127][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 69 and human policies 1. +[2026-03-26 08:39:01,128][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:39:21,682][__main__][INFO] - Number of regex retries in iteration 693: 0 +[2026-03-26 08:39:21,683][__main__][INFO] - agents played in iteration 693 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:39:22,452][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:39:22,476][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:39:22,499][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:39:22,523][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.57%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:39:22,523][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:39:22,524][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:39:23,371][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:39:23,824][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:39:24,329][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:39:24,829][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:39:25,329][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:39:25,839][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:39:26,340][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:39:26,841][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:39:27,342][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:39:27,843][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:39:28,339][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:39:28,836][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:39:29,340][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:39:29,843][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:39:30,338][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:39:30,834][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:39:31,330][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:39:31,824][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:39:32,325][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:39:32,821][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:39:33,315][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:39:33,812][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:39:34,306][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:39:34,801][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:39:35,297][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:39:35,795][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:39:36,295][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:39:36,793][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:39:37,291][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:39:37,789][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:39:38,290][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:39:38,790][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:39:39,289][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:39:39,788][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:39:40,285][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:39:40,782][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:39:41,281][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:39:41,786][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:39:42,285][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:39:42,783][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:39:43,281][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:39:43,778][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:39:44,278][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:39:44,778][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:39:45,279][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:39:45,780][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:39:46,283][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:39:46,784][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:39:47,281][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:39:47,780][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:39:48,278][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:39:48,777][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:39:49,276][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:39:49,774][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:39:50,277][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:39:50,775][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:39:51,278][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:39:51,780][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:39:52,283][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:39:52,788][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:39:53,286][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:39:53,784][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:39:54,284][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:39:54,778][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:39:55,277][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9993 tokens. +[2026-03-26 08:39:56,161][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.14%, ΔTime: 00:00:32 +[2026-03-26 08:39:56,928][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:39:56,931][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:39:56,932][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:39:57,670][__main__][INFO] - Iteration 694 took 56s (36.35% Gen, 62.34% Train). Generation: 20s, Training: 35s. Estimated remaining time: 36h 28m 2s. Estimated total time: 47h 7m 11s. Time estimates for 10 more iterations: 9m 25s, 100 more iterations: 1h 34m 14s, 500 more iterations: 7h 51m 11s. +[2026-03-26 08:39:57,673][__main__][INFO] - Starting iteration 694. +[2026-03-26 08:39:58,090][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 69 and human policies 1. +[2026-03-26 08:39:58,091][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:40:14,011][__main__][INFO] - Number of regex retries in iteration 694: 0 +[2026-03-26 08:40:14,012][__main__][INFO] - agents played in iteration 694 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:40:14,800][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:40:14,824][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:40:14,848][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:40:14,871][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:40:14,872][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:40:14,872][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:40:15,733][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:40:16,188][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:40:16,690][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:40:17,187][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:40:17,684][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:40:18,185][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:40:18,683][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:40:19,182][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:40:19,685][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:40:20,183][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:40:20,682][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:40:21,181][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:40:21,680][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:40:22,176][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:40:22,674][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:40:23,177][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:40:23,675][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:40:24,174][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:40:24,672][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:40:25,174][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:40:25,676][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:40:26,174][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:40:26,671][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:40:27,171][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:40:27,670][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:40:28,168][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:40:28,666][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:40:29,162][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:40:29,657][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:40:30,157][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:40:30,660][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:40:31,157][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:40:31,654][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:40:32,152][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:40:32,653][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:40:33,154][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:40:33,652][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:40:34,149][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:40:34,648][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:40:35,149][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:40:35,647][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:40:36,142][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:40:36,637][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:40:37,129][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:40:37,620][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:40:38,114][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:40:38,609][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:40:39,104][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:40:39,598][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:40:40,096][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:40:40,593][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:40:41,091][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:40:41,589][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:40:42,087][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:40:42,584][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:40:43,080][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:40:43,579][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:40:44,076][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:40:44,572][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:40:45,072][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:40:45,568][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:40:46,065][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:40:46,562][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:40:47,062][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:40:47,562][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9876 tokens. +[2026-03-26 08:40:48,451][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.00%, Current % of VRAM taken: 60.43%, Block Peak % of device VRAM: 61.96%, ΔTime: 00:00:32 +[2026-03-26 08:40:49,215][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:40:49,217][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:40:49,218][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:40:49,991][__main__][INFO] - Iteration 695 took 51s (30.68% Gen, 67.83% Train). Generation: 15s, Training: 35s. Estimated remaining time: 32h 35m 4s. Estimated total time: 43h 15m 5s. Time estimates for 10 more iterations: 8m 39s, 100 more iterations: 1h 26m 30s, 500 more iterations: 7h 12m 30s. +[2026-03-26 08:40:49,994][__main__][INFO] - Starting iteration 695. +[2026-03-26 08:40:50,395][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 69 and human policies 1. +[2026-03-26 08:40:50,395][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:40:55,505][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:41:06,518][__main__][INFO] - Number of regex retries in iteration 695: 1 +[2026-03-26 08:41:06,519][__main__][INFO] - agents played in iteration 695 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:41:07,290][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:41:07,314][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:41:07,338][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:41:07,361][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:41:07,361][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:41:07,362][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:41:08,207][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:41:08,660][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:41:09,161][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:41:09,656][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:41:10,152][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:41:10,650][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:41:11,149][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:41:11,647][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:41:12,143][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:41:12,641][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:41:13,140][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:41:13,638][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:41:14,136][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:41:14,633][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:41:15,131][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:41:15,628][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:41:16,130][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:41:16,631][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:41:17,136][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:41:17,636][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:41:18,135][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:41:18,635][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:41:19,132][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:41:19,624][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:41:20,122][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:41:20,624][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:41:21,122][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:41:21,619][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:41:22,114][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:41:22,610][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:41:23,107][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:41:23,603][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:41:24,100][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:41:24,596][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:41:25,095][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:41:25,592][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:41:26,088][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:41:26,585][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:41:27,081][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:41:27,577][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:41:28,073][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:41:28,571][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:41:29,068][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:41:29,565][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:41:30,063][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:41:30,560][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:41:31,057][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:41:31,556][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:41:32,053][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:41:32,550][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:41:33,046][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:41:33,543][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:41:34,039][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:41:34,536][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:41:35,034][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:41:35,530][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:41:36,027][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:41:36,523][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:41:37,018][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:41:37,514][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:41:38,009][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:41:38,504][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:41:38,999][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:41:39,498][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:41:39,995][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9833 tokens. +[2026-03-26 08:41:40,876][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.98%, Current % of VRAM taken: 60.41%, Block Peak % of device VRAM: 61.86%, ΔTime: 00:00:32 +[2026-03-26 08:41:41,702][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:41:41,704][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:41:41,706][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:41:42,376][__main__][INFO] - Iteration 696 took 51s (31.02% Gen, 67.69% Train). Generation: 16s, Training: 35s. Estimated remaining time: 32h 38m 12s. Estimated total time: 43h 19m 6s. Time estimates for 10 more iterations: 8m 39s, 100 more iterations: 1h 26m 38s, 500 more iterations: 7h 13m 11s. +[2026-03-26 08:41:42,379][__main__][INFO] - Starting iteration 696. +[2026-03-26 08:41:42,778][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 69 and human policies 1. +[2026-03-26 08:41:42,778][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:41:58,437][__main__][INFO] - Number of regex retries in iteration 696: 0 +[2026-03-26 08:41:58,438][__main__][INFO] - agents played in iteration 696 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:41:59,204][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:41:59,227][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:41:59,251][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:41:59,274][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:41:59,274][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:41:59,275][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:42:00,117][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:42:00,568][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:42:01,069][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:42:01,566][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:42:02,060][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:42:02,556][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:42:03,054][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:42:03,552][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:42:04,049][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:42:04,554][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:42:05,059][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:42:05,564][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:42:06,067][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:42:06,573][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:42:07,078][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:42:07,594][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:42:08,093][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:42:08,591][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:42:09,092][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:42:09,594][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:42:10,092][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:42:10,594][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:42:11,095][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:42:11,597][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:42:12,092][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:42:12,589][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:42:13,091][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:42:13,592][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:42:14,093][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:42:14,593][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:42:15,095][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:42:15,595][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:42:16,095][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:42:16,591][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:42:17,084][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:42:17,582][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:42:18,086][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:42:18,589][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:42:19,091][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:42:19,587][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:42:20,087][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:42:20,585][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:42:21,082][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:42:21,581][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:42:22,081][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:42:22,574][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:42:23,075][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:42:23,577][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:42:24,076][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:42:24,577][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:42:25,078][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:42:25,579][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:42:26,078][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:42:26,579][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:42:27,076][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:42:27,578][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:42:28,076][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:42:28,574][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:42:29,072][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:42:29,573][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:42:30,078][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:42:30,578][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:42:31,078][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:42:31,575][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:42:32,072][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 10036 tokens. +[2026-03-26 08:42:33,073][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.03%, Current % of VRAM taken: 60.46%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 08:42:33,845][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:42:33,847][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:42:33,849][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:42:34,714][__main__][INFO] - Iteration 697 took 51s (30.15% Gen, 68.18% Train). Generation: 15s, Training: 35s. Estimated remaining time: 32h 35m 6s. Estimated total time: 43h 16m 52s. Time estimates for 10 more iterations: 8m 39s, 100 more iterations: 1h 26m 33s, 500 more iterations: 7h 12m 48s. +[2026-03-26 08:42:34,717][__main__][INFO] - Starting iteration 697. +[2026-03-26 08:42:35,118][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 69 and human policies 1. +[2026-03-26 08:42:35,119][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:42:51,394][__main__][INFO] - Number of regex retries in iteration 697: 0 +[2026-03-26 08:42:51,395][__main__][INFO] - agents played in iteration 697 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:42:52,166][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:42:52,190][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:42:52,213][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:42:52,236][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:42:52,237][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:42:52,238][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:42:53,080][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:42:53,536][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:42:54,040][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:42:54,541][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:42:55,043][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:42:55,545][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:42:56,044][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:42:56,544][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:42:57,043][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:42:57,540][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:42:58,038][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:42:58,535][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:42:59,034][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:42:59,529][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:43:00,025][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:43:00,521][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:43:01,022][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:43:01,525][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:43:02,024][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:43:02,522][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:43:03,020][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:43:03,519][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:43:04,017][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:43:04,517][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:43:05,016][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:43:05,515][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:43:06,015][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:43:06,520][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:43:07,027][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:43:07,531][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:43:08,030][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:43:08,533][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:43:09,033][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:43:09,535][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:43:10,037][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:43:10,538][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:43:11,040][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:43:11,541][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:43:12,041][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:43:12,542][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:43:13,042][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:43:13,545][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:43:14,045][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:43:14,546][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:43:15,041][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:43:15,541][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:43:16,043][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:43:16,548][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:43:17,050][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:43:17,554][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:43:18,055][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:43:18,554][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:43:19,054][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:43:19,553][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:43:20,053][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:43:20,552][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:43:21,050][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:43:21,546][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:43:22,043][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:43:22,540][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:43:23,037][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:43:23,540][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:43:24,038][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:43:24,534][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:43:25,035][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9950 tokens. +[2026-03-26 08:43:25,923][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.26%, ΔTime: 00:00:32 +[2026-03-26 08:43:26,690][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:43:26,693][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:43:26,694][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:43:27,419][__main__][INFO] - Iteration 698 took 52s (31.12% Gen, 67.49% Train). Generation: 16s, Training: 35s. Estimated remaining time: 32h 52m 25s. Estimated total time: 43h 35m 4s. Time estimates for 10 more iterations: 8m 43s, 100 more iterations: 1h 27m 10s, 500 more iterations: 7h 15m 50s. +[2026-03-26 08:43:27,421][__main__][INFO] - Starting iteration 698. +[2026-03-26 08:43:27,822][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 69 and human policies 1. +[2026-03-26 08:43:27,822][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:43:43,494][__main__][INFO] - Number of regex retries in iteration 698: 0 +[2026-03-26 08:43:43,495][__main__][INFO] - agents played in iteration 698 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:43:44,274][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:43:44,298][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:43:44,322][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:43:44,345][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:43:44,345][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:43:44,346][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:43:45,189][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:43:45,643][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:43:46,143][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:43:46,640][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:43:47,138][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:43:47,639][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:43:48,137][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:43:48,642][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:43:49,141][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:43:49,646][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:43:50,146][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:43:50,649][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:43:51,146][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:43:51,650][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:43:52,158][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:43:52,661][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:43:53,162][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:43:53,659][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:43:54,161][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:43:54,665][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:43:55,168][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:43:55,666][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:43:56,164][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:43:56,663][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:43:57,176][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:43:57,674][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:43:58,174][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:43:58,679][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:43:59,179][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:43:59,680][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:44:00,181][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:44:00,678][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:44:01,181][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:44:01,688][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:44:02,189][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:44:02,693][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:44:03,194][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:44:03,697][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:44:04,200][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:44:04,701][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:44:05,201][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:44:05,697][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:44:06,196][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:44:06,707][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:44:07,208][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:44:07,711][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:44:08,205][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:44:08,702][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:44:09,203][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:44:09,704][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:44:10,202][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:44:10,698][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:44:11,197][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:44:11,696][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:44:12,199][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:44:12,702][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:44:13,206][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:44:13,704][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:44:14,203][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:44:14,703][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:44:15,201][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:44:15,703][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:44:16,201][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:44:16,701][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:44:17,202][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9994 tokens. +[2026-03-26 08:44:18,091][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.00%, Current % of VRAM taken: 60.42%, Block Peak % of device VRAM: 62.09%, ΔTime: 00:00:32 +[2026-03-26 08:44:18,852][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:44:18,854][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:44:18,856][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:44:19,697][__main__][INFO] - Iteration 699 took 51s (30.21% Gen, 68.16% Train). Generation: 15s, Training: 35s. Estimated remaining time: 32h 30m 18s. Estimated total time: 43h 13m 49s. Time estimates for 10 more iterations: 8m 38s, 100 more iterations: 1h 26m 27s, 500 more iterations: 7h 12m 18s. +[2026-03-26 08:44:19,700][__main__][INFO] - Starting iteration 699. +[2026-03-26 08:44:20,109][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 69 and human policies 1. +[2026-03-26 08:44:20,110][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:44:36,134][__main__][INFO] - Number of regex retries in iteration 699: 0 +[2026-03-26 08:44:36,135][__main__][INFO] - agents played in iteration 699 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:44:36,900][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:44:36,924][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:44:36,948][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:44:36,971][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:44:36,972][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:44:36,972][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:44:37,829][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:44:38,282][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:44:38,784][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:44:39,282][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:44:39,780][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:44:40,278][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:44:40,777][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:44:41,275][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:44:41,773][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:44:42,269][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:44:42,766][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:44:43,263][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:44:43,764][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:44:44,269][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:44:44,766][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:44:45,272][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:44:45,777][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:44:46,279][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:44:46,776][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:44:47,275][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:44:47,775][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:44:48,276][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:44:48,771][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:44:49,268][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:44:49,766][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:44:50,274][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:44:50,780][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:44:51,285][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:44:51,785][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:44:52,284][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:44:52,787][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:44:53,288][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:44:53,790][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:44:54,294][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:44:54,804][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:44:55,307][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:44:55,807][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:44:56,311][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:44:56,811][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:44:57,308][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:44:57,807][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:44:58,301][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:44:58,795][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:44:59,288][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:44:59,784][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:45:00,279][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:45:00,774][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:45:01,267][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:45:01,765][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:45:02,263][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:45:02,758][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:45:03,256][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:45:03,752][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:45:04,248][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:45:04,744][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:45:05,240][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:45:05,738][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:45:06,241][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:45:06,745][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:45:07,243][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:45:07,741][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:45:08,241][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:45:08,743][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:45:09,242][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:45:09,739][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9861 tokens. +[2026-03-26 08:45:10,611][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.01%, Current % of VRAM taken: 60.44%, Block Peak % of device VRAM: 62.04%, ΔTime: 00:00:32 +[2026-03-26 08:45:11,351][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:45:11,353][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:45:11,355][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:45:13,117][__main__][INFO] - Iteration 700 took 53s (30.23% Gen, 66.44% Train). Generation: 16s, Training: 35s. Estimated remaining time: 33h 26m 1s. Estimated total time: 44h 10m 26s. Time estimates for 10 more iterations: 8m 50s, 100 more iterations: 1h 28m 20s, 500 more iterations: 7h 21m 44s. +[2026-03-26 08:45:13,119][__main__][INFO] - Starting iteration 700. +[2026-03-26 08:45:13,521][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 69 and human policies 1. +[2026-03-26 08:45:13,522][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:45:27,930][__main__][INFO] - Number of regex retries in iteration 700: 0 +[2026-03-26 08:45:27,931][__main__][INFO] - agents played in iteration 700 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:45:28,688][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:45:28,712][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:45:28,737][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:45:28,761][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:45:28,761][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:45:28,762][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:45:29,607][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:45:30,059][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:45:30,563][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:45:31,060][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:45:31,559][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:45:32,057][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:45:32,559][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:45:33,056][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:45:33,552][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:45:34,059][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:45:34,558][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:45:35,056][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:45:35,555][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:45:36,053][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:45:36,551][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:45:37,048][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:45:37,555][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:45:38,061][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:45:38,570][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:45:39,073][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:45:39,574][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:45:40,075][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:45:40,575][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:45:41,077][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:45:41,581][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:45:42,083][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:45:42,585][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:45:43,086][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:45:43,584][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:45:44,083][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:45:44,586][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:45:45,088][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:45:45,590][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:45:46,091][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:45:46,590][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:45:47,088][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:45:47,587][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:45:48,088][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:45:48,588][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:45:49,084][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:45:49,582][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:45:50,084][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:45:50,582][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:45:51,081][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:45:51,579][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:45:52,075][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:45:52,575][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:45:53,075][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:45:53,573][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:45:54,072][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:45:54,577][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:45:55,075][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:45:55,576][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:45:56,083][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:45:56,581][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:45:57,079][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:45:57,581][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:45:58,081][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:45:58,579][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:45:59,083][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:45:59,584][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:46:00,086][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:46:00,589][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:46:01,093][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:46:01,594][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9936 tokens. +[2026-03-26 08:46:02,473][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.19%, Current % of VRAM taken: 60.61%, Block Peak % of device VRAM: 62.30%, ΔTime: 00:00:32 +[2026-03-26 08:46:03,259][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:46:03,263][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:46:03,265][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:46:04,566][__main__][INFO] - Iteration 701 took 51s (28.23% Gen, 69.22% Train). Generation: 14s, Training: 35s. Estimated remaining time: 31h 47m 0s. Estimated total time: 42h 32m 16s. Time estimates for 10 more iterations: 8m 30s, 100 more iterations: 1h 25m 4s, 500 more iterations: 7h 5m 22s. +[2026-03-26 08:46:04,568][__main__][INFO] - Starting iteration 701. +[2026-03-26 08:46:04,974][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 70 and human policies 1. +[2026-03-26 08:46:04,974][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:46:20,836][__main__][INFO] - Number of regex retries in iteration 701: 0 +[2026-03-26 08:46:20,837][__main__][INFO] - agents played in iteration 701 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:46:21,598][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:46:21,621][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:46:21,645][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:46:21,668][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:46:21,669][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:46:21,670][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:46:22,511][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:46:22,963][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:46:23,461][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:46:23,953][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:46:24,457][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:46:24,954][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:46:25,451][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:46:25,943][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:46:26,439][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:46:26,938][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:46:27,441][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:46:27,940][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:46:28,440][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:46:28,943][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:46:29,446][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:46:29,948][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:46:30,448][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:46:30,947][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:46:31,445][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:46:31,939][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:46:32,431][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:46:32,926][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:46:33,425][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:46:33,926][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:46:34,427][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:46:34,939][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:46:35,447][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:46:35,951][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:46:36,456][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:46:36,961][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:46:37,463][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:46:37,969][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:46:38,475][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:46:38,976][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:46:39,478][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:46:39,983][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:46:40,486][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:46:40,984][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:46:41,482][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:46:41,980][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:46:42,480][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:46:43,000][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:46:43,502][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:46:43,999][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:46:44,498][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:46:44,998][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:46:45,497][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:46:45,998][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:46:46,504][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:46:47,005][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:46:47,505][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:46:48,007][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:46:48,507][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:46:49,008][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:46:49,511][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:46:50,014][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:46:50,518][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:46:51,016][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:46:51,511][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:46:52,006][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:46:52,504][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:46:53,000][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:46:53,499][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:46:54,001][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:46:54,499][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9894 tokens. +[2026-03-26 08:46:55,391][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.04%, Current % of VRAM taken: 60.47%, Block Peak % of device VRAM: 62.18%, ΔTime: 00:00:32 +[2026-03-26 08:46:56,158][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:46:56,160][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:46:56,162][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:46:56,931][__main__][INFO] - Iteration 702 took 51s (30.53% Gen, 67.99% Train). Generation: 15s, Training: 35s. Estimated remaining time: 32h 31m 46s. Estimated total time: 43h 17m 54s. Time estimates for 10 more iterations: 8m 39s, 100 more iterations: 1h 26m 35s, 500 more iterations: 7h 12m 59s. +[2026-03-26 08:46:56,933][__main__][INFO] - Starting iteration 702. +[2026-03-26 08:46:57,338][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 70 and human policies 1. +[2026-03-26 08:46:57,338][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:47:02,064][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:47:13,163][__main__][INFO] - Number of regex retries in iteration 702: 1 +[2026-03-26 08:47:13,164][__main__][INFO] - agents played in iteration 702 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:47:13,926][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:47:13,950][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:47:13,973][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:47:13,997][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:47:13,998][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:47:13,998][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:47:14,850][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:47:15,303][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:47:15,806][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:47:16,311][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:47:16,810][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:47:17,311][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:47:17,811][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:47:18,312][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:47:18,816][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:47:19,321][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:47:19,825][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:47:20,324][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:47:20,818][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:47:21,314][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:47:21,809][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:47:22,304][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:47:22,802][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:47:23,312][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:47:23,813][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:47:24,312][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:47:24,808][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:47:25,307][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:47:25,805][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:47:26,310][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:47:26,815][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:47:27,315][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:47:27,816][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:47:28,327][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:47:28,828][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:47:29,329][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:47:29,831][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:47:30,332][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:47:30,832][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:47:31,362][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:47:31,867][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:47:32,370][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:47:32,875][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:47:33,376][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:47:33,877][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:47:34,372][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:47:34,873][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:47:35,374][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:47:35,868][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:47:36,369][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:47:36,871][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:47:37,372][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:47:37,871][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:47:38,372][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:47:38,867][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:47:39,366][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:47:39,865][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:47:40,361][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:47:40,859][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:47:41,362][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:47:41,859][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:47:42,358][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:47:42,857][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:47:43,358][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:47:43,860][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:47:44,361][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:47:44,861][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:47:45,359][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:47:45,859][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:47:46,360][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:47:46,857][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9921 tokens. +[2026-03-26 08:47:47,742][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.27%, ΔTime: 00:00:32 +[2026-03-26 08:47:48,575][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:47:48,578][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:47:48,580][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:47:49,276][__main__][INFO] - Iteration 703 took 51s (30.47% Gen, 68.19% Train). Generation: 15s, Training: 35s. Estimated remaining time: 32h 29m 56s. Estimated total time: 43h 16m 57s. Time estimates for 10 more iterations: 8m 39s, 100 more iterations: 1h 26m 33s, 500 more iterations: 7h 12m 49s. +[2026-03-26 08:47:49,279][__main__][INFO] - Starting iteration 703. +[2026-03-26 08:47:49,685][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 70 and human policies 1. +[2026-03-26 08:47:49,685][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:48:06,016][__main__][INFO] - Number of regex retries in iteration 703: 0 +[2026-03-26 08:48:06,017][__main__][INFO] - agents played in iteration 703 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:48:06,789][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:48:06,813][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:48:06,836][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:48:06,859][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:48:06,860][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:48:06,861][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:48:07,697][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:48:08,148][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:48:08,648][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:48:09,145][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:48:09,646][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:48:10,154][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:48:10,663][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:48:11,158][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:48:11,656][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:48:12,148][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:48:12,643][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:48:13,134][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:48:13,628][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:48:14,127][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:48:14,622][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:48:15,114][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:48:15,604][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:48:16,099][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:48:16,592][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:48:17,088][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:48:17,589][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:48:18,089][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:48:18,589][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:48:19,090][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:48:19,587][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:48:20,088][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:48:20,591][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:48:21,093][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:48:21,599][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:48:22,109][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:48:22,611][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:48:23,115][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:48:23,616][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:48:24,115][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:48:24,615][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:48:25,109][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:48:25,611][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:48:26,110][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:48:26,606][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:48:27,099][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:48:27,595][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:48:28,094][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:48:28,593][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:48:29,088][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:48:29,588][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:48:30,080][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:48:30,578][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:48:31,071][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:48:31,566][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:48:32,063][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:48:32,558][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:48:33,054][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:48:33,553][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:48:34,050][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:48:34,551][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:48:35,053][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:48:35,555][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:48:36,052][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:48:36,553][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:48:37,058][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:48:37,554][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:48:38,049][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:48:38,543][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:48:39,040][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:48:39,538][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9757 tokens. +[2026-03-26 08:48:40,413][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.10%, ΔTime: 00:00:32 +[2026-03-26 08:48:41,181][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:48:41,183][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:48:41,185][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:48:41,968][__main__][INFO] - Iteration 704 took 52s (31.24% Gen, 67.26% Train). Generation: 16s, Training: 35s. Estimated remaining time: 32h 46m 18s. Estimated total time: 43h 34m 12s. Time estimates for 10 more iterations: 8m 42s, 100 more iterations: 1h 27m 8s, 500 more iterations: 7h 15m 42s. +[2026-03-26 08:48:41,970][__main__][INFO] - Starting iteration 704. +[2026-03-26 08:48:42,370][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 70 and human policies 1. +[2026-03-26 08:48:42,370][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:48:46,788][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:48:58,463][__main__][INFO] - Number of regex retries in iteration 704: 1 +[2026-03-26 08:48:58,464][__main__][INFO] - agents played in iteration 704 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:48:59,228][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:48:59,253][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:48:59,278][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:48:59,302][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:48:59,303][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:48:59,303][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:49:00,154][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:49:00,610][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:49:01,123][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:49:01,621][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:49:02,130][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:49:02,628][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:49:03,126][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:49:03,623][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:49:04,120][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:49:04,618][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:49:05,113][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:49:05,607][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:49:06,107][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:49:06,609][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:49:07,103][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:49:07,599][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:49:08,095][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:49:08,593][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:49:09,091][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:49:09,585][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:49:10,087][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:49:10,589][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:49:11,090][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:49:11,585][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:49:12,085][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:49:12,587][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:49:13,089][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:49:13,589][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:49:14,091][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:49:14,591][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:49:15,093][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:49:15,594][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:49:16,092][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:49:16,591][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:49:17,089][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:49:17,590][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:49:18,100][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:49:18,606][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:49:19,106][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:49:19,603][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:49:20,103][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:49:20,603][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:49:21,102][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:49:21,599][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:49:22,095][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:49:22,592][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:49:23,085][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:49:23,580][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:49:24,078][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:49:24,575][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:49:25,073][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:49:25,570][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:49:26,066][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:49:26,565][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:49:27,066][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:49:27,565][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:49:28,066][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:49:28,569][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:49:29,067][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:49:29,564][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:49:30,062][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:49:30,560][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:49:31,057][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:49:31,559][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:49:32,056][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9837 tokens. +[2026-03-26 08:49:32,941][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.04%, ΔTime: 00:00:32 +[2026-03-26 08:49:33,694][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:49:33,696][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:49:33,698][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:49:34,592][__main__][INFO] - Iteration 705 took 52s (30.82% Gen, 67.47% Train). Generation: 16s, Training: 35s. Estimated remaining time: 32h 42m 23s. Estimated total time: 43h 31m 9s. Time estimates for 10 more iterations: 8m 42s, 100 more iterations: 1h 27m 2s, 500 more iterations: 7h 15m 11s. +[2026-03-26 08:49:34,594][__main__][INFO] - Starting iteration 705. +[2026-03-26 08:49:35,003][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 70 and human policies 1. +[2026-03-26 08:49:35,003][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:49:49,176][__main__][INFO] - Number of regex retries in iteration 705: 0 +[2026-03-26 08:49:49,177][__main__][INFO] - agents played in iteration 705 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:49:49,939][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:49:49,963][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:49:49,987][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:49:50,011][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:49:50,011][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:49:50,012][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:49:50,858][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:49:51,314][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:49:51,826][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:49:52,325][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:49:52,827][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:49:53,332][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:49:53,833][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:49:54,335][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:49:54,837][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:49:55,345][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:49:55,844][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:49:56,344][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:49:56,843][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:49:57,348][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:49:57,848][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:49:58,347][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:49:58,846][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:49:59,345][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:49:59,841][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:50:00,342][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:50:00,839][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:50:01,335][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:50:01,843][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:50:02,339][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:50:02,834][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:50:03,331][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:50:04,839][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:50:05,338][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:50:05,840][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:50:06,346][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:50:06,846][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:50:07,344][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:50:07,843][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:50:08,349][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:50:08,852][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:50:09,357][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:50:09,871][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:50:10,375][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:50:10,877][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:50:11,379][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:50:11,881][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:50:12,379][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:50:12,877][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:50:13,374][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:50:13,875][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:50:14,374][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:50:14,873][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:50:15,368][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:50:15,862][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:50:16,358][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:50:16,853][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:50:17,348][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:50:17,843][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:50:18,339][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:50:18,835][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:50:19,331][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:50:19,827][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:50:20,325][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:50:20,827][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:50:21,326][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:50:21,825][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:50:22,325][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:50:22,829][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:50:23,330][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:50:23,827][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9900 tokens. +[2026-03-26 08:50:24,722][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:33 +[2026-03-26 08:50:25,519][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:50:25,521][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:50:25,523][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:50:26,243][__main__][INFO] - Iteration 706 took 51s (27.66% Gen, 70.93% Train). Generation: 14s, Training: 36s. Estimated remaining time: 31h 52m 25s. Estimated total time: 42h 42m 3s. Time estimates for 10 more iterations: 8m 32s, 100 more iterations: 1h 25m 24s, 500 more iterations: 7h 7m 0s. +[2026-03-26 08:50:26,245][__main__][INFO] - Starting iteration 706. +[2026-03-26 08:50:26,663][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 70 and human policies 1. +[2026-03-26 08:50:26,664][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:50:44,101][__main__][INFO] - Number of regex retries in iteration 706: 0 +[2026-03-26 08:50:44,102][__main__][INFO] - agents played in iteration 706 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:50:44,868][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:50:44,892][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:50:44,916][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:50:44,940][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:50:44,940][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:50:44,941][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:50:45,774][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:50:46,224][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:50:46,719][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:50:47,221][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:50:47,715][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:50:48,209][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:50:48,706][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:50:49,207][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:50:49,707][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:50:50,208][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:50:50,706][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:50:51,211][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:50:51,717][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:50:52,214][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:50:52,709][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:50:53,209][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:50:53,709][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:50:54,212][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:50:54,713][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:50:55,216][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:50:55,719][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:50:56,227][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:50:56,732][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:50:57,238][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:50:57,743][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:50:58,246][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:50:58,747][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:50:59,246][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:50:59,748][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:51:00,251][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:51:00,754][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:51:01,257][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:51:01,760][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:51:02,260][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:51:02,762][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:51:03,260][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:51:03,757][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:51:04,259][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:51:04,756][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:51:05,255][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:51:05,754][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:51:06,254][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:51:06,755][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:51:07,254][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:51:07,753][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:51:08,257][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:51:08,757][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:51:09,251][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:51:09,750][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:51:10,247][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:51:10,742][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:51:11,240][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:51:11,736][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:51:12,236][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:51:12,736][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:51:13,237][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:51:13,735][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:51:14,235][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:51:14,734][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:51:15,231][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:51:15,729][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:51:16,226][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:51:16,727][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:51:17,229][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:51:17,728][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9954 tokens. +[2026-03-26 08:51:18,619][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.21%, ΔTime: 00:00:32 +[2026-03-26 08:51:19,398][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:51:19,401][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:51:19,402][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:51:20,120][__main__][INFO] - Iteration 707 took 53s (32.62% Gen, 66.03% Train). Generation: 17s, Training: 35s. Estimated remaining time: 33h 42m 23s. Estimated total time: 44h 32m 54s. Time estimates for 10 more iterations: 8m 54s, 100 more iterations: 1h 29m 5s, 500 more iterations: 7h 25m 29s. +[2026-03-26 08:51:20,123][__main__][INFO] - Starting iteration 707. +[2026-03-26 08:51:20,524][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 70 and human policies 1. +[2026-03-26 08:51:20,525][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:51:36,234][__main__][INFO] - Number of regex retries in iteration 707: 0 +[2026-03-26 08:51:36,235][__main__][INFO] - agents played in iteration 707 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:51:37,006][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:51:37,030][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:51:37,053][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:51:37,077][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:51:37,078][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:51:37,078][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:51:37,919][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:51:38,376][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:51:38,880][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:51:39,378][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:51:39,874][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:51:40,375][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:51:40,876][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:51:41,375][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:51:41,873][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:51:42,374][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:51:42,870][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:51:43,371][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:51:43,871][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:51:44,364][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:51:44,863][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:51:45,363][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:51:45,860][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:51:46,358][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:51:46,855][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:51:47,359][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:51:47,863][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:51:48,364][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:51:48,864][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:51:49,364][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:51:49,866][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:51:50,369][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:51:50,876][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:51:51,376][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:51:51,883][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:51:52,386][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:51:52,888][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:51:53,389][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:51:53,888][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:51:54,386][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:51:54,885][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:51:55,384][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:51:55,883][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:51:56,379][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:51:56,878][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:51:57,377][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:51:57,876][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:51:58,371][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:51:58,868][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:51:59,366][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:51:59,864][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:52:00,363][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:52:00,855][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:52:01,347][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:52:01,839][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:52:02,338][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:52:02,837][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:52:03,337][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:52:03,839][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:52:04,340][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:52:04,843][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:52:05,346][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:52:05,845][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:52:06,342][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:52:06,840][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:52:07,339][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:52:07,836][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:52:08,334][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:52:08,833][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:52:09,331][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:52:09,828][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9955 tokens. +[2026-03-26 08:52:10,706][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.20%, ΔTime: 00:00:32 +[2026-03-26 08:52:11,484][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:52:11,486][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:52:11,488][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:52:12,361][__main__][INFO] - Iteration 708 took 51s (30.31% Gen, 68.01% Train). Generation: 15s, Training: 35s. Estimated remaining time: 32h 20m 27s. Estimated total time: 43h 11m 51s. Time estimates for 10 more iterations: 8m 38s, 100 more iterations: 1h 26m 23s, 500 more iterations: 7h 11m 58s. +[2026-03-26 08:52:12,363][__main__][INFO] - Starting iteration 708. +[2026-03-26 08:52:12,761][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 70 and human policies 1. +[2026-03-26 08:52:12,762][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:52:16,451][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:52:16,453][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:52:27,076][__main__][INFO] - Number of regex retries in iteration 708: 2 +[2026-03-26 08:52:27,077][__main__][INFO] - agents played in iteration 708 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:52:27,843][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:52:27,866][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:52:27,893][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:52:27,917][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:52:27,918][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:52:27,918][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:52:28,780][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:52:29,235][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:52:29,743][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:52:30,240][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:52:30,738][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:52:31,237][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:52:31,735][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:52:32,233][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:52:32,731][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:52:33,225][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:52:33,726][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:52:34,222][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:52:34,722][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:52:35,224][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:52:35,719][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:52:36,216][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:52:36,714][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:52:37,215][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:52:37,718][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:52:38,223][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:52:38,726][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:52:39,230][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:52:39,731][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:52:40,236][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:52:40,740][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:52:41,242][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:52:41,745][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:52:42,248][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:52:42,750][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:52:43,255][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:52:43,761][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:52:44,264][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:52:44,768][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:52:45,268][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:52:45,777][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:52:46,275][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:52:46,774][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:52:47,273][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:52:47,772][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:52:48,268][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:52:48,763][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:52:49,264][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:52:49,769][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:52:50,275][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:52:50,773][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:52:51,274][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:52:51,776][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:52:52,274][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:52:52,774][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:52:53,272][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:52:53,771][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:52:54,269][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:52:54,770][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:52:55,273][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:52:55,771][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:52:56,270][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:52:56,770][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:52:57,275][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:52:57,785][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:52:58,282][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:52:58,780][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:52:59,277][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:52:59,774][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:53:00,270][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:53:00,765][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9906 tokens. +[2026-03-26 08:53:01,660][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.99%, Current % of VRAM taken: 60.41%, Block Peak % of device VRAM: 62.13%, ΔTime: 00:00:32 +[2026-03-26 08:53:02,437][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:53:02,439][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:53:02,441][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:53:03,246][__main__][INFO] - Iteration 709 took 50s (28.36% Gen, 70.05% Train). Generation: 14s, Training: 35s. Estimated remaining time: 31h 12m 1s. Estimated total time: 42h 4m 15s. Time estimates for 10 more iterations: 8m 24s, 100 more iterations: 1h 24m 8s, 500 more iterations: 7h 0m 42s. +[2026-03-26 08:53:03,249][__main__][INFO] - Starting iteration 709. +[2026-03-26 08:53:03,657][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 70 and human policies 1. +[2026-03-26 08:53:03,657][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:53:23,597][__main__][INFO] - Number of regex retries in iteration 709: 0 +[2026-03-26 08:53:23,598][__main__][INFO] - agents played in iteration 709 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:53:24,371][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:53:24,394][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:53:24,417][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:53:24,441][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:53:24,441][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:53:24,442][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:53:25,296][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:53:25,752][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:53:26,250][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:53:26,744][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:53:27,244][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:53:27,738][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:53:28,233][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:53:28,727][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:53:29,222][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:53:29,720][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:53:30,221][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:53:30,725][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:53:31,227][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:53:31,729][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:53:32,236][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:53:32,736][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:53:33,239][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:53:33,746][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:53:34,245][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:53:34,743][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:53:35,240][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:53:35,736][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:53:36,233][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:53:36,734][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:53:37,234][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:53:37,725][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:53:38,222][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:53:38,719][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:53:39,215][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:53:39,711][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:53:40,211][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:53:40,712][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:53:41,214][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:53:41,711][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:53:42,209][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:53:42,711][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:53:43,203][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:53:43,700][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:53:44,221][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:53:44,718][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:53:45,216][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:53:45,725][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:53:46,227][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:53:46,726][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:53:47,230][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:53:47,732][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:53:48,235][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:53:48,737][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:53:49,239][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:53:49,736][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:53:50,234][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:53:50,733][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:53:51,231][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:53:51,728][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:53:52,227][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:53:52,724][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:53:53,229][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:53:53,800][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:53:54,299][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:53:54,799][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:53:55,300][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:53:55,798][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:53:56,292][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:53:56,787][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:53:57,284][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9822 tokens. +[2026-03-26 08:53:58,171][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.09%, ΔTime: 00:00:32 +[2026-03-26 08:53:58,936][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:53:58,938][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:53:58,940][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:53:59,682][__main__][INFO] - Iteration 710 took 56s (35.59% Gen, 63.08% Train). Generation: 19s, Training: 35s. Estimated remaining time: 35h 48m 7s. Estimated total time: 46h 41m 18s. Time estimates for 10 more iterations: 9m 20s, 100 more iterations: 1h 33m 22s, 500 more iterations: 7h 46m 53s. +[2026-03-26 08:53:59,685][__main__][INFO] - Starting iteration 710. +[2026-03-26 08:54:00,084][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 70 and human policies 1. +[2026-03-26 08:54:00,084][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:54:08,487][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:54:17,906][__main__][INFO] - Number of regex retries in iteration 710: 1 +[2026-03-26 08:54:17,907][__main__][INFO] - agents played in iteration 710 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:54:18,683][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:54:18,706][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:54:18,730][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:54:18,753][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:54:18,754][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:54:18,754][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:54:19,593][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:54:20,050][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:54:20,563][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:54:21,060][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:54:21,560][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:54:22,063][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:54:22,558][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:54:23,053][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:54:23,558][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:54:24,055][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:54:24,556][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:54:25,056][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:54:25,554][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:54:26,051][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:54:26,546][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:54:27,052][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:54:27,549][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:54:28,043][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:54:28,543][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:54:29,043][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:54:29,542][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:54:30,044][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:54:30,540][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:54:31,040][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:54:31,546][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:54:32,046][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:54:32,545][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:54:33,039][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:54:33,537][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:54:34,031][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:54:34,524][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:54:35,019][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:54:35,510][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:54:36,009][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:54:36,517][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:54:37,020][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:54:37,524][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:54:38,028][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:54:38,531][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:54:39,029][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:54:39,527][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:54:40,025][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:54:40,523][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:54:41,020][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:54:41,518][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:54:42,020][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:54:42,526][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:54:43,022][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:54:43,521][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:54:44,016][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:54:44,507][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:54:45,005][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:54:45,501][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:54:45,997][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:54:46,488][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:54:46,978][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:54:47,469][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:54:47,964][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:54:48,459][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:54:48,961][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:54:49,465][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:54:49,960][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:54:50,457][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:54:50,952][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:54:51,446][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9699 tokens. +[2026-03-26 08:54:52,335][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.02%, Current % of VRAM taken: 60.45%, Block Peak % of device VRAM: 62.15%, ΔTime: 00:00:32 +[2026-03-26 08:54:53,097][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:54:53,099][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:54:53,101][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:54:54,450][__main__][INFO] - Iteration 711 took 54s (32.78% Gen, 64.73% Train). Generation: 17s, Training: 35s. Estimated remaining time: 34h 24m 16s. Estimated total time: 45h 18m 22s. Time estimates for 10 more iterations: 9m 3s, 100 more iterations: 1h 30m 36s, 500 more iterations: 7h 33m 3s. +[2026-03-26 08:54:54,453][__main__][INFO] - Starting iteration 711. +[2026-03-26 08:54:54,852][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 71 and human policies 1. +[2026-03-26 08:54:54,853][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:55:28,013][__main__][INFO] - Number of regex retries in iteration 711: 0 +[2026-03-26 08:55:28,014][__main__][INFO] - agents played in iteration 711 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:55:28,794][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:55:28,818][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:55:28,841][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:55:28,865][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:55:28,865][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:55:28,866][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:55:29,722][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:55:30,171][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:55:30,665][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:55:31,162][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:55:31,658][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:55:32,153][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:55:32,645][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:55:33,150][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:55:33,643][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:55:34,138][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:55:34,635][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:55:35,139][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:55:35,637][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:55:36,134][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:55:36,631][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:55:37,151][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:55:37,653][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:55:38,151][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:55:38,648][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:55:39,144][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:55:39,644][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:55:40,146][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:55:40,648][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:55:41,148][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:55:41,647][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:55:42,151][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:55:42,658][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:55:43,159][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:55:43,662][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:55:44,166][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:55:44,671][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:55:45,177][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:55:45,684][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:55:46,187][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:55:46,687][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:55:47,188][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:55:47,695][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:55:48,200][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:55:48,696][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:55:49,195][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:55:49,695][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:55:50,190][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:55:50,685][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:55:51,183][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:55:51,680][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:55:52,175][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:55:52,669][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:55:53,163][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:55:53,657][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:55:54,152][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:55:54,645][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:55:55,137][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:55:55,636][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:55:56,127][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:55:56,618][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:55:57,109][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:55:57,599][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:55:58,096][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:55:58,598][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:55:59,098][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:55:59,595][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:56:00,092][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:56:00,587][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:56:01,085][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:56:01,583][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9836 tokens. +[2026-03-26 08:56:02,480][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.35%, ΔTime: 00:00:32 +[2026-03-26 08:56:03,219][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:56:03,222][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:56:03,223][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:56:04,093][__main__][INFO] - Iteration 712 took 1m 9s (47.89% Gen, 50.85% Train). Generation: 33s, Training: 35s. Estimated remaining time: 46h 46m 48s. Estimated total time: 57h 42m 3s. Time estimates for 10 more iterations: 11m 32s, 100 more iterations: 1h 55m 24s, 500 more iterations: 9h 37m 0s. +[2026-03-26 08:56:04,095][__main__][INFO] - Starting iteration 712. +[2026-03-26 08:56:04,496][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 71 and human policies 1. +[2026-03-26 08:56:04,497][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:56:19,860][__main__][INFO] - Number of regex retries in iteration 712: 0 +[2026-03-26 08:56:19,861][__main__][INFO] - agents played in iteration 712 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:56:20,647][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:56:20,671][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:56:20,695][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:56:20,718][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:56:20,719][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:56:20,719][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:56:21,569][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:56:22,024][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:56:22,527][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:56:23,024][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:56:23,523][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:56:24,022][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:56:24,520][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:56:25,025][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:56:25,527][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:56:26,021][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:56:26,525][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:56:27,025][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:56:27,521][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:56:28,019][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:56:28,518][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:56:29,015][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:56:29,510][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:56:30,002][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:56:30,498][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:56:30,995][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:56:31,488][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:56:31,983][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:56:32,486][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:56:32,976][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:56:33,466][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:56:33,963][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:56:34,461][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:56:34,954][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:56:35,452][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:56:35,950][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:56:36,448][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:56:36,945][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:56:37,443][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:56:37,940][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:56:38,437][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:56:38,938][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:56:39,443][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:56:39,941][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:56:40,440][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:56:40,937][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:56:41,433][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:56:41,928][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:56:42,423][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:56:42,921][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:56:43,418][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:56:43,922][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:56:44,426][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:56:44,927][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:56:45,428][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:56:45,928][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:56:46,428][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:56:46,927][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:56:47,427][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:56:47,924][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:56:48,426][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:56:48,924][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:56:49,422][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:56:49,915][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:56:50,408][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:56:50,905][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:56:51,398][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:56:51,896][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:56:52,399][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:56:52,897][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:56:53,394][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9791 tokens. +[2026-03-26 08:56:54,310][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.92%, Current % of VRAM taken: 60.34%, Block Peak % of device VRAM: 62.14%, ΔTime: 00:00:32 +[2026-03-26 08:56:55,055][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:56:55,057][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:56:55,058][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:56:55,919][__main__][INFO] - Iteration 713 took 51s (29.88% Gen, 68.45% Train). Generation: 15s, Training: 35s. Estimated remaining time: 31h 55m 2s. Estimated total time: 42h 51m 9s. Time estimates for 10 more iterations: 8m 34s, 100 more iterations: 1h 25m 42s, 500 more iterations: 7h 8m 31s. +[2026-03-26 08:56:55,921][__main__][INFO] - Starting iteration 713. +[2026-03-26 08:56:56,323][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 71 and human policies 1. +[2026-03-26 08:56:56,323][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:57:15,599][__main__][INFO] - Number of regex retries in iteration 713: 0 +[2026-03-26 08:57:15,599][__main__][INFO] - agents played in iteration 713 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:57:16,388][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:57:16,411][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:57:16,435][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:57:16,458][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:57:16,459][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:57:16,459][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:57:17,313][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:57:17,768][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:57:18,267][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:57:18,763][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:57:19,258][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:57:19,754][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:57:20,250][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:57:20,751][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:57:21,253][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:57:21,752][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:57:22,247][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:57:22,744][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:57:23,243][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:57:23,740][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:57:24,235][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:57:24,727][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:57:25,224][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:57:25,720][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:57:26,221][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:57:26,714][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:57:27,206][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:57:27,700][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:57:28,195][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:57:28,698][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:57:29,200][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:57:29,697][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:57:30,195][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:57:30,692][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:57:31,185][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:57:31,685][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:57:32,190][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:57:32,683][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:57:33,180][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:57:33,676][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:57:34,174][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:57:34,669][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:57:35,166][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:57:35,663][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:57:36,159][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:57:36,662][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:57:37,157][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:57:37,658][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:57:38,152][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:57:38,646][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:57:39,149][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:57:39,649][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:57:40,145][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:57:40,646][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:57:41,142][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:57:41,642][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:57:42,147][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:57:42,650][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:57:43,154][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:57:43,654][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:57:44,153][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:57:44,656][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:57:45,159][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:57:45,661][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:57:46,164][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:57:46,662][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:57:47,156][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:57:47,651][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:57:48,142][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:57:48,632][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:57:49,128][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9785 tokens. +[2026-03-26 08:57:50,006][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.02%, Current % of VRAM taken: 60.45%, Block Peak % of device VRAM: 61.90%, ΔTime: 00:00:32 +[2026-03-26 08:57:50,761][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:57:50,763][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:57:50,765][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:57:51,519][__main__][INFO] - Iteration 714 took 55s (34.92% Gen, 63.71% Train). Generation: 19s, Training: 35s. Estimated remaining time: 35h 2m 48s. Estimated total time: 45h 59m 51s. Time estimates for 10 more iterations: 9m 11s, 100 more iterations: 1h 31m 59s, 500 more iterations: 7h 39m 58s. +[2026-03-26 08:57:51,521][__main__][INFO] - Starting iteration 714. +[2026-03-26 08:57:51,924][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 71 and human policies 1. +[2026-03-26 08:57:51,924][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:58:12,096][__main__][INFO] - Number of regex retries in iteration 714: 0 +[2026-03-26 08:58:12,097][__main__][INFO] - agents played in iteration 714 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:58:12,873][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:58:12,896][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:58:12,920][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:58:12,944][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:58:12,944][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:58:12,945][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:58:13,781][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:58:14,233][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:58:14,729][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:58:15,226][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:58:15,718][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:58:16,209][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:58:16,702][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:58:17,193][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:58:17,683][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:58:18,183][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:58:18,687][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:58:19,190][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:58:19,687][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:58:20,186][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:58:20,679][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:58:21,176][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:58:21,672][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:58:22,170][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:58:22,664][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:58:23,160][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:58:23,665][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:58:24,157][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:58:24,649][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:58:25,146][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:58:25,644][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:58:26,141][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:58:26,636][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:58:27,128][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:58:27,624][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:58:28,125][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:58:28,616][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:58:29,107][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:58:29,609][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:58:30,116][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:58:30,614][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:58:31,115][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:58:31,614][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:58:32,112][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:58:32,608][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:58:33,105][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:58:33,603][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:58:34,117][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:58:34,619][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:58:35,112][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:58:35,604][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:58:36,097][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:58:36,599][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:58:37,098][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:58:37,607][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:58:38,106][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:58:38,605][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:58:39,104][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:58:39,602][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:58:40,102][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:58:40,600][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:58:41,098][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:58:41,595][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:58:42,094][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:58:42,592][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:58:43,089][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:58:43,588][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:58:44,086][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:58:44,582][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:58:45,086][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:58:45,586][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9729 tokens. +[2026-03-26 08:58:46,503][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.05%, Current % of VRAM taken: 60.48%, Block Peak % of device VRAM: 62.00%, ΔTime: 00:00:32 +[2026-03-26 08:58:47,265][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:58:47,267][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:58:47,269][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:58:48,001][__main__][INFO] - Iteration 715 took 56s (35.97% Gen, 62.72% Train). Generation: 20s, Training: 35s. Estimated remaining time: 35h 45m 54s. Estimated total time: 46h 43m 53s. Time estimates for 10 more iterations: 9m 20s, 100 more iterations: 1h 33m 27s, 500 more iterations: 7h 47m 18s. +[2026-03-26 08:58:48,003][__main__][INFO] - Starting iteration 715. +[2026-03-26 08:58:48,403][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 71 and human policies 1. +[2026-03-26 08:58:48,404][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 08:59:11,191][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 08:59:16,521][__main__][INFO] - Number of regex retries in iteration 715: 1 +[2026-03-26 08:59:16,522][__main__][INFO] - agents played in iteration 715 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 08:59:17,297][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:59:17,321][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:59:17,344][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:59:17,368][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 08:59:17,368][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 08:59:17,369][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 08:59:18,218][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 08:59:18,668][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 08:59:19,166][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 08:59:19,665][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 08:59:20,161][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 08:59:20,653][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 08:59:21,147][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 08:59:21,642][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 08:59:22,136][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 08:59:22,632][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 08:59:23,129][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 08:59:23,626][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 08:59:24,123][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 08:59:24,619][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 08:59:25,116][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 08:59:25,619][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 08:59:26,117][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 08:59:26,615][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 08:59:27,139][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 08:59:27,638][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 08:59:28,141][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 08:59:28,639][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 08:59:29,137][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 08:59:29,635][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 08:59:30,133][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 08:59:30,632][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 08:59:31,128][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 08:59:31,628][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 08:59:32,127][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 08:59:32,623][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 08:59:33,124][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 08:59:33,626][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 08:59:34,130][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 08:59:34,632][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 08:59:35,131][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 08:59:35,634][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 08:59:36,138][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 08:59:36,639][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 08:59:37,136][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 08:59:37,633][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 08:59:38,130][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 08:59:38,630][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 08:59:39,132][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 08:59:39,641][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 08:59:40,145][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 08:59:40,648][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 08:59:41,149][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 08:59:41,656][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 08:59:42,158][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 08:59:42,657][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 08:59:43,155][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 08:59:43,654][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 08:59:44,154][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 08:59:44,663][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 08:59:45,163][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 08:59:45,662][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 08:59:46,161][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 08:59:46,657][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 08:59:47,159][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 08:59:47,654][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 08:59:48,152][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 08:59:48,647][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 08:59:49,145][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 08:59:49,640][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 08:59:50,138][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9857 tokens. +[2026-03-26 08:59:51,036][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 12.97%, Current % of VRAM taken: 59.40%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 08:59:51,798][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 08:59:51,801][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 08:59:51,802][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 08:59:52,660][__main__][INFO] - Iteration 716 took 1m 4s (43.76% Gen, 54.91% Train). Generation: 28s, Training: 35s. Estimated remaining time: 42h 33m 48s. Estimated total time: 53h 32m 52s. Time estimates for 10 more iterations: 10m 42s, 100 more iterations: 1h 47m 5s, 500 more iterations: 8h 55m 28s. +[2026-03-26 08:59:52,662][__main__][INFO] - Starting iteration 716. +[2026-03-26 08:59:53,064][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 71 and human policies 1. +[2026-03-26 08:59:53,065][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:00:11,380][__main__][INFO] - Number of regex retries in iteration 716: 0 +[2026-03-26 09:00:11,381][__main__][INFO] - agents played in iteration 716 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:00:12,147][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:00:12,171][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:00:12,195][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:00:12,218][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:00:12,219][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:00:12,219][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:00:13,060][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:00:13,510][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:00:14,011][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:00:14,509][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:00:15,012][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:00:15,508][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:00:16,005][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:00:16,503][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:00:16,994][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:00:17,489][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:00:17,986][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:00:18,482][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:00:18,976][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:00:19,474][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:00:19,971][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:00:20,466][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:00:20,960][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:00:21,451][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:00:21,943][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:00:22,441][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:00:22,936][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:00:23,430][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:00:23,926][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:00:24,417][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:00:24,917][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:00:25,412][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:00:25,907][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:00:26,402][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:00:26,897][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:00:27,391][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:00:27,886][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:00:28,387][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:00:28,885][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:00:29,384][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:00:29,881][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:00:30,389][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:00:30,891][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:00:31,398][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:00:31,897][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:00:32,396][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:00:32,896][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:00:33,399][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:00:33,894][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:00:34,392][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:00:34,890][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:00:35,389][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:00:35,887][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:00:36,386][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:00:36,882][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:00:37,379][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:00:37,880][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:00:38,377][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:00:38,875][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:00:39,371][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:00:39,868][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:00:40,359][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:00:40,855][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:00:41,348][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:00:41,845][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:00:42,342][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:00:42,840][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:00:43,338][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:00:43,838][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:00:44,339][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:00:44,834][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9786 tokens. +[2026-03-26 09:00:45,717][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.76%, Current % of VRAM taken: 60.19%, Block Peak % of device VRAM: 62.19%, ΔTime: 00:00:32 +[2026-03-26 09:00:46,470][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:00:46,472][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:00:46,474][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:00:47,138][__main__][INFO] - Iteration 717 took 54s (33.87% Gen, 64.90% Train). Generation: 18s, Training: 35s. Estimated remaining time: 34h 3m 44s. Estimated total time: 45h 3m 43s. Time estimates for 10 more iterations: 9m 0s, 100 more iterations: 1h 30m 7s, 500 more iterations: 7h 30m 37s. +[2026-03-26 09:00:47,140][__main__][INFO] - Starting iteration 717. +[2026-03-26 09:00:47,541][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 71 and human policies 1. +[2026-03-26 09:00:47,542][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:01:01,683][__main__][INFO] - Number of regex retries in iteration 717: 0 +[2026-03-26 09:01:01,684][__main__][INFO] - agents played in iteration 717 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:01:02,454][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.33%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:01:02,478][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.33%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:01:02,502][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.33%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:01:02,525][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.33%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:01:02,526][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:01:02,527][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:01:03,372][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:01:03,822][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:01:04,326][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:01:04,824][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:01:05,321][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:01:05,814][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:01:06,309][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:01:06,810][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:01:07,306][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:01:07,805][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:01:08,300][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:01:08,793][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:01:09,294][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:01:09,788][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:01:10,285][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:01:10,778][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:01:11,274][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:01:11,765][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:01:12,259][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:01:12,750][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:01:13,246][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:01:13,741][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:01:14,231][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:01:14,722][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:01:15,217][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:01:15,716][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:01:16,216][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:01:16,715][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:01:17,211][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:01:17,709][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:01:18,208][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:01:18,705][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:01:19,202][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:01:19,697][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:01:20,194][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:01:20,693][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:01:21,196][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:01:21,698][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:01:22,201][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:01:22,698][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:01:23,196][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:01:23,697][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:01:24,201][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:01:24,700][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:01:25,198][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:01:25,697][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:01:26,196][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:01:26,694][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:01:27,191][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:01:27,684][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:01:28,179][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:01:28,674][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:01:29,174][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:01:29,670][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:01:30,171][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:01:30,666][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:01:31,159][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:01:31,657][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:01:32,151][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:01:32,651][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:01:33,154][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:01:33,656][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:01:34,156][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:01:34,655][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:01:35,155][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9645 tokens. +[2026-03-26 09:01:36,034][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 61.97%, ΔTime: 00:00:32 +[2026-03-26 09:01:36,785][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:01:36,787][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:01:36,789][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:01:37,546][__main__][INFO] - Iteration 718 took 50s (28.28% Gen, 70.20% Train). Generation: 14s, Training: 35s. Estimated remaining time: 30h 39m 29s. Estimated total time: 41h 40m 18s. Time estimates for 10 more iterations: 8m 20s, 100 more iterations: 1h 23m 20s, 500 more iterations: 6h 56m 43s. +[2026-03-26 09:01:37,549][__main__][INFO] - Starting iteration 718. +[2026-03-26 09:01:37,949][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 71 and human policies 1. +[2026-03-26 09:01:37,949][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:01:58,171][__main__][INFO] - Number of regex retries in iteration 718: 0 +[2026-03-26 09:01:58,172][__main__][INFO] - agents played in iteration 718 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:01:58,940][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:01:58,963][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:01:58,987][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:01:59,010][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:01:59,011][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:01:59,012][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:01:59,864][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:02:00,313][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:02:00,815][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:02:01,314][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:02:01,806][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:02:02,302][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:02:02,793][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:02:03,283][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:02:03,774][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:02:04,272][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:02:04,769][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:02:05,267][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:02:05,765][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:02:06,257][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:02:06,754][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:02:07,246][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:02:07,743][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:02:08,240][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:02:08,741][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:02:09,242][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:02:09,741][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:02:10,242][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:02:10,742][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:02:11,243][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:02:11,744][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:02:12,239][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:02:12,734][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:02:13,235][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:02:13,731][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:02:14,225][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:02:14,725][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:02:15,244][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:02:15,741][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:02:16,236][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:02:16,730][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:02:17,227][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:02:17,724][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:02:18,221][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:02:18,717][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:02:19,215][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:02:19,713][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:02:20,213][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:02:20,712][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:02:21,208][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:02:21,706][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:02:22,203][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:02:22,696][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:02:23,193][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:02:23,692][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:02:24,189][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:02:24,687][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:02:25,187][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:02:25,684][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:02:26,182][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:02:26,682][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:02:27,181][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:02:27,678][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:02:28,176][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:02:28,673][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:02:29,169][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:02:29,666][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:02:30,162][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:02:30,655][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:02:31,151][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:02:31,647][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9718 tokens. +[2026-03-26 09:02:32,510][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.04%, Current % of VRAM taken: 60.47%, Block Peak % of device VRAM: 61.98%, ΔTime: 00:00:32 +[2026-03-26 09:02:33,263][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:02:33,266][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:02:33,267][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:02:33,933][__main__][INFO] - Iteration 719 took 55s (36.12% Gen, 62.69% Train). Generation: 20s, Training: 35s. Estimated remaining time: 35h 37m 30s. Estimated total time: 46h 39m 15s. Time estimates for 10 more iterations: 9m 19s, 100 more iterations: 1h 33m 18s, 500 more iterations: 7h 46m 32s. +[2026-03-26 09:02:33,936][__main__][INFO] - Starting iteration 719. +[2026-03-26 09:02:34,335][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 71 and human policies 1. +[2026-03-26 09:02:34,336][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:02:50,220][__main__][INFO] - Number of regex retries in iteration 719: 0 +[2026-03-26 09:02:50,221][__main__][INFO] - agents played in iteration 719 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:02:50,984][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:02:51,008][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:02:51,032][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:02:51,055][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:02:51,056][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:02:51,056][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:02:51,918][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:02:52,365][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:02:52,868][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:02:53,361][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:02:53,856][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:02:54,355][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:02:54,848][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:02:55,344][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:02:55,838][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:02:56,337][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:02:56,840][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:02:57,335][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:02:57,830][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:02:58,326][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:02:58,824][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:02:59,321][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:02:59,819][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:03:00,325][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:03:00,821][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:03:01,319][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:03:01,819][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:03:02,316][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:03:02,807][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:03:03,303][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:03:03,798][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:03:04,299][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:03:04,802][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:03:05,304][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:03:05,805][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:03:06,303][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:03:06,801][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:03:07,299][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:03:07,797][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:03:08,330][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:03:08,833][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:03:09,338][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:03:09,837][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:03:10,336][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:03:10,839][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:03:11,342][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:03:11,844][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:03:12,344][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:03:12,842][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:03:13,340][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:03:13,842][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:03:14,342][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:03:14,841][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:03:15,339][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:03:15,841][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:03:16,342][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:03:16,842][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:03:17,337][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:03:17,831][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:03:18,326][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:03:18,823][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:03:19,322][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:03:19,817][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:03:20,318][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:03:20,814][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:03:21,305][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:03:21,804][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:03:22,300][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:03:22,795][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:03:23,290][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:03:23,785][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9831 tokens. +[2026-03-26 09:03:24,664][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.01%, Current % of VRAM taken: 59.43%, Block Peak % of device VRAM: 62.17%, ΔTime: 00:00:32 +[2026-03-26 09:03:25,418][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:03:25,420][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:03:25,422][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:03:26,243][__main__][INFO] - Iteration 720 took 51s (30.60% Gen, 67.81% Train). Generation: 15s, Training: 35s. Estimated remaining time: 32h 12m 46s. Estimated total time: 43h 15m 24s. Time estimates for 10 more iterations: 8m 39s, 100 more iterations: 1h 26m 30s, 500 more iterations: 7h 12m 34s. +[2026-03-26 09:03:26,245][__main__][INFO] - Starting iteration 720. +[2026-03-26 09:03:26,644][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 71 and human policies 1. +[2026-03-26 09:03:26,644][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:03:41,449][__main__][INFO] - Number of regex retries in iteration 720: 0 +[2026-03-26 09:03:41,451][__main__][INFO] - agents played in iteration 720 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:03:42,214][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:03:42,238][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:03:42,261][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:03:42,285][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:03:42,286][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:03:42,286][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:03:43,140][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:03:43,597][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:03:44,106][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:03:44,607][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:03:45,107][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:03:45,606][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:03:46,106][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:03:46,601][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:03:47,098][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:03:47,595][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:03:48,090][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:03:48,582][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:03:49,080][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:03:49,576][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:03:50,079][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:03:50,580][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:03:51,079][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:03:51,575][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:03:52,071][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:03:52,570][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:03:53,070][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:03:53,579][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:03:54,076][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:03:54,575][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:03:55,072][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:03:55,564][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:03:56,061][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:03:56,559][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:03:57,058][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:03:57,552][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:03:58,050][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:03:58,548][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:03:59,043][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:03:59,545][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:04:00,044][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:04:00,543][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:04:01,048][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:04:01,552][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:04:02,054][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:04:02,553][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:04:03,057][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:04:03,556][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:04:04,057][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:04:04,562][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:04:05,060][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:04:05,557][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:04:06,056][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:04:06,552][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:04:07,047][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:04:07,546][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:04:08,044][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:04:08,544][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:04:09,043][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:04:09,543][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:04:10,042][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:04:10,545][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:04:11,052][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:04:11,548][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:04:12,044][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:04:12,540][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:04:13,035][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:04:13,526][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:04:14,022][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:04:14,515][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:04:15,010][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9756 tokens. +[2026-03-26 09:04:15,905][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.98%, Current % of VRAM taken: 60.41%, Block Peak % of device VRAM: 61.88%, ΔTime: 00:00:32 +[2026-03-26 09:04:16,651][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:04:16,653][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:04:16,655][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:04:17,961][__main__][INFO] - Iteration 721 took 51s (28.85% Gen, 68.60% Train). Generation: 14s, Training: 35s. Estimated remaining time: 31h 42m 24s. Estimated total time: 42h 45m 53s. Time estimates for 10 more iterations: 8m 33s, 100 more iterations: 1h 25m 31s, 500 more iterations: 7h 7m 38s. +[2026-03-26 09:04:17,963][__main__][INFO] - Starting iteration 721. +[2026-03-26 09:04:18,362][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 72 and human policies 1. +[2026-03-26 09:04:18,363][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:04:42,017][__main__][INFO] - Number of regex retries in iteration 721: 0 +[2026-03-26 09:04:42,018][__main__][INFO] - agents played in iteration 721 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:04:42,788][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:04:42,812][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:04:42,836][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:04:42,860][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:04:42,860][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:04:42,861][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:04:43,714][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:04:44,164][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:04:44,661][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:04:45,167][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:04:45,659][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:04:46,159][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:04:46,660][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:04:47,156][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:04:47,652][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:04:48,153][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:04:48,679][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:04:49,181][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:04:49,685][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:04:50,187][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:04:50,686][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:04:51,183][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:04:51,684][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:04:52,184][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:04:52,687][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:04:53,192][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:04:53,694][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:04:54,194][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:04:54,695][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:04:55,197][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:04:55,701][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:04:56,203][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:04:56,705][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:04:57,212][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:04:57,720][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:04:58,219][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:04:58,720][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:04:59,215][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:04:59,717][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:05:00,209][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:05:00,701][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:05:01,192][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:05:01,690][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:05:02,194][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:05:02,686][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:05:03,183][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:05:03,675][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:05:04,172][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:05:04,671][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:05:05,168][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:05:05,665][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:05:06,172][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:05:06,670][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:05:07,168][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:05:07,666][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:05:08,163][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:05:08,661][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:05:09,159][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:05:09,657][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:05:10,158][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:05:10,657][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:05:11,152][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:05:11,651][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:05:12,150][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:05:12,649][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:05:13,151][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:05:13,652][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:05:14,156][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:05:14,655][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:05:15,153][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:05:15,650][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9815 tokens. +[2026-03-26 09:05:16,540][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 12.00%, Current % of VRAM taken: 58.42%, Block Peak % of device VRAM: 62.04%, ΔTime: 00:00:32 +[2026-03-26 09:05:17,291][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:05:17,293][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:05:17,294][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:05:18,114][__main__][INFO] - Iteration 722 took 59s (39.59% Gen, 59.04% Train). Generation: 23s, Training: 35s. Estimated remaining time: 38h 43m 7s. Estimated total time: 49h 47m 37s. Time estimates for 10 more iterations: 9m 57s, 100 more iterations: 1h 39m 35s, 500 more iterations: 8h 17m 56s. +[2026-03-26 09:05:18,116][__main__][INFO] - Starting iteration 722. +[2026-03-26 09:05:18,517][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 72 and human policies 1. +[2026-03-26 09:05:18,517][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:05:33,511][__main__][INFO] - Number of regex retries in iteration 722: 0 +[2026-03-26 09:05:33,512][__main__][INFO] - agents played in iteration 722 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:05:34,278][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 51.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:05:34,302][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 51.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:05:34,325][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 51.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:05:34,349][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 51.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:05:34,349][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:05:34,350][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:05:35,217][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:05:35,669][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:05:36,174][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:05:36,665][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:05:37,156][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:05:37,647][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:05:38,140][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:05:38,631][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:05:39,129][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:05:39,629][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:05:40,128][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:05:40,627][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:05:41,126][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:05:41,624][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:05:42,123][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:05:42,627][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:05:43,125][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:05:43,645][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:05:44,150][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:05:44,648][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:05:45,151][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:05:45,663][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:05:46,169][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:05:46,675][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:05:47,178][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:05:47,677][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:05:48,177][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:05:48,672][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:05:49,167][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:05:49,663][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:05:50,160][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:05:50,655][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:05:51,155][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:05:51,657][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:05:52,159][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:05:52,662][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:05:53,164][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:05:53,671][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:05:54,171][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:05:54,669][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:05:55,164][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:05:55,661][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:05:56,154][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:05:56,652][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:05:57,150][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:05:57,644][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:05:58,138][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:05:58,638][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:05:59,136][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:05:59,636][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:06:00,141][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:06:00,639][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:06:01,139][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:06:01,635][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:06:02,135][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:06:02,638][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:06:03,140][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:06:03,638][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:06:04,132][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:06:04,637][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:06:05,130][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:06:05,627][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:06:06,126][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:06:06,622][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:06:07,121][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9800 tokens. +[2026-03-26 09:06:08,028][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.12%, ΔTime: 00:00:32 +[2026-03-26 09:06:08,796][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:06:08,798][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:06:08,800][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:06:09,543][__main__][INFO] - Iteration 723 took 51s (29.39% Gen, 69.15% Train). Generation: 14s, Training: 35s. Estimated remaining time: 31h 25m 58s. Estimated total time: 42h 31m 19s. Time estimates for 10 more iterations: 8m 30s, 100 more iterations: 1h 25m 2s, 500 more iterations: 7h 5m 13s. +[2026-03-26 09:06:09,545][__main__][INFO] - Starting iteration 723. +[2026-03-26 09:06:09,945][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 72 and human policies 1. +[2026-03-26 09:06:09,946][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:06:20,193][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 09:06:29,958][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 09:06:31,133][__main__][INFO] - Number of regex retries in iteration 723: 2 +[2026-03-26 09:06:31,134][__main__][INFO] - agents played in iteration 723 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:06:31,897][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:06:31,921][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:06:31,945][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:06:31,969][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:06:31,969][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:06:31,970][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:06:32,840][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:06:33,292][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:06:33,792][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:06:34,292][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:06:34,790][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:06:35,288][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:06:35,789][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:06:36,287][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:06:36,791][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:06:37,291][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:06:37,782][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:06:38,277][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:06:38,774][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:06:39,266][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:06:39,761][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:06:40,256][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:06:40,758][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:06:41,259][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:06:41,759][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:06:42,259][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:06:42,759][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:06:43,257][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:06:43,764][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:06:44,262][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:06:44,759][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:06:45,260][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:06:45,760][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:06:46,260][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:06:46,765][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:06:47,272][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:06:47,780][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:06:48,290][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:06:48,790][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:06:49,293][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:06:49,795][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:06:50,322][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:06:50,816][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:06:51,316][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:06:51,814][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:06:52,312][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:06:52,813][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:06:53,316][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:06:53,813][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:06:54,312][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:06:54,809][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:06:55,306][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:06:55,804][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:06:56,300][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:06:56,798][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:06:57,292][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:06:57,788][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:06:58,286][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:06:58,782][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:06:59,287][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:06:59,785][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:07:00,287][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:07:00,788][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:07:01,280][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:07:01,780][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:07:02,275][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:07:02,775][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:07:03,274][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:07:03,770][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:07:04,266][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:07:04,763][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9755 tokens. +[2026-03-26 09:07:05,660][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.14%, ΔTime: 00:00:32 +[2026-03-26 09:07:06,417][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:07:06,419][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:07:06,421][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:07:07,192][__main__][INFO] - Iteration 724 took 57s (37.01% Gen, 61.64% Train). Generation: 21s, Training: 35s. Estimated remaining time: 36h 36m 3s. Estimated total time: 47h 42m 22s. Time estimates for 10 more iterations: 9m 32s, 100 more iterations: 1h 35m 24s, 500 more iterations: 7h 57m 3s. +[2026-03-26 09:07:07,194][__main__][INFO] - Starting iteration 724. +[2026-03-26 09:07:07,605][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 72 and human policies 1. +[2026-03-26 09:07:07,606][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:07:27,053][__main__][INFO] - Number of regex retries in iteration 724: 0 +[2026-03-26 09:07:27,054][__main__][INFO] - agents played in iteration 724 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:07:27,818][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:07:27,842][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:07:27,866][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:07:27,889][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:07:27,890][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:07:27,890][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:07:28,752][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:07:29,207][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:07:29,711][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:07:30,207][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:07:30,704][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:07:31,201][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:07:31,708][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:07:32,207][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:07:32,704][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:07:33,201][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:07:33,702][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:07:34,198][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:07:34,701][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:07:35,203][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:07:35,709][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:07:36,208][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:07:36,736][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:07:37,237][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:07:37,742][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:07:38,245][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:07:38,749][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:07:39,250][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:07:39,752][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:07:40,253][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:07:40,757][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:07:41,259][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:07:41,753][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:07:42,248][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:07:42,740][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:07:43,240][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:07:43,744][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:07:44,243][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:07:44,737][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:07:45,238][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:07:45,739][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:07:46,240][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:07:46,741][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:07:47,239][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:07:47,735][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:07:48,231][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:07:48,733][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:07:49,229][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:07:49,727][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:07:50,220][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:07:50,715][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:07:51,206][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:07:51,703][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:07:52,213][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:07:52,707][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:07:53,202][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:07:53,703][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:07:54,203][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:07:54,699][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:07:55,203][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:07:55,697][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:07:56,197][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:07:56,697][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:07:57,197][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:07:57,699][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:07:58,193][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:07:58,694][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:07:59,196][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:07:59,697][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:08:00,196][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:08:00,692][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9756 tokens. +[2026-03-26 09:08:01,596][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.03%, Current % of VRAM taken: 60.46%, Block Peak % of device VRAM: 62.25%, ΔTime: 00:00:32 +[2026-03-26 09:08:02,356][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:08:02,359][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:08:02,360][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:08:03,137][__main__][INFO] - Iteration 725 took 55s (35.02% Gen, 63.58% Train). Generation: 19s, Training: 35s. Estimated remaining time: 35h 9m 23s. Estimated total time: 46h 16m 37s. Time estimates for 10 more iterations: 9m 15s, 100 more iterations: 1h 32m 33s, 500 more iterations: 7h 42m 46s. +[2026-03-26 09:08:03,139][__main__][INFO] - Starting iteration 725. +[2026-03-26 09:08:03,540][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 72 and human policies 1. +[2026-03-26 09:08:03,540][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:08:19,093][__main__][INFO] - Number of regex retries in iteration 725: 0 +[2026-03-26 09:08:19,094][__main__][INFO] - agents played in iteration 725 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:08:19,865][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:08:19,889][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:08:19,912][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:08:19,936][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:08:19,936][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:08:19,937][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:08:20,799][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:08:21,253][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:08:21,755][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:08:22,253][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:08:22,755][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:08:23,250][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:08:23,746][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:08:24,245][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:08:24,743][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:08:25,246][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:08:25,749][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:08:26,250][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:08:26,749][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:08:27,246][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:08:27,748][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:08:28,248][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:08:28,750][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:08:29,249][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:08:29,751][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:08:30,252][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:08:30,754][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:08:31,258][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:08:31,758][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:08:32,258][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:08:32,758][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:08:33,253][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:08:33,750][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:08:34,246][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:08:34,743][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:08:35,241][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:08:35,738][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:08:36,231][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:08:36,728][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:08:37,226][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:08:37,724][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:08:38,223][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:08:38,721][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:08:39,220][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:08:39,719][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:08:40,220][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:08:40,719][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:08:41,225][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:08:41,725][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:08:42,222][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:08:42,721][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:08:43,217][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:08:43,712][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:08:44,209][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:08:44,706][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:08:45,205][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:08:45,705][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:08:46,202][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:08:46,700][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:08:47,200][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:08:47,699][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:08:48,199][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:08:48,695][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:08:49,193][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:08:49,688][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:08:50,188][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:08:50,681][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:08:51,175][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:08:51,670][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:08:52,164][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:08:52,660][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9767 tokens. +[2026-03-26 09:08:53,557][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 61.98%, ΔTime: 00:00:32 +[2026-03-26 09:08:54,303][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:08:54,306][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:08:54,308][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:08:55,103][__main__][INFO] - Iteration 726 took 51s (30.16% Gen, 68.29% Train). Generation: 15s, Training: 35s. Estimated remaining time: 31h 50m 5s. Estimated total time: 42h 58m 12s. Time estimates for 10 more iterations: 8m 35s, 100 more iterations: 1h 25m 56s, 500 more iterations: 7h 9m 42s. +[2026-03-26 09:08:55,105][__main__][INFO] - Starting iteration 726. +[2026-03-26 09:08:55,505][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 72 and human policies 1. +[2026-03-26 09:08:55,506][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:09:11,187][__main__][INFO] - Number of regex retries in iteration 726: 0 +[2026-03-26 09:09:11,188][__main__][INFO] - agents played in iteration 726 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:09:11,953][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:09:11,977][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:09:12,000][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:09:12,023][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:09:12,024][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:09:12,024][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:09:12,890][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:09:13,339][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:09:13,834][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:09:14,333][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:09:14,834][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:09:15,328][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:09:15,827][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:09:16,326][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:09:16,830][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:09:17,334][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:09:17,837][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:09:18,330][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:09:18,832][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:09:19,333][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:09:19,834][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:09:20,339][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:09:20,841][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:09:21,343][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:09:21,845][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:09:22,346][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:09:22,847][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:09:23,345][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:09:23,841][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:09:24,339][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:09:24,842][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:09:25,341][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:09:25,839][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:09:26,340][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:09:26,839][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:09:27,338][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:09:27,867][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:09:28,366][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:09:28,864][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:09:29,364][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:09:29,864][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:09:30,364][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:09:30,863][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:09:31,366][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:09:31,862][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:09:32,362][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:09:32,863][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:09:33,361][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:09:33,863][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:09:34,362][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:09:34,860][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:09:35,359][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:09:35,857][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:09:36,355][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:09:36,853][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:09:37,354][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:09:37,854][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:09:38,354][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:09:38,855][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:09:39,356][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:09:39,857][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:09:40,358][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:09:40,859][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:09:41,358][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:09:41,857][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:09:42,350][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:09:42,846][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:09:43,344][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:09:43,842][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:09:44,334][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:09:44,830][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9921 tokens. +[2026-03-26 09:09:45,729][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.19%, ΔTime: 00:00:32 +[2026-03-26 09:09:46,481][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:09:46,483][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:09:46,485][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:09:47,208][__main__][INFO] - Iteration 727 took 51s (30.33% Gen, 68.27% Train). Generation: 15s, Training: 35s. Estimated remaining time: 31h 56m 11s. Estimated total time: 43h 5m 10s. Time estimates for 10 more iterations: 8m 37s, 100 more iterations: 1h 26m 10s, 500 more iterations: 7h 10m 51s. +[2026-03-26 09:09:47,210][__main__][INFO] - Starting iteration 727. +[2026-03-26 09:09:47,611][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 72 and human policies 1. +[2026-03-26 09:09:47,612][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:09:58,304][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 09:10:03,320][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 0 balls + +This proposal is made in a way that reflects the equal but opposite strategy to both of our high-value items, avoiding any allocation and hoping to force Alice into a similar zero-proposal in the next round or changing her strategy. did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 09:10:04,322][__main__][INFO] - Number of regex retries in iteration 727: 2 +[2026-03-26 09:10:04,323][__main__][INFO] - agents played in iteration 727 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:10:05,101][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:10:05,126][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:10:05,150][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:10:05,175][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:10:05,175][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:10:05,176][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:10:06,036][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:10:06,491][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:10:06,992][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:10:07,493][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:10:07,995][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:10:08,490][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:10:08,986][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:10:09,488][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:10:09,988][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:10:10,482][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:10:10,976][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:10:11,476][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:10:11,993][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:10:12,494][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:10:12,991][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:10:13,492][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:10:13,993][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:10:14,488][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:10:14,980][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:10:15,469][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:10:15,960][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:10:16,452][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:10:16,949][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:10:17,441][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:10:17,933][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:10:18,428][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:10:18,925][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:10:19,421][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:10:19,921][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:10:20,427][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:10:20,920][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:10:21,425][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:10:21,922][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:10:22,421][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:10:22,919][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:10:23,418][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:10:23,917][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:10:24,416][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:10:24,914][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:10:25,408][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:10:25,903][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:10:26,404][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:10:26,910][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:10:27,408][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:10:27,906][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:10:28,402][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:10:28,895][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:10:29,386][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:10:29,883][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:10:30,403][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:10:30,901][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:10:31,401][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:10:31,899][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:10:32,397][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:10:32,899][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:10:33,398][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:10:33,896][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:10:34,393][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:10:34,884][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:10:35,380][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:10:35,877][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:10:36,372][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:10:36,864][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:10:37,363][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:10:37,857][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9658 tokens. +[2026-03-26 09:10:38,756][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.92%, Current % of VRAM taken: 60.35%, Block Peak % of device VRAM: 62.05%, ΔTime: 00:00:32 +[2026-03-26 09:10:39,489][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:10:39,491][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:10:39,493][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:10:40,269][__main__][INFO] - Iteration 728 took 52s (31.74% Gen, 66.79% Train). Generation: 16s, Training: 35s. Estimated remaining time: 32h 43m 1s. Estimated total time: 43h 52m 53s. Time estimates for 10 more iterations: 8m 46s, 100 more iterations: 1h 27m 45s, 500 more iterations: 7h 18m 48s. +[2026-03-26 09:10:40,273][__main__][INFO] - Starting iteration 728. +[2026-03-26 09:10:40,673][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 72 and human policies 1. +[2026-03-26 09:10:40,674][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:10:44,907][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 09:10:44,909][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 09:10:46,906][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 09:10:56,008][__main__][INFO] - Number of regex retries in iteration 728: 3 +[2026-03-26 09:10:56,009][__main__][INFO] - agents played in iteration 728 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:10:56,770][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:10:56,794][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:10:56,817][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:10:56,840][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:10:56,841][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:10:56,842][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:10:57,697][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:10:58,153][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:10:58,658][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:10:59,165][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:10:59,672][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:11:00,177][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:11:00,677][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:11:01,181][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:11:01,684][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:11:02,186][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:11:02,689][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:11:03,191][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:11:03,698][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:11:04,202][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:11:04,711][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:11:05,211][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:11:05,713][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:11:06,209][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:11:06,701][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:11:07,196][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:11:07,694][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:11:08,190][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:11:08,693][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:11:09,187][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:11:09,680][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:11:10,177][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:11:10,675][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:11:11,171][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:11:11,669][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:11:12,164][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:11:12,665][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:11:13,163][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:11:13,662][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:11:14,159][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:11:14,657][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:11:15,158][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:11:15,657][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:11:16,157][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:11:16,657][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:11:17,164][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:11:17,662][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:11:18,161][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:11:18,660][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:11:19,159][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:11:19,663][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:11:20,164][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:11:20,666][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:11:21,166][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:11:21,665][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:11:22,164][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:11:22,661][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:11:23,163][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:11:23,661][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:11:24,163][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:11:24,661][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:11:25,159][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:11:25,655][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:11:26,152][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:11:26,648][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:11:27,145][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:11:27,641][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:11:28,137][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:11:28,629][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:11:29,125][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:11:29,622][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9962 tokens. +[2026-03-26 09:11:30,514][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.05%, Current % of VRAM taken: 60.48%, Block Peak % of device VRAM: 62.24%, ΔTime: 00:00:32 +[2026-03-26 09:11:31,248][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:11:31,250][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:11:31,252][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:11:32,023][__main__][INFO] - Iteration 729 took 51s (29.86% Gen, 68.63% Train). Generation: 15s, Training: 35s. Estimated remaining time: 31h 36m 49s. Estimated total time: 42h 47m 32s. Time estimates for 10 more iterations: 8m 33s, 100 more iterations: 1h 25m 35s, 500 more iterations: 7h 7m 55s. +[2026-03-26 09:11:32,026][__main__][INFO] - Starting iteration 729. +[2026-03-26 09:11:32,426][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 72 and human policies 1. +[2026-03-26 09:11:32,427][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:11:39,000][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 0 balls Humanities, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 09:11:47,055][__main__][INFO] - Number of regex retries in iteration 729: 1 +[2026-03-26 09:11:47,056][__main__][INFO] - agents played in iteration 729 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:11:47,814][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:11:47,838][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:11:47,861][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:11:47,885][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:11:47,885][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:11:47,886][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:11:48,757][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:11:49,216][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:11:49,729][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:11:50,233][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:11:50,732][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:11:51,230][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:11:51,727][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:11:52,229][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:11:52,728][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:11:53,224][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:11:53,724][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:11:54,224][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:11:54,723][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:11:55,222][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:11:55,719][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:11:56,218][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:11:56,715][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:11:57,207][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:11:57,708][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:11:58,205][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:11:58,703][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:11:59,196][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:11:59,692][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:12:00,192][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:12:00,685][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:12:01,185][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:12:01,688][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:12:02,190][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:12:02,686][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:12:03,183][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:12:03,679][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:12:04,178][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:12:04,674][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:12:05,172][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:12:05,668][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:12:06,167][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:12:06,665][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:12:07,163][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:12:07,663][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:12:08,164][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:12:08,663][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:12:09,161][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:12:09,659][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:12:10,158][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:12:10,656][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:12:11,154][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:12:11,650][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:12:12,147][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:12:12,644][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:12:13,141][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:12:13,633][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:12:14,122][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:12:14,618][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:12:15,116][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:12:15,607][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:12:16,103][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:12:16,603][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:12:17,095][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:12:17,587][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:12:18,078][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:12:18,576][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:12:19,068][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:12:19,559][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:12:20,052][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:12:20,554][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9662 tokens. +[2026-03-26 09:12:21,449][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.12%, ΔTime: 00:00:32 +[2026-03-26 09:12:22,214][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:12:22,216][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:12:22,218][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:12:22,962][__main__][INFO] - Iteration 730 took 50s (28.95% Gen, 69.58% Train). Generation: 14s, Training: 35s. Estimated remaining time: 30h 55m 13s. Estimated total time: 42h 6m 47s. Time estimates for 10 more iterations: 8m 25s, 100 more iterations: 1h 24m 13s, 500 more iterations: 7h 1m 7s. +[2026-03-26 09:12:22,964][__main__][INFO] - Starting iteration 730. +[2026-03-26 09:12:23,363][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 72 and human policies 1. +[2026-03-26 09:12:23,364][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:12:27,867][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 09:12:39,678][__main__][INFO] - Number of regex retries in iteration 730: 1 +[2026-03-26 09:12:39,679][__main__][INFO] - agents played in iteration 730 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:12:40,456][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:12:40,480][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:12:40,503][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:12:40,527][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:12:40,527][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:12:40,528][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:12:41,390][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:12:41,841][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:12:42,346][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:12:42,847][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:12:43,347][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:12:43,845][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:12:44,346][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:12:44,842][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:12:45,345][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:12:45,848][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:12:46,350][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:12:46,852][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:12:47,354][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:12:47,853][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:12:48,351][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:12:48,848][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:12:49,346][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:12:49,839][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:12:50,361][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:12:50,855][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:12:51,356][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:12:51,847][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:12:52,343][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:12:52,837][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:12:53,332][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:12:53,827][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:12:54,325][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:12:54,819][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:12:55,315][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:12:55,811][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:12:56,305][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:12:56,800][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:12:57,294][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:12:57,790][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:12:58,289][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:12:58,785][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:12:59,278][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:12:59,772][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:13:00,266][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:13:00,762][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:13:01,258][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:13:01,750][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:13:02,241][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:13:02,732][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:13:03,235][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:13:03,726][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:13:04,226][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:13:04,717][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:13:05,213][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:13:05,703][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:13:06,198][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:13:06,688][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:13:07,183][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:13:07,672][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:13:08,167][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:13:08,656][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:13:09,152][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:13:09,648][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:13:10,143][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:13:10,637][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:13:11,128][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:13:11,623][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:13:12,115][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:13:12,605][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:13:13,101][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9683 tokens. +[2026-03-26 09:13:13,993][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.01%, Current % of VRAM taken: 60.44%, Block Peak % of device VRAM: 61.96%, ΔTime: 00:00:32 +[2026-03-26 09:13:14,809][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:13:14,812][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:13:14,814][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:13:16,459][__main__][INFO] - Iteration 731 took 53s (30.73% Gen, 66.17% Train). Generation: 16s, Training: 35s. Estimated remaining time: 33h 2m 23s. Estimated total time: 44h 14m 51s. Time estimates for 10 more iterations: 8m 50s, 100 more iterations: 1h 28m 29s, 500 more iterations: 7h 22m 28s. +[2026-03-26 09:13:16,462][__main__][INFO] - Starting iteration 731. +[2026-03-26 09:13:16,864][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 73 and human policies 1. +[2026-03-26 09:13:16,864][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:13:32,389][__main__][INFO] - Number of regex retries in iteration 731: 0 +[2026-03-26 09:13:32,390][__main__][INFO] - agents played in iteration 731 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:13:33,170][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:13:33,194][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:13:33,217][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:13:33,241][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:13:33,242][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:13:33,243][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:13:34,104][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:13:34,560][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:13:35,068][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:13:35,570][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:13:36,085][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:13:36,588][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:13:37,098][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:13:37,601][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:13:38,105][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:13:38,608][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:13:39,111][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:13:39,623][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:13:40,124][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:13:40,624][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:13:41,118][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:13:41,616][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:13:42,116][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:13:42,613][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:13:43,104][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:13:43,595][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:13:44,089][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:13:44,589][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:13:45,083][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:13:45,580][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:13:46,077][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:13:46,578][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:13:47,076][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:13:47,571][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:13:48,068][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:13:48,572][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:13:49,070][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:13:49,573][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:13:50,070][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:13:50,566][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:13:51,064][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:13:51,563][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:13:52,064][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:13:52,563][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:13:53,057][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:13:53,554][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:13:54,050][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:13:54,543][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:13:55,038][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:13:55,537][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:13:56,036][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:13:56,535][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:13:57,028][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:13:57,528][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:13:58,021][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:13:58,514][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:13:59,009][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:13:59,505][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:14:00,001][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:14:00,498][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:14:00,994][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:14:01,490][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:14:01,985][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:14:02,485][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:14:02,983][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:14:03,478][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:14:03,974][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:14:04,473][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:14:04,973][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:14:05,467][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:14:05,966][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9744 tokens. +[2026-03-26 09:14:06,849][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.18%, Current % of VRAM taken: 60.61%, Block Peak % of device VRAM: 62.14%, ΔTime: 00:00:32 +[2026-03-26 09:14:07,558][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:14:07,560][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:14:07,562][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:14:08,285][__main__][INFO] - Iteration 732 took 51s (30.19% Gen, 68.40% Train). Generation: 15s, Training: 35s. Estimated remaining time: 31h 37m 45s. Estimated total time: 42h 51m 5s. Time estimates for 10 more iterations: 8m 34s, 100 more iterations: 1h 25m 42s, 500 more iterations: 7h 8m 30s. +[2026-03-26 09:14:08,287][__main__][INFO] - Starting iteration 732. +[2026-03-26 09:14:08,687][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 73 and human policies 1. +[2026-03-26 09:14:08,688][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:14:14,787][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 09:14:23,038][__main__][INFO] - Number of regex retries in iteration 732: 1 +[2026-03-26 09:14:23,039][__main__][INFO] - agents played in iteration 732 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:14:23,807][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:14:23,833][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:14:23,858][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:14:23,884][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:14:23,884][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:14:23,885][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:14:24,751][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:14:25,206][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:14:25,723][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:14:26,220][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:14:26,719][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:14:27,223][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:14:27,731][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:14:28,241][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:14:28,738][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:14:29,244][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:14:29,749][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:14:30,253][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:14:30,759][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:14:31,264][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:14:31,768][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:14:32,278][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:14:32,780][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:14:33,281][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:14:33,780][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:14:34,279][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:14:34,779][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:14:35,278][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:14:35,775][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:14:36,272][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:14:36,770][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:14:37,271][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:14:37,768][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:14:38,270][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:14:38,765][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:14:39,261][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:14:39,767][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:14:40,261][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:14:40,758][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:14:41,258][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:14:41,757][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:14:42,256][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:14:42,757][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:14:43,259][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:14:43,759][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:14:44,258][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:14:44,756][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:14:45,253][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:14:45,748][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:14:46,248][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:14:46,747][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:14:47,246][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:14:47,749][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:14:48,256][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:14:48,758][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:14:49,254][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:14:49,749][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:14:50,245][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:14:50,745][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:14:51,239][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:14:51,731][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:14:52,225][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:14:52,722][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:14:53,220][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:14:53,719][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:14:54,219][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:14:54,718][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:14:55,217][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:14:55,708][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:14:56,206][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:14:56,703][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9773 tokens. +[2026-03-26 09:14:57,595][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 61.98%, ΔTime: 00:00:32 +[2026-03-26 09:14:58,328][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:14:58,330][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:14:58,332][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:14:59,111][__main__][INFO] - Iteration 733 took 50s (28.46% Gen, 69.99% Train). Generation: 14s, Training: 35s. Estimated remaining time: 30h 47m 3s. Estimated total time: 42h 1m 13s. Time estimates for 10 more iterations: 8m 24s, 100 more iterations: 1h 24m 2s, 500 more iterations: 7h 0m 12s. +[2026-03-26 09:14:59,114][__main__][INFO] - Starting iteration 733. +[2026-03-26 09:14:59,514][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 73 and human policies 1. +[2026-03-26 09:14:59,515][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:15:15,289][__main__][INFO] - Number of regex retries in iteration 733: 0 +[2026-03-26 09:15:15,289][__main__][INFO] - agents played in iteration 733 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:15:16,064][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:15:16,087][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:15:16,111][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:15:16,135][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:15:16,135][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:15:16,136][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:15:17,008][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:15:17,467][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:15:17,973][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:15:18,473][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:15:18,971][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:15:19,473][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:15:19,973][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:15:20,474][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:15:20,973][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:15:21,468][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:15:21,969][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:15:22,477][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:15:22,978][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:15:23,472][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:15:23,969][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:15:24,469][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:15:24,982][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:15:25,505][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:15:26,010][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:15:26,511][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:15:27,012][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:15:27,515][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:15:28,016][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:15:28,518][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:15:29,019][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:15:29,515][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:15:30,011][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:15:30,508][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:15:31,001][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:15:31,498][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:15:31,996][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:15:32,488][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:15:32,984][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:15:33,477][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:15:33,975][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:15:34,467][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:15:34,964][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:15:35,463][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:15:35,961][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:15:36,452][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:15:36,949][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:15:37,447][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:15:37,945][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:15:38,443][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:15:38,938][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:15:39,445][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:15:39,944][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:15:40,443][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:15:40,936][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:15:41,429][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:15:41,927][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:15:42,420][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:15:42,916][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:15:43,410][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:15:43,904][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:15:44,405][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:15:44,899][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:15:45,397][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:15:45,891][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:15:46,388][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:15:46,886][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:15:47,384][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:15:47,882][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:15:48,375][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:15:48,871][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9722 tokens. +[2026-03-26 09:15:49,752][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.31%, ΔTime: 00:00:32 +[2026-03-26 09:15:50,498][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:15:50,500][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:15:50,502][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:15:51,258][__main__][INFO] - Iteration 734 took 51s (30.49% Gen, 68.05% Train). Generation: 15s, Training: 35s. Estimated remaining time: 31h 52m 12s. Estimated total time: 43h 7m 15s. Time estimates for 10 more iterations: 8m 37s, 100 more iterations: 1h 26m 14s, 500 more iterations: 7h 11m 12s. +[2026-03-26 09:15:51,261][__main__][INFO] - Starting iteration 734. +[2026-03-26 09:15:51,660][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 73 and human policies 1. +[2026-03-26 09:15:51,660][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:16:03,008][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 09:16:07,041][__main__][INFO] - Number of regex retries in iteration 734: 1 +[2026-03-26 09:16:07,042][__main__][INFO] - agents played in iteration 734 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:16:07,837][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:16:07,860][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:16:07,884][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:16:07,907][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.55%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:16:07,908][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:16:07,908][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:16:08,781][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:16:09,234][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:16:09,740][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:16:10,243][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:16:10,746][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:16:11,249][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:16:11,752][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:16:12,254][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:16:12,757][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:16:13,253][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:16:13,755][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:16:14,256][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:16:14,749][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:16:15,244][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:16:15,739][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:16:16,231][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:16:16,725][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:16:17,217][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:16:17,713][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:16:18,215][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:16:18,712][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:16:19,203][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:16:19,694][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:16:20,194][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:16:20,685][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:16:21,188][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:16:21,686][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:16:22,181][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:16:22,679][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:16:23,178][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:16:23,676][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:16:24,171][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:16:24,670][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:16:25,170][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:16:25,669][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:16:26,164][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:16:26,660][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:16:27,156][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:16:27,652][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:16:28,150][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:16:28,644][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:16:29,140][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:16:29,638][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:16:30,138][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:16:30,637][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:16:31,136][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:16:31,635][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:16:32,133][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:16:32,630][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:16:33,128][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:16:33,626][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:16:34,125][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:16:34,623][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:16:35,122][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:16:35,621][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:16:36,119][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:16:36,621][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:16:37,119][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:16:37,614][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:16:38,106][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:16:38,597][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:16:39,097][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:16:39,596][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:16:40,088][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:16:40,590][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9745 tokens. +[2026-03-26 09:16:41,484][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 61.96%, ΔTime: 00:00:32 +[2026-03-26 09:16:42,284][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:16:42,287][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:16:42,288][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:16:43,069][__main__][INFO] - Iteration 735 took 51s (29.92% Gen, 68.56% Train). Generation: 15s, Training: 35s. Estimated remaining time: 31h 34m 36s. Estimated total time: 42h 50m 30s. Time estimates for 10 more iterations: 8m 34s, 100 more iterations: 1h 25m 41s, 500 more iterations: 7h 8m 25s. +[2026-03-26 09:16:43,072][__main__][INFO] - Starting iteration 735. +[2026-03-26 09:16:43,472][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 73 and human policies 1. +[2026-03-26 09:16:43,473][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:16:57,975][__main__][INFO] - Number of regex retries in iteration 735: 0 +[2026-03-26 09:16:57,976][__main__][INFO] - agents played in iteration 735 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:16:58,738][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:16:58,762][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:16:58,786][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:16:58,809][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:16:58,810][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:16:58,811][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:16:59,670][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:17:00,126][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:17:00,629][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:17:01,129][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:17:01,632][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:17:02,132][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:17:02,634][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:17:03,137][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:17:03,639][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:17:04,146][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:17:04,645][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:17:05,149][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:17:05,654][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:17:06,158][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:17:06,661][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:17:07,161][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:17:07,661][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:17:08,159][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:17:08,657][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:17:09,156][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:17:09,653][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:17:10,150][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:17:10,648][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:17:11,146][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:17:11,644][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:17:12,137][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:17:12,630][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:17:13,123][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:17:13,617][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:17:14,109][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:17:14,604][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:17:15,096][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:17:15,589][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:17:16,088][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:17:16,588][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:17:17,085][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:17:17,585][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:17:18,085][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:17:18,588][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:17:19,090][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:17:19,590][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:17:20,088][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:17:20,585][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:17:21,083][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:17:21,581][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:17:22,078][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:17:22,577][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:17:23,074][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:17:23,570][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:17:24,067][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:17:24,561][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:17:25,057][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:17:25,553][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:17:26,048][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:17:26,544][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:17:27,041][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:17:27,538][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:17:28,034][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:17:28,533][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:17:29,032][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:17:29,530][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:17:30,029][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:17:30,530][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:17:31,030][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:17:31,528][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9860 tokens. +[2026-03-26 09:17:32,437][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 09:17:33,185][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:17:33,188][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:17:33,189][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:17:33,947][__main__][INFO] - Iteration 736 took 50s (28.73% Gen, 69.76% Train). Generation: 14s, Training: 35s. Estimated remaining time: 30h 47m 1s. Estimated total time: 42h 3m 46s. Time estimates for 10 more iterations: 8m 24s, 100 more iterations: 1h 24m 7s, 500 more iterations: 7h 0m 37s. +[2026-03-26 09:17:33,949][__main__][INFO] - Starting iteration 736. +[2026-03-26 09:17:34,349][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 73 and human policies 1. +[2026-03-26 09:17:34,350][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:17:49,481][__main__][INFO] - Number of regex retries in iteration 736: 0 +[2026-03-26 09:17:49,482][__main__][INFO] - agents played in iteration 736 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:17:50,251][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:17:50,274][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:17:50,297][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:17:50,321][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:17:50,321][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:17:50,322][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:17:51,172][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:17:51,632][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:17:52,140][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:17:52,643][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:17:53,147][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:17:53,645][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:17:54,147][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:17:54,644][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:17:55,140][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:17:55,644][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:17:56,148][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:17:56,655][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:17:57,159][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:17:57,661][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:17:58,159][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:17:58,657][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:17:59,155][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:17:59,654][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:18:00,154][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:18:00,651][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:18:01,147][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:18:01,641][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:18:02,140][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:18:02,660][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:18:03,165][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:18:03,663][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:18:04,163][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:18:04,666][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:18:05,166][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:18:05,669][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:18:06,168][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:18:06,666][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:18:07,165][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:18:07,664][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:18:08,165][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:18:08,669][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:18:09,168][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:18:09,669][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:18:10,172][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:18:10,678][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:18:11,177][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:18:11,670][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:18:12,164][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:18:12,656][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:18:13,148][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:18:13,639][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:18:14,136][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:18:14,637][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:18:15,135][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:18:15,627][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:18:16,125][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:18:16,624][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:18:17,117][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:18:17,609][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:18:18,100][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:18:18,592][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:18:19,088][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:18:19,585][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:18:20,088][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:18:20,587][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:18:21,085][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:18:21,583][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:18:22,082][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:18:22,580][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:18:23,081][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9804 tokens. +[2026-03-26 09:18:23,975][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.20%, Current % of VRAM taken: 60.63%, Block Peak % of device VRAM: 62.06%, ΔTime: 00:00:32 +[2026-03-26 09:18:24,731][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:18:24,733][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:18:24,735][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:18:25,596][__main__][INFO] - Iteration 737 took 51s (29.53% Gen, 68.79% Train). Generation: 15s, Training: 35s. Estimated remaining time: 31h 24m 47s. Estimated total time: 42h 42m 24s. Time estimates for 10 more iterations: 8m 32s, 100 more iterations: 1h 25m 24s, 500 more iterations: 7h 7m 4s. +[2026-03-26 09:18:25,599][__main__][INFO] - Starting iteration 737. +[2026-03-26 09:18:25,997][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 73 and human policies 1. +[2026-03-26 09:18:25,998][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:18:43,032][__main__][INFO] - Number of regex retries in iteration 737: 0 +[2026-03-26 09:18:43,033][__main__][INFO] - agents played in iteration 737 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:18:43,847][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:18:43,870][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:18:43,894][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:18:43,918][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:18:43,919][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:18:43,919][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:18:44,810][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:18:45,265][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:18:45,778][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:18:46,279][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:18:46,784][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:18:47,281][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:18:47,785][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:18:48,287][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:18:48,794][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:18:49,293][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:18:49,790][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:18:50,288][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:18:50,787][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:18:51,285][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:18:51,786][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:18:52,284][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:18:52,783][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:18:53,277][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:18:53,779][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:18:54,277][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:18:54,776][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:18:55,279][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:18:55,777][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:18:56,269][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:18:56,761][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:18:57,259][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:18:57,758][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:18:58,256][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:18:58,755][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:18:59,253][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:18:59,751][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:19:00,250][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:19:00,751][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:19:01,247][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:19:01,748][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:19:02,250][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:19:02,746][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:19:03,244][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:19:03,739][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:19:04,240][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:19:04,733][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:19:05,232][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:19:05,731][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:19:06,231][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:19:06,731][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:19:07,231][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:19:07,731][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:19:08,231][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:19:08,730][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:19:09,228][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:19:09,722][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:19:10,215][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:19:10,708][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:19:11,206][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:19:11,700][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:19:12,192][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:19:12,694][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:19:13,191][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:19:13,690][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:19:14,189][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:19:14,687][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:19:15,186][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:19:15,685][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:19:16,179][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:19:16,677][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9774 tokens. +[2026-03-26 09:19:17,569][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.57%, Block Peak % of device VRAM: 62.06%, ΔTime: 00:00:32 +[2026-03-26 09:19:18,345][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:19:18,347][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:19:18,349][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:19:19,090][__main__][INFO] - Iteration 738 took 53s (32.08% Gen, 66.52% Train). Generation: 17s, Training: 35s. Estimated remaining time: 32h 56m 10s. Estimated total time: 44h 14m 41s. Time estimates for 10 more iterations: 8m 50s, 100 more iterations: 1h 28m 29s, 500 more iterations: 7h 22m 26s. +[2026-03-26 09:19:19,092][__main__][INFO] - Starting iteration 738. +[2026-03-26 09:19:19,490][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 73 and human policies 1. +[2026-03-26 09:19:19,491][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:19:33,737][__main__][INFO] - Number of regex retries in iteration 738: 0 +[2026-03-26 09:19:33,738][__main__][INFO] - agents played in iteration 738 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:19:34,500][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:19:34,524][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:19:34,548][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:19:34,571][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:19:34,572][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:19:34,572][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:19:35,426][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:19:35,882][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:19:36,396][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:19:36,901][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:19:37,405][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:19:37,905][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:19:38,409][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:19:38,915][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:19:39,419][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:19:39,921][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:19:40,424][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:19:40,926][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:19:41,429][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:19:41,935][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:19:42,434][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:19:42,935][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:19:43,433][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:19:43,932][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:19:44,430][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:19:44,927][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:19:45,425][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:19:45,919][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:19:46,419][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:19:46,918][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:19:47,424][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:19:47,923][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:19:48,422][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:19:48,921][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:19:49,420][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:19:49,919][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:19:50,415][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:19:50,912][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:19:51,411][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:19:51,904][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:19:52,400][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:19:52,894][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:19:53,389][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:19:53,883][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:19:54,377][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:19:54,868][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:19:55,367][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:19:55,864][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:19:56,360][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:19:56,858][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:19:57,353][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:19:57,850][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:19:58,346][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:19:58,843][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:19:59,340][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:19:59,832][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:20:00,325][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:20:00,822][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:20:01,318][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:20:01,816][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:20:02,315][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:20:02,812][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:20:03,309][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:20:03,806][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:20:04,303][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:20:04,802][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:20:05,300][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:20:05,797][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:20:06,297][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:20:06,796][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:20:07,294][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9786 tokens. +[2026-03-26 09:20:08,181][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 61.92%, ΔTime: 00:00:32 +[2026-03-26 09:20:08,937][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:20:08,939][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:20:08,940][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:20:09,698][__main__][INFO] - Iteration 739 took 50s (28.38% Gen, 70.11% Train). Generation: 14s, Training: 35s. Estimated remaining time: 30h 31m 4s. Estimated total time: 41h 50m 25s. Time estimates for 10 more iterations: 8m 22s, 100 more iterations: 1h 23m 40s, 500 more iterations: 6h 58m 24s. +[2026-03-26 09:20:09,700][__main__][INFO] - Starting iteration 739. +[2026-03-26 09:20:10,100][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 73 and human policies 1. +[2026-03-26 09:20:10,100][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:20:18,941][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 09:20:27,426][__main__][INFO] - Number of regex retries in iteration 739: 1 +[2026-03-26 09:20:27,427][__main__][INFO] - agents played in iteration 739 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:20:28,218][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:20:28,242][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:20:28,266][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:20:28,290][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:20:28,290][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:20:28,291][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:20:29,166][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:20:29,621][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:20:30,133][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:20:30,633][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:20:31,134][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:20:31,636][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:20:32,134][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:20:32,639][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:20:33,135][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:20:33,633][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:20:34,128][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:20:34,624][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:20:35,120][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:20:35,617][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:20:36,113][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:20:36,609][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:20:37,106][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:20:37,603][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:20:38,098][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:20:38,598][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:20:39,094][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:20:39,591][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:20:40,087][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:20:40,581][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:20:41,076][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:20:41,575][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:20:42,075][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:20:42,577][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:20:43,080][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:20:43,580][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:20:44,083][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:20:44,588][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:20:45,089][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:20:45,588][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:20:46,088][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:20:46,579][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:20:47,070][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:20:47,566][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:20:48,068][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:20:48,566][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:20:49,060][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:20:49,557][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:20:50,055][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:20:50,553][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:20:51,050][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:20:51,547][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:20:52,044][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:20:52,541][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:20:53,038][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:20:53,535][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:20:54,032][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:20:54,526][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:20:55,026][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:20:55,519][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:20:56,013][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:20:56,506][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:20:57,004][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:20:57,504][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:20:58,003][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:20:58,502][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:20:59,007][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:20:59,507][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:21:00,007][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:21:00,507][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:21:01,006][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9896 tokens. +[2026-03-26 09:21:01,889][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.33%, ΔTime: 00:00:32 +[2026-03-26 09:21:02,645][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:21:02,647][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:21:02,648][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:21:03,379][__main__][INFO] - Iteration 740 took 53s (32.52% Gen, 66.11% Train). Generation: 17s, Training: 35s. Estimated remaining time: 33h 3m 46s. Estimated total time: 44h 24m 1s. Time estimates for 10 more iterations: 8m 52s, 100 more iterations: 1h 28m 48s, 500 more iterations: 7h 24m 0s. +[2026-03-26 09:21:03,382][__main__][INFO] - Starting iteration 740. +[2026-03-26 09:21:03,782][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 73 and human policies 1. +[2026-03-26 09:21:03,783][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:21:18,016][__main__][INFO] - Number of regex retries in iteration 740: 0 +[2026-03-26 09:21:18,017][__main__][INFO] - agents played in iteration 740 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:21:18,812][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:21:18,836][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:21:18,860][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:21:18,883][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:21:18,884][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:21:18,884][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:21:19,772][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:21:20,231][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:21:20,735][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:21:21,230][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:21:21,729][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:21:22,230][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:21:22,731][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:21:23,231][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:21:23,732][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:21:24,228][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:21:24,724][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:21:25,217][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:21:25,709][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:21:26,204][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:21:26,697][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:21:27,190][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:21:27,688][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:21:28,185][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:21:28,682][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:21:29,179][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:21:29,679][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:21:30,177][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:21:30,672][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:21:31,169][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:21:31,666][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:21:32,169][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:21:32,670][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:21:33,166][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:21:33,666][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:21:34,167][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:21:34,667][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:21:35,163][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:21:35,666][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:21:36,162][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:21:36,662][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:21:37,161][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:21:37,656][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:21:38,153][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:21:38,650][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:21:39,146][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:21:39,643][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:21:40,139][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:21:40,635][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:21:41,129][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:21:41,620][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:21:42,118][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:21:42,618][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:21:43,124][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:21:43,623][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:21:44,121][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:21:44,618][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:21:45,118][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:21:45,616][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:21:46,115][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:21:46,613][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:21:47,110][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:21:47,608][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:21:48,107][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:21:48,607][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:21:49,127][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:21:49,627][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:21:50,129][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:21:50,630][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:21:51,130][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:21:51,630][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9770 tokens. +[2026-03-26 09:21:52,512][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.17%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.05%, ΔTime: 00:00:32 +[2026-03-26 09:21:53,272][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:21:53,274][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:21:53,276][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:21:54,804][__main__][INFO] - Iteration 741 took 51s (27.90% Gen, 69.11% Train). Generation: 14s, Training: 35s. Estimated remaining time: 31h 9m 59s. Estimated total time: 42h 31m 5s. Time estimates for 10 more iterations: 8m 30s, 100 more iterations: 1h 25m 2s, 500 more iterations: 7h 5m 10s. +[2026-03-26 09:21:54,806][__main__][INFO] - Starting iteration 741. +[2026-03-26 09:21:55,204][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 74 and human policies 1. +[2026-03-26 09:21:55,205][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:22:11,247][__main__][INFO] - Number of regex retries in iteration 741: 0 +[2026-03-26 09:22:11,248][__main__][INFO] - agents played in iteration 741 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:22:12,033][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:22:12,057][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:22:12,081][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:22:12,105][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:22:12,106][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:22:12,106][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:22:12,997][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:22:13,446][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:22:13,950][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:22:14,446][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:22:14,947][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:22:15,444][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:22:15,947][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:22:16,449][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:22:16,944][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:22:17,441][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:22:17,942][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:22:18,440][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:22:18,937][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:22:19,438][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:22:19,936][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:22:20,436][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:22:20,957][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:22:21,448][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:22:21,949][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:22:22,447][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:22:22,944][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:22:23,440][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:22:23,937][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:22:24,429][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:22:24,925][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:22:25,421][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:22:25,914][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:22:26,412][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:22:26,913][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:22:27,415][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:22:27,912][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:22:28,412][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:22:28,909][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:22:29,407][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:22:29,900][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:22:30,397][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:22:30,897][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:22:31,394][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:22:31,890][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:22:32,387][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:22:32,880][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:22:33,387][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:22:33,883][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:22:34,380][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:22:34,878][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:22:35,379][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:22:35,877][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:22:36,376][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:22:36,871][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:22:37,369][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:22:37,868][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:22:38,367][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:22:38,863][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:22:39,360][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:22:39,859][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:22:40,357][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:22:40,858][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:22:41,354][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:22:41,853][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:22:42,351][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:22:42,850][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:22:43,349][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:22:43,847][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:22:44,350][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:22:44,847][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9789 tokens. +[2026-03-26 09:22:45,742][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.54%, Block Peak % of device VRAM: 62.16%, ΔTime: 00:00:32 +[2026-03-26 09:22:46,497][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:22:46,500][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:22:46,502][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:22:47,226][__main__][INFO] - Iteration 742 took 52s (30.84% Gen, 67.77% Train). Generation: 16s, Training: 35s. Estimated remaining time: 31h 59m 7s. Estimated total time: 43h 21m 6s. Time estimates for 10 more iterations: 8m 40s, 100 more iterations: 1h 26m 42s, 500 more iterations: 7h 13m 31s. +[2026-03-26 09:22:47,228][__main__][INFO] - Starting iteration 742. +[2026-03-26 09:22:47,628][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 74 and human policies 1. +[2026-03-26 09:22:47,629][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:22:58,900][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 09:23:02,242][__main__][INFO] - Number of regex retries in iteration 742: 1 +[2026-03-26 09:23:02,243][__main__][INFO] - agents played in iteration 742 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:23:03,046][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:23:03,070][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:23:03,094][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:23:03,119][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:23:03,119][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:23:03,120][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:23:04,014][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:23:04,478][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:23:04,989][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:23:05,497][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:23:06,002][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:23:06,507][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:23:07,012][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:23:07,526][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:23:08,031][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:23:08,524][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:23:09,024][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:23:09,523][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:23:10,023][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:23:10,523][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:23:11,020][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:23:11,524][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:23:12,023][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:23:12,529][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:23:13,028][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:23:13,528][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:23:14,027][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:23:14,526][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:23:15,025][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:23:15,526][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:23:16,029][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:23:16,528][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:23:17,029][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:23:17,533][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:23:18,033][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:23:18,537][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:23:19,038][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:23:19,540][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:23:20,043][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:23:20,545][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:23:21,046][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:23:21,553][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:23:22,054][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:23:22,553][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:23:23,052][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:23:23,552][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:23:24,051][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:23:24,546][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:23:25,041][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:23:25,539][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:23:26,037][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:23:26,534][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:23:27,031][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:23:27,528][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:23:28,025][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:23:28,524][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:23:29,028][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:23:29,525][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:23:30,023][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:23:30,523][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:23:31,022][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:23:31,520][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:23:32,019][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:23:32,517][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:23:33,019][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:23:33,513][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:23:34,010][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:23:34,507][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:23:35,014][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:23:35,512][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:23:36,005][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9948 tokens. +[2026-03-26 09:23:36,907][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.92%, Current % of VRAM taken: 60.35%, Block Peak % of device VRAM: 62.06%, ΔTime: 00:00:32 +[2026-03-26 09:23:37,661][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:23:37,664][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:23:37,665][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:23:38,399][__main__][INFO] - Iteration 743 took 50s (28.78% Gen, 69.77% Train). Generation: 14s, Training: 35s. Estimated remaining time: 30h 55m 44s. Estimated total time: 42h 18m 33s. Time estimates for 10 more iterations: 8m 27s, 100 more iterations: 1h 24m 37s, 500 more iterations: 7h 3m 5s. +[2026-03-26 09:23:38,401][__main__][INFO] - Starting iteration 743. +[2026-03-26 09:23:38,802][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 74 and human policies 1. +[2026-03-26 09:23:38,802][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:23:51,551][__main__][INFO] - Number of regex retries in iteration 743: 0 +[2026-03-26 09:23:51,552][__main__][INFO] - agents played in iteration 743 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:23:52,319][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:23:52,343][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:23:52,367][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:23:52,391][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:23:52,392][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:23:52,392][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:23:53,270][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:23:53,721][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:23:54,226][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:23:54,721][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:23:55,222][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:23:55,720][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:23:56,226][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:23:56,727][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:23:57,223][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:23:57,725][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:23:58,227][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:23:58,729][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:23:59,230][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:23:59,732][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:24:00,235][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:24:00,735][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:24:01,235][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:24:01,735][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:24:02,235][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:24:02,753][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:24:03,258][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:24:03,758][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:24:04,254][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:24:04,753][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:24:05,252][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:24:05,752][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:24:06,244][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:24:06,745][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:24:07,239][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:24:07,732][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:24:08,224][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:24:08,716][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:24:09,207][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:24:09,707][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:24:10,210][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:24:10,710][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:24:11,212][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:24:11,712][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:24:12,219][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:24:12,719][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:24:13,219][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:24:13,717][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:24:14,217][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:24:14,722][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:24:15,220][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:24:15,719][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:24:16,217][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:24:16,717][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:24:17,218][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:24:17,715][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:24:18,214][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:24:18,710][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:24:19,205][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:24:19,699][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:24:20,199][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:24:20,697][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:24:21,195][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:24:21,697][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:24:22,199][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:24:22,694][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:24:23,188][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:24:23,683][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:24:24,182][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:24:24,678][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:24:25,173][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9809 tokens. +[2026-03-26 09:24:26,070][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.98%, Current % of VRAM taken: 60.41%, Block Peak % of device VRAM: 62.12%, ΔTime: 00:00:32 +[2026-03-26 09:24:26,839][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:24:26,842][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:24:26,843][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:24:27,583][__main__][INFO] - Iteration 744 took 48s (26.14% Gen, 72.35% Train). Generation: 12s, Training: 35s. Estimated remaining time: 29h 15m 27s. Estimated total time: 40h 39m 6s. Time estimates for 10 more iterations: 8m 7s, 100 more iterations: 1h 21m 18s, 500 more iterations: 6h 46m 31s. +[2026-03-26 09:24:27,585][__main__][INFO] - Starting iteration 744. +[2026-03-26 09:24:27,987][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 74 and human policies 1. +[2026-03-26 09:24:27,988][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:24:41,720][__main__][INFO] - Number of regex retries in iteration 744: 0 +[2026-03-26 09:24:41,721][__main__][INFO] - agents played in iteration 744 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:24:42,477][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:24:42,500][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:24:42,524][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:24:42,547][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:24:42,548][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:24:42,548][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:24:43,411][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:24:43,863][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:24:44,370][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:24:44,873][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:24:45,381][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:24:45,883][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:24:46,387][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:24:46,888][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:24:47,390][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:24:47,897][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:24:48,402][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:24:48,905][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:24:49,409][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:24:49,910][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:24:50,413][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:24:50,907][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:24:51,401][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:24:51,900][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:24:52,400][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:24:52,900][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:24:53,399][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:24:53,900][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:24:54,399][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:24:54,899][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:24:55,399][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:24:55,899][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:24:56,398][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:24:56,897][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:24:57,395][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:24:57,893][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:24:58,391][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:24:58,889][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:24:59,387][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:24:59,885][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:25:00,383][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:25:00,880][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:25:01,378][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:25:01,875][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:25:02,377][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:25:02,874][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:25:03,370][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:25:03,867][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:25:04,365][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:25:04,866][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:25:05,364][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:25:05,872][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:25:06,370][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:25:06,869][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:25:07,367][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:25:07,864][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:25:08,360][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:25:08,861][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:25:09,361][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:25:09,860][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:25:10,357][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:25:10,851][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:25:11,346][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:25:11,847][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:25:12,348][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:25:12,844][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:25:13,345][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:25:13,845][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:25:14,347][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:25:14,847][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:25:15,347][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9816 tokens. +[2026-03-26 09:25:16,248][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.13%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 61.90%, ΔTime: 00:00:32 +[2026-03-26 09:25:17,067][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:25:17,069][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:25:17,071][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:25:17,824][__main__][INFO] - Iteration 745 took 49s (27.56% Gen, 70.93% Train). Generation: 13s, Training: 35s. Estimated remaining time: 30h 7m 24s. Estimated total time: 41h 31m 54s. Time estimates for 10 more iterations: 8m 18s, 100 more iterations: 1h 23m 3s, 500 more iterations: 6h 55m 19s. +[2026-03-26 09:25:17,827][__main__][INFO] - Starting iteration 745. +[2026-03-26 09:25:18,226][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 74 and human policies 1. +[2026-03-26 09:25:18,227][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:25:32,891][__main__][INFO] - Number of regex retries in iteration 745: 0 +[2026-03-26 09:25:32,892][__main__][INFO] - agents played in iteration 745 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:25:33,648][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:25:33,672][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:25:33,696][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:25:33,720][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:25:33,720][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:25:33,721][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:25:34,588][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:25:35,046][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:25:35,556][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:25:36,058][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:25:36,560][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:25:37,060][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:25:37,566][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:25:38,074][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:25:38,573][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:25:39,071][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:25:39,575][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:25:40,071][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:25:40,569][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:25:41,068][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:25:41,567][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:25:42,067][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:25:42,569][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:25:43,069][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:25:43,571][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:25:44,074][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:25:44,576][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:25:45,076][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:25:45,574][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:25:46,073][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:25:46,572][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:25:47,070][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:25:47,569][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:25:48,068][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:25:48,563][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:25:49,060][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:25:49,559][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:25:50,052][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:25:50,551][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:25:51,054][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:25:51,552][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:25:52,054][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:25:52,555][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:25:53,055][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:25:53,556][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:25:54,057][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:25:54,553][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:25:55,052][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:25:55,552][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:25:56,050][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:25:56,550][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:25:57,045][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:25:57,544][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:25:58,042][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:25:58,538][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:25:59,039][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:25:59,539][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:26:00,039][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:26:00,540][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:26:01,040][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:26:01,534][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:26:02,036][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:26:02,536][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:26:03,032][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:26:03,529][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:26:04,025][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:26:04,521][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:26:05,019][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:26:05,516][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:26:06,012][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:26:06,508][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9840 tokens. +[2026-03-26 09:26:07,393][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.01%, Current % of VRAM taken: 60.44%, Block Peak % of device VRAM: 61.92%, ΔTime: 00:00:32 +[2026-03-26 09:26:08,146][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:26:08,148][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:26:08,150][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:26:08,914][__main__][INFO] - Iteration 746 took 50s (28.93% Gen, 69.56% Train). Generation: 14s, Training: 35s. Estimated remaining time: 30h 49m 5s. Estimated total time: 42h 14m 25s. Time estimates for 10 more iterations: 8m 26s, 100 more iterations: 1h 24m 28s, 500 more iterations: 7h 2m 24s. +[2026-03-26 09:26:08,916][__main__][INFO] - Starting iteration 746. +[2026-03-26 09:26:09,316][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 74 and human policies 1. +[2026-03-26 09:26:09,317][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:26:28,804][__main__][INFO] - Number of regex retries in iteration 746: 0 +[2026-03-26 09:26:28,805][__main__][INFO] - agents played in iteration 746 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:26:29,579][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:26:29,603][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:26:29,626][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:26:29,652][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:26:29,653][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:26:29,653][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:26:30,520][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:26:30,977][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:26:31,482][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:26:31,982][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:26:32,484][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:26:32,991][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:26:33,497][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:26:33,993][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:26:34,491][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:26:34,992][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:26:35,489][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:26:35,989][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:26:36,487][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:26:36,986][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:26:37,483][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:26:37,980][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:26:38,482][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:26:38,985][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:26:39,489][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:26:39,991][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:26:40,492][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:26:40,991][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:26:41,487][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:26:41,984][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:26:42,481][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:26:42,980][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:26:43,479][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:26:43,979][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:26:44,480][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:26:44,979][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:26:45,479][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:26:45,978][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:26:46,484][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:26:46,983][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:26:47,483][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:26:47,982][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:26:48,481][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:26:48,979][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:26:49,478][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:26:49,978][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:26:50,481][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:26:50,977][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:26:51,477][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:26:51,976][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:26:52,479][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:26:52,973][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:26:53,470][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:26:53,968][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:26:54,466][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:26:54,963][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:26:55,461][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:26:55,959][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:26:56,458][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:26:56,957][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:26:57,454][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:26:57,953][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:26:58,454][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:26:58,948][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:26:59,447][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:26:59,942][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:27:00,435][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:27:00,934][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:27:01,427][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:27:01,921][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:27:02,420][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9832 tokens. +[2026-03-26 09:27:03,311][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.10%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.05%, ΔTime: 00:00:32 +[2026-03-26 09:27:04,044][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:27:04,046][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:27:04,048][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:27:04,835][__main__][INFO] - Iteration 747 took 55s (35.10% Gen, 63.48% Train). Generation: 19s, Training: 35s. Estimated remaining time: 34h 49m 40s. Estimated total time: 46h 15m 56s. Time estimates for 10 more iterations: 9m 15s, 100 more iterations: 1h 32m 31s, 500 more iterations: 7h 42m 39s. +[2026-03-26 09:27:04,837][__main__][INFO] - Starting iteration 747. +[2026-03-26 09:27:05,241][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 74 and human policies 1. +[2026-03-26 09:27:05,242][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:27:25,500][__main__][INFO] - Number of regex retries in iteration 747: 0 +[2026-03-26 09:27:25,501][__main__][INFO] - agents played in iteration 747 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:27:26,305][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:27:26,329][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:27:26,353][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:27:26,377][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:27:26,377][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:27:26,378][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:27:27,267][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:27:27,722][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:27:28,230][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:27:28,733][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:27:29,238][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:27:29,742][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:27:30,245][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:27:30,748][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:27:31,247][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:27:31,752][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:27:32,256][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:27:32,759][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:27:33,258][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:27:33,753][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:27:34,252][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:27:34,752][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:27:35,251][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:27:35,757][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:27:36,261][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:27:36,766][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:27:37,267][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:27:37,766][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:27:38,264][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:27:38,763][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:27:39,262][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:27:39,761][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:27:40,262][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:27:40,764][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:27:41,289][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:27:41,788][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:27:42,286][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:27:42,786][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:27:43,286][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:27:43,789][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:27:44,287][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:27:44,785][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:27:45,284][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:27:45,784][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:27:46,284][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:27:46,778][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:27:47,276][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:27:47,774][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:27:48,274][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:27:48,772][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:27:49,269][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:27:49,766][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:27:50,264][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:27:50,761][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:27:51,258][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:27:51,755][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:27:52,252][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:27:52,750][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:27:53,247][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:27:53,744][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:27:54,242][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:27:54,739][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:27:55,237][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:27:55,729][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:27:56,226][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:27:56,723][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:27:57,217][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:27:57,708][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:27:58,198][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:27:58,691][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:27:59,188][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9909 tokens. +[2026-03-26 09:28:00,085][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.03%, Current % of VRAM taken: 60.45%, Block Peak % of device VRAM: 62.13%, ΔTime: 00:00:32 +[2026-03-26 09:28:00,843][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:28:00,845][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:28:00,847][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:28:01,575][__main__][INFO] - Iteration 748 took 56s (35.96% Gen, 62.74% Train). Generation: 20s, Training: 35s. Estimated remaining time: 35h 29m 30s. Estimated total time: 46h 56m 43s. Time estimates for 10 more iterations: 9m 23s, 100 more iterations: 1h 33m 53s, 500 more iterations: 7h 49m 27s. +[2026-03-26 09:28:01,577][__main__][INFO] - Starting iteration 748. +[2026-03-26 09:28:01,974][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 74 and human policies 1. +[2026-03-26 09:28:01,975][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:28:19,305][__main__][INFO] - Number of regex retries in iteration 748: 0 +[2026-03-26 09:28:19,306][__main__][INFO] - agents played in iteration 748 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:28:20,070][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:28:20,094][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:28:20,117][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:28:20,141][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:28:20,141][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:28:20,142][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:28:21,008][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:28:21,471][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:28:21,979][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:28:22,484][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:28:22,985][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:28:23,488][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:28:23,990][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:28:24,493][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:28:24,996][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:28:25,500][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:28:26,004][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:28:26,504][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:28:27,007][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:28:27,510][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:28:28,017][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:28:28,521][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:28:29,022][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:28:29,520][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:28:30,020][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:28:30,521][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:28:31,019][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:28:31,518][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:28:32,015][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:28:32,512][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:28:33,010][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:28:33,508][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:28:34,011][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:28:34,509][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:28:35,012][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:28:35,511][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:28:36,018][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:28:36,519][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:28:37,020][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:28:37,519][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:28:38,017][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:28:38,515][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:28:39,016][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:28:39,517][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:28:40,012][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:28:40,511][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:28:41,007][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:28:41,506][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:28:42,003][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:28:42,501][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:28:42,998][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:28:43,490][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:28:43,986][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:28:44,482][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:28:44,974][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:28:45,465][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:28:45,962][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:28:46,455][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:28:46,947][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:28:47,443][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:28:47,942][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:28:48,440][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:28:48,938][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:28:49,436][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:28:49,933][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:28:50,432][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:28:50,929][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:28:51,426][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:28:51,923][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:28:52,420][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:28:52,921][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9925 tokens. +[2026-03-26 09:28:53,806][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.19%, Current % of VRAM taken: 60.62%, Block Peak % of device VRAM: 62.21%, ΔTime: 00:00:32 +[2026-03-26 09:28:54,559][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:28:54,562][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:28:54,563][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:28:55,415][__main__][INFO] - Iteration 749 took 53s (32.43% Gen, 65.97% Train). Generation: 17s, Training: 35s. Estimated remaining time: 33h 3m 57s. Estimated total time: 44h 32m 4s. Time estimates for 10 more iterations: 8m 54s, 100 more iterations: 1h 29m 4s, 500 more iterations: 7h 25m 20s. +[2026-03-26 09:28:55,417][__main__][INFO] - Starting iteration 749. +[2026-03-26 09:28:55,817][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 74 and human policies 1. +[2026-03-26 09:28:55,818][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:29:03,111][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 09:29:13,481][__main__][INFO] - Number of regex retries in iteration 749: 1 +[2026-03-26 09:29:13,482][__main__][INFO] - agents played in iteration 749 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:29:14,259][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:29:14,282][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:29:14,307][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:29:14,330][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.54%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:29:14,331][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:29:14,331][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:29:15,226][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:29:15,680][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:29:16,188][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:29:16,690][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:29:17,193][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:29:17,694][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:29:18,192][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:29:18,690][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:29:19,190][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:29:19,687][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:29:20,186][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:29:20,683][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:29:21,181][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:29:21,679][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:29:22,177][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:29:22,670][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:29:23,167][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:29:23,669][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:29:24,167][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:29:24,665][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:29:25,163][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:29:25,662][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:29:26,164][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:29:26,667][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:29:27,167][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:29:27,668][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:29:28,168][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:29:28,666][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:29:29,162][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:29:29,660][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:29:30,158][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:29:30,651][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:29:31,148][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:29:31,643][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:29:32,145][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:29:32,642][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:29:33,141][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:29:33,640][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:29:34,138][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:29:34,632][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:29:35,133][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:29:35,630][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:29:36,127][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:29:36,625][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:29:37,123][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:29:37,621][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:29:38,120][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:29:38,621][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:29:39,118][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:29:39,613][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:29:40,114][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:29:40,611][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:29:41,119][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:29:41,616][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:29:42,113][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:29:42,606][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:29:43,098][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:29:43,597][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:29:44,095][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:29:44,593][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:29:45,091][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:29:45,583][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:29:46,080][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:29:46,578][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:29:47,079][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9893 tokens. +[2026-03-26 09:29:47,976][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.24%, Current % of VRAM taken: 60.66%, Block Peak % of device VRAM: 62.18%, ΔTime: 00:00:32 +[2026-03-26 09:29:48,718][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:29:48,720][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:29:48,722][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:29:49,457][__main__][INFO] - Iteration 750 took 53s (32.93% Gen, 65.70% Train). Generation: 17s, Training: 35s. Estimated remaining time: 33h 13m 1s. Estimated total time: 44h 42m 2s. Time estimates for 10 more iterations: 8m 56s, 100 more iterations: 1h 29m 24s, 500 more iterations: 7h 27m 0s. +[2026-03-26 09:29:49,460][__main__][INFO] - Starting iteration 750. +[2026-03-26 09:29:49,889][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 74 and human policies 1. +[2026-03-26 09:29:49,890][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:30:04,926][__main__][INFO] - Number of regex retries in iteration 750: 0 +[2026-03-26 09:30:04,927][__main__][INFO] - agents played in iteration 750 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:30:05,685][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:30:05,708][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:30:05,732][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:30:05,756][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.56%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:30:05,757][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:30:05,757][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:30:06,617][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:30:07,072][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:30:07,573][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:30:08,087][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:30:08,585][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:30:09,084][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:30:09,582][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:30:10,081][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:30:10,573][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:30:11,069][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:30:11,560][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:30:12,054][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:30:12,558][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:30:13,054][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:30:13,554][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:30:14,057][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:30:14,562][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:30:15,061][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:30:15,563][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:30:16,065][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:30:16,561][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:30:17,059][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:30:17,557][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:30:18,058][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:30:18,559][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:30:19,064][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:30:19,569][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:30:20,074][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:30:20,579][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:30:21,084][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:30:21,587][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:30:22,091][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:30:22,592][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:30:23,091][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:30:23,589][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:30:24,088][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:30:24,581][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:30:25,079][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:30:25,573][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:30:26,072][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:30:26,571][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:30:27,071][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:30:27,567][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:30:28,065][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:30:28,565][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:30:29,065][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:30:29,564][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:30:30,063][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:30:30,562][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:30:31,068][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:30:31,567][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:30:32,066][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:30:32,567][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:30:33,067][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:30:33,566][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:30:34,065][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:30:34,566][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:30:35,061][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:30:35,557][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:30:36,062][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:30:36,562][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:30:37,061][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:30:37,560][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:30:38,059][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:30:38,557][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9832 tokens. +[2026-03-26 09:30:39,458][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.95%, Current % of VRAM taken: 60.38%, Block Peak % of device VRAM: 62.15%, ΔTime: 00:00:32 +[2026-03-26 09:30:40,215][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:30:40,217][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:30:40,219][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:30:41,631][__main__][INFO] - Iteration 751 took 51s (29.06% Gen, 68.21% Train). Generation: 15s, Training: 35s. Estimated remaining time: 31h 37m 16s. Estimated total time: 43h 7m 9s. Time estimates for 10 more iterations: 8m 37s, 100 more iterations: 1h 26m 14s, 500 more iterations: 7h 11m 11s. +[2026-03-26 09:30:41,633][__main__][INFO] - Starting iteration 751. +[2026-03-26 09:30:42,035][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 75 and human policies 1. +[2026-03-26 09:30:42,035][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:30:56,948][__main__][INFO] - Number of regex retries in iteration 751: 0 +[2026-03-26 09:30:56,949][__main__][INFO] - agents played in iteration 751 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:30:57,710][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:30:57,737][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:30:57,764][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:30:57,790][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.43%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:30:57,791][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:30:57,791][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:30:58,665][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:30:59,122][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:30:59,624][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:31:00,132][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:31:00,631][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:31:01,129][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:31:01,628][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:31:02,128][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:31:02,626][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:31:03,127][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:31:03,625][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:31:04,129][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:31:04,629][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:31:05,135][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:31:05,638][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:31:06,138][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:31:06,637][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:31:07,137][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:31:07,636][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:31:08,151][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:31:08,658][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:31:09,161][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:31:09,663][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:31:10,166][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:31:10,672][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:31:11,172][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:31:11,675][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:31:12,177][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:31:12,682][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:31:13,187][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:31:13,696][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:31:14,197][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:31:14,696][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:31:15,195][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:31:15,693][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:31:16,191][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:31:16,689][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:31:17,187][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:31:17,686][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:31:18,184][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:31:18,683][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:31:19,182][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:31:19,681][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:31:20,186][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:31:20,685][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:31:21,182][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:31:21,680][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:31:22,179][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:31:22,677][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:31:23,178][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:31:23,670][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:31:24,167][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:31:24,663][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:31:25,163][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:31:25,663][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:31:26,161][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:31:26,659][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:31:27,156][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:31:27,652][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:31:28,148][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:31:28,644][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:31:29,140][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:31:29,637][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:31:30,134][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:31:30,629][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9890 tokens. +[2026-03-26 09:31:31,524][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.97%, Current % of VRAM taken: 60.40%, Block Peak % of device VRAM: 62.22%, ΔTime: 00:00:32 +[2026-03-26 09:31:32,278][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:31:32,280][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:31:32,282][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:31:33,068][__main__][INFO] - Iteration 752 took 51s (29.22% Gen, 69.24% Train). Generation: 14s, Training: 35s. Estimated remaining time: 31h 0m 56s. Estimated total time: 42h 31m 40s. Time estimates for 10 more iterations: 8m 30s, 100 more iterations: 1h 25m 3s, 500 more iterations: 7h 5m 16s. +[2026-03-26 09:31:33,070][__main__][INFO] - Starting iteration 752. +[2026-03-26 09:31:33,472][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 75 and human policies 1. +[2026-03-26 09:31:33,473][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:31:52,599][__main__][INFO] - Number of regex retries in iteration 752: 0 +[2026-03-26 09:31:52,600][__main__][INFO] - agents played in iteration 752 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:31:53,362][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:31:53,386][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:31:53,410][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:31:53,434][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:31:53,435][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:31:53,435][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:31:54,302][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:31:54,755][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:31:55,256][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:31:55,759][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:31:56,259][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:31:56,757][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:31:57,257][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:31:57,755][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:31:58,256][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:31:58,758][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:31:59,267][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:31:59,770][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:32:00,274][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:32:00,785][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:32:01,289][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:32:01,789][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:32:02,292][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:32:02,796][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:32:03,300][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:32:03,804][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:32:04,307][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:32:04,804][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:32:05,302][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:32:05,798][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:32:06,297][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:32:06,796][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:32:07,294][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:32:07,789][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:32:08,288][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:32:08,794][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:32:09,289][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:32:09,786][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:32:10,283][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:32:10,782][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:32:11,276][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:32:11,774][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:32:12,273][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:32:12,772][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:32:13,271][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:32:13,769][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:32:14,268][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:32:14,765][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:32:15,264][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:32:15,765][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:32:16,264][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:32:16,762][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:32:17,265][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:32:17,770][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:32:18,269][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:32:18,768][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:32:19,267][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:32:19,767][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:32:20,268][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:32:20,766][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:32:21,262][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:32:21,756][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:32:22,249][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:32:22,747][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:32:23,246][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:32:23,741][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:32:24,237][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:32:24,737][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:32:25,245][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:32:25,745][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:32:26,240][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9779 tokens. +[2026-03-26 09:32:27,153][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.93%, Current % of VRAM taken: 60.36%, Block Peak % of device VRAM: 61.81%, ΔTime: 00:00:32 +[2026-03-26 09:32:27,909][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:32:27,912][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:32:27,913][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:32:28,624][__main__][INFO] - Iteration 753 took 55s (34.68% Gen, 64.03% Train). Generation: 19s, Training: 35s. Estimated remaining time: 34h 25m 57s. Estimated total time: 45h 57m 37s. Time estimates for 10 more iterations: 9m 11s, 100 more iterations: 1h 31m 55s, 500 more iterations: 7h 39m 36s. +[2026-03-26 09:32:28,626][__main__][INFO] - Starting iteration 753. +[2026-03-26 09:32:29,034][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 75 and human policies 1. +[2026-03-26 09:32:29,034][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:32:42,745][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 09:32:43,619][__main__][INFO] - Number of regex retries in iteration 753: 1 +[2026-03-26 09:32:43,620][__main__][INFO] - agents played in iteration 753 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:32:44,384][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:32:44,408][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:32:44,431][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:32:44,455][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:32:44,455][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:32:44,456][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:32:45,325][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:32:45,779][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:32:46,282][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:32:46,781][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:32:47,280][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:32:47,778][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:32:48,276][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:32:48,773][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:32:49,275][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:32:49,784][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:32:50,287][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:32:50,792][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:32:51,296][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:32:51,802][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:32:52,306][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:32:52,804][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:32:53,305][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:32:53,806][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:32:54,305][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:32:54,805][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:32:55,309][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:32:55,807][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:32:56,306][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:32:56,830][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:32:57,331][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:32:57,829][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:32:58,329][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:32:58,826][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:32:59,325][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:32:59,820][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:33:00,318][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:33:00,815][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:33:01,311][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:33:01,809][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:33:02,306][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:33:02,798][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:33:03,296][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:33:03,793][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:33:04,290][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:33:04,789][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:33:05,286][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:33:05,790][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:33:06,290][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:33:06,789][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:33:07,288][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:33:07,787][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:33:08,286][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:33:08,785][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:33:09,284][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:33:09,782][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:33:10,279][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:33:10,776][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:33:11,275][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:33:11,773][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:33:12,276][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:33:12,774][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:33:13,269][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:33:13,766][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:33:14,263][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:33:14,761][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:33:15,260][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:33:15,758][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:33:16,257][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:33:16,754][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:33:17,255][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9910 tokens. +[2026-03-26 09:33:18,158][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.19%, Current % of VRAM taken: 60.61%, Block Peak % of device VRAM: 61.98%, ΔTime: 00:00:32 +[2026-03-26 09:33:18,919][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:33:18,922][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:33:18,923][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:37:39,064][__main__][INFO] - Iteration 754 took 5m 10s (4.70% Gen, 11.39% Train). Generation: 14s, Training: 35s. Estimated remaining time: 246h 44m 44s. Estimated total time: 258h 21m 34s. Time estimates for 10 more iterations: 51m 40s, 100 more iterations: 8h 36m 43s, 500 more iterations: 43h 3m 35s. +[2026-03-26 09:37:39,066][__main__][INFO] - Starting iteration 754. +[2026-03-26 09:37:39,466][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 75 and human policies 1. +[2026-03-26 09:37:39,467][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:38:47,545][__main__][INFO] - Number of regex retries in iteration 754: 0 +[2026-03-26 09:38:47,546][__main__][INFO] - agents played in iteration 754 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:38:48,320][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:38:48,343][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:38:48,367][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:38:48,390][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:38:48,391][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:38:48,391][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:38:49,249][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:38:49,701][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:38:50,206][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:38:50,698][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:38:51,194][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:38:51,691][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:38:52,187][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:38:52,684][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:38:53,183][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:38:53,679][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:38:54,178][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:38:54,679][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:38:55,183][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:38:55,687][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:38:56,186][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:38:56,686][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:38:57,184][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:38:57,681][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:38:58,177][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:38:58,667][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:38:59,164][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:38:59,660][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:39:00,161][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:39:00,661][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:39:01,158][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:39:01,662][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:39:02,165][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:39:02,678][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:39:03,180][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:39:03,682][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:39:04,188][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:39:04,687][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:39:05,188][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:39:05,690][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:39:06,192][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:39:06,693][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:39:07,190][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:39:07,687][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:39:08,184][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:39:08,687][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:39:09,188][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:39:09,689][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:39:10,188][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:39:10,686][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:39:11,184][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:39:11,684][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:39:12,182][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:39:12,681][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:39:13,180][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:39:13,680][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:39:14,179][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:39:14,678][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:39:15,177][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:39:15,676][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:39:16,176][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:39:16,675][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:39:17,175][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:39:17,674][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:39:18,168][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:39:18,666][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:39:19,164][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:39:19,662][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:39:20,165][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:39:20,664][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:39:21,165][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9903 tokens. +[2026-03-26 09:39:22,070][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.06%, Current % of VRAM taken: 60.49%, Block Peak % of device VRAM: 62.10%, ΔTime: 00:00:32 +[2026-03-26 09:39:22,846][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:39:22,849][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:39:22,850][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:39:23,606][__main__][INFO] - Iteration 755 took 1m 44s (65.37% Gen, 33.90% Train). Generation: 1m 8s, Training: 35s. Estimated remaining time: 75h 8m 26s. Estimated total time: 86h 47m 1s. Time estimates for 10 more iterations: 17m 21s, 100 more iterations: 2h 53m 34s, 500 more iterations: 14h 27m 50s. +[2026-03-26 09:39:23,608][__main__][INFO] - Starting iteration 755. +[2026-03-26 09:39:24,010][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 75 and human policies 1. +[2026-03-26 09:39:24,011][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:39:53,016][__main__][INFO] - Number of regex retries in iteration 755: 0 +[2026-03-26 09:39:53,016][__main__][INFO] - agents played in iteration 755 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:39:53,784][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:39:53,808][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:39:53,832][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:39:53,856][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.46%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:39:53,856][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:39:53,857][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:39:54,737][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:39:55,196][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:39:55,694][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:39:56,192][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:39:56,691][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:39:57,188][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:39:57,682][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:39:58,173][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:39:58,675][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:39:59,166][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:39:59,658][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:40:00,151][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:40:00,649][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:40:01,150][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:40:01,651][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:40:02,147][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:40:02,649][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:40:03,154][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:40:03,659][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:40:04,163][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:40:04,666][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:40:05,166][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:40:05,665][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:40:06,160][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:40:06,659][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:40:07,152][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:40:07,642][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:40:08,134][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:40:08,624][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:40:09,116][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:40:09,613][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:40:10,104][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:40:10,595][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:40:11,100][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:40:11,598][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:40:12,093][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:40:12,588][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:40:13,086][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:40:13,585][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:40:14,080][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:40:14,579][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:40:15,080][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:40:15,580][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:40:16,078][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:40:16,578][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:40:17,076][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:40:17,574][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:40:18,072][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:40:18,565][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:40:19,061][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:40:19,557][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:40:20,055][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:40:20,552][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:40:21,049][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:40:21,545][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:40:22,042][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:40:22,540][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:40:23,038][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:40:23,535][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:40:24,033][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:40:24,532][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:40:25,030][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:40:25,526][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:40:26,023][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:40:26,521][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9727 tokens. +[2026-03-26 09:40:27,408][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 61.95%, ΔTime: 00:00:32 +[2026-03-26 09:40:28,163][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:40:28,167][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:40:28,169][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:40:29,029][__main__][INFO] - Iteration 756 took 1m 5s (44.61% Gen, 54.06% Train). Generation: 29s, Training: 35s. Estimated remaining time: 42h 31m 18s. Estimated total time: 54h 10m 58s. Time estimates for 10 more iterations: 10m 50s, 100 more iterations: 1h 48m 21s, 500 more iterations: 9h 1m 49s. +[2026-03-26 09:40:29,031][__main__][INFO] - Starting iteration 756. +[2026-03-26 09:40:29,441][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 75 and human policies 1. +[2026-03-26 09:40:29,442][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:40:54,202][__main__][INFO] - Number of regex retries in iteration 756: 0 +[2026-03-26 09:40:54,203][__main__][INFO] - agents played in iteration 756 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:40:54,955][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:40:54,978][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:40:55,002][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:40:55,025][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:40:55,026][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:40:55,026][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:40:55,881][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:40:56,335][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:40:56,833][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:40:57,332][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:40:57,834][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:40:58,336][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:40:58,836][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:40:59,333][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:40:59,833][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:41:00,335][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:41:00,832][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:41:01,332][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:41:01,834][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:41:02,337][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:41:02,839][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:41:03,340][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:41:03,840][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:41:04,339][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:41:04,837][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:41:05,335][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:41:05,832][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:41:06,330][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:41:06,827][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:41:07,324][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:41:07,822][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:41:08,320][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:41:08,822][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:41:09,321][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:41:09,823][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:41:10,322][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:41:10,821][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:41:11,321][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:41:11,820][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:41:12,314][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:41:12,811][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:41:13,309][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:41:13,809][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:41:14,308][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:41:14,806][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:41:15,305][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:41:15,805][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:41:16,304][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:41:16,803][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:41:17,300][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:41:17,801][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:41:18,297][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:41:18,798][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:41:19,294][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:41:19,790][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:41:20,289][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:41:20,787][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:41:21,286][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:41:21,784][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:41:22,281][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:41:22,778][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:41:23,277][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:41:23,774][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:41:24,269][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:41:24,763][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:41:25,259][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:41:25,755][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:41:26,251][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:41:26,745][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:41:27,240][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:41:27,737][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9918 tokens. +[2026-03-26 09:41:28,622][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.99%, Current % of VRAM taken: 60.41%, Block Peak % of device VRAM: 62.08%, ΔTime: 00:00:32 +[2026-03-26 09:41:29,376][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:41:29,378][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:41:29,380][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:41:30,046][__main__][INFO] - Iteration 757 took 1m 0s (40.85% Gen, 58.03% Train). Generation: 24s, Training: 35s. Estimated remaining time: 38h 49m 57s. Estimated total time: 50h 30m 39s. Time estimates for 10 more iterations: 10m 6s, 100 more iterations: 1h 41m 1s, 500 more iterations: 8h 25m 6s. +[2026-03-26 09:41:30,048][__main__][INFO] - Starting iteration 757. +[2026-03-26 09:41:30,449][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 75 and human policies 1. +[2026-03-26 09:41:30,450][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:42:11,944][__main__][INFO] - Number of regex retries in iteration 757: 0 +[2026-03-26 09:42:11,945][__main__][INFO] - agents played in iteration 757 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:42:12,699][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:42:12,723][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:42:12,746][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:42:12,770][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.42%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:42:12,770][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:42:12,771][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:42:13,640][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:42:14,096][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:42:14,597][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:42:15,100][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:42:15,600][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:42:16,101][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:42:16,601][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:42:17,102][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:42:17,601][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:42:18,099][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:42:18,597][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:42:19,095][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:42:19,587][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:42:20,082][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:42:20,580][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:42:21,075][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:42:21,568][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:42:22,067][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:42:22,566][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:42:23,066][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:42:23,564][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:42:24,067][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:42:24,563][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:42:25,062][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:42:25,561][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:42:26,058][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:42:26,555][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:42:27,052][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:42:27,544][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:42:28,036][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:42:28,532][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:42:29,028][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:42:29,526][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:42:30,027][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:42:30,527][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:42:31,026][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:42:31,527][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:42:32,026][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:42:32,526][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:42:33,026][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:42:33,525][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:42:34,023][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:42:34,516][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:42:35,010][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:42:35,501][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:42:35,993][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:42:36,490][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:42:36,988][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:42:37,486][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:42:37,979][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:42:38,473][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:42:38,965][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:42:39,462][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:42:39,957][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:42:40,451][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:42:40,944][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:42:41,441][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:42:41,936][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:42:42,433][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:42:42,930][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:42:43,428][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:42:43,929][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:42:44,432][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:42:44,937][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:42:45,433][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9804 tokens. +[2026-03-26 09:42:46,371][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.02%, Current % of VRAM taken: 60.45%, Block Peak % of device VRAM: 62.30%, ΔTime: 00:00:32 +[2026-03-26 09:42:47,135][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:42:47,138][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:42:47,140][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:42:47,814][__main__][INFO] - Iteration 758 took 1m 17s (53.63% Gen, 45.49% Train). Generation: 41s, Training: 35s. Estimated remaining time: 52h 46m 17s. Estimated total time: 64h 28m 16s. Time estimates for 10 more iterations: 12m 53s, 100 more iterations: 2h 8m 56s, 500 more iterations: 10h 44m 42s. +[2026-03-26 09:42:47,817][__main__][INFO] - Starting iteration 758. +[2026-03-26 09:42:48,217][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 75 and human policies 1. +[2026-03-26 09:42:48,218][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:43:03,245][__main__][INFO] - Number of regex retries in iteration 758: 0 +[2026-03-26 09:43:03,246][__main__][INFO] - agents played in iteration 758 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:43:04,009][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:43:04,033][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:43:04,056][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:43:04,080][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:43:04,080][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:43:04,081][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:43:04,953][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:43:05,406][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:43:05,907][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:43:06,405][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:43:06,903][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:43:07,405][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:43:07,909][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:43:08,408][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:43:08,909][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:43:09,407][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:43:09,905][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:43:10,404][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:43:10,902][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:43:11,400][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:43:11,898][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:43:12,396][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:43:12,892][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:43:13,395][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:43:13,895][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:43:14,398][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:43:14,898][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:43:15,398][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:43:15,902][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:43:16,409][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:43:16,909][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:43:17,410][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:43:17,908][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:43:18,406][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:43:18,905][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:43:19,422][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:43:19,924][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:43:20,425][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:43:20,923][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:43:21,421][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:43:21,922][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:43:22,426][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:43:22,926][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:43:23,426][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:43:23,925][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:43:24,424][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:43:24,923][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:43:25,422][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:43:25,919][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:43:26,418][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:43:26,916][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:43:27,414][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:43:27,912][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:43:28,409][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:43:28,907][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:43:29,404][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:43:29,897][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:43:30,393][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:43:30,890][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:43:31,387][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:43:31,884][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:43:32,384][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:43:32,888][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:43:33,388][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:43:33,887][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:43:34,387][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:43:34,887][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:43:35,390][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:43:35,893][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:43:36,398][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:43:36,902][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9904 tokens. +[2026-03-26 09:43:37,838][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 61.99%, ΔTime: 00:00:32 +[2026-03-26 09:43:38,597][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:43:38,599][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:43:38,600][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:43:39,480][__main__][INFO] - Iteration 759 took 51s (29.31% Gen, 68.97% Train). Generation: 15s, Training: 35s. Estimated remaining time: 31h 0m 22s. Estimated total time: 42h 43m 12s. Time estimates for 10 more iterations: 8m 32s, 100 more iterations: 1h 25m 26s, 500 more iterations: 7h 7m 12s. +[2026-03-26 09:43:39,482][__main__][INFO] - Starting iteration 759. +[2026-03-26 09:43:39,880][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 75 and human policies 1. +[2026-03-26 09:43:39,881][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:44:05,760][__main__][INFO] - Number of regex retries in iteration 759: 0 +[2026-03-26 09:44:05,761][__main__][INFO] - agents played in iteration 759 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:44:06,520][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:44:06,544][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:44:06,567][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:44:06,590][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:44:06,591][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:44:06,592][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:44:07,465][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:44:07,921][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:44:08,425][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:44:08,940][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:44:09,438][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:44:09,941][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:44:10,437][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:44:10,932][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:44:11,436][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:44:11,937][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:44:12,437][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:44:12,938][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:44:13,440][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:44:13,936][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:44:14,435][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:44:14,934][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:44:15,435][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:44:15,942][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:44:16,441][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:44:16,941][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:44:17,442][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:44:17,942][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:44:18,441][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:44:18,937][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:44:19,435][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:44:19,937][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:44:20,451][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:44:20,952][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:44:21,455][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:44:21,957][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:44:22,460][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:44:22,961][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:44:23,462][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:44:23,961][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:44:24,463][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:44:24,963][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:44:25,463][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:44:25,968][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:44:26,472][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:44:26,972][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:44:27,472][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:44:27,971][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:44:28,466][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:44:28,965][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:44:29,464][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:44:29,961][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:44:30,459][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:44:30,957][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:44:31,453][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:44:31,949][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:44:32,442][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:44:32,936][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:44:33,438][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:44:33,938][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:44:34,438][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:44:34,939][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:44:35,443][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:44:35,949][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:44:36,445][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:44:36,940][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:44:37,433][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:44:37,926][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:44:38,421][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:44:38,920][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:44:39,418][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9767 tokens. +[2026-03-26 09:44:40,388][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.96%, Current % of VRAM taken: 60.38%, Block Peak % of device VRAM: 62.18%, ΔTime: 00:00:32 +[2026-03-26 09:44:41,151][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:44:41,153][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:44:41,155][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:44:41,892][__main__][INFO] - Iteration 760 took 1m 2s (41.73% Gen, 57.08% Train). Generation: 25s, Training: 35s. Estimated remaining time: 39h 56m 46s. Estimated total time: 51h 40m 39s. Time estimates for 10 more iterations: 10m 20s, 100 more iterations: 1h 43m 21s, 500 more iterations: 8h 36m 46s. +[2026-03-26 09:44:41,895][__main__][INFO] - Starting iteration 760. +[2026-03-26 09:44:42,294][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 75 and human policies 1. +[2026-03-26 09:44:42,295][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:44:56,127][__main__][INFO] - Number of regex retries in iteration 760: 0 +[2026-03-26 09:44:56,128][__main__][INFO] - agents played in iteration 760 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:44:56,889][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:44:56,913][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:44:56,936][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:44:56,960][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:44:56,961][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:44:56,962][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:44:57,831][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:44:58,287][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:44:58,792][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:44:59,293][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:44:59,793][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:45:00,292][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:45:00,791][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:45:01,290][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:45:01,791][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:45:02,291][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:45:02,789][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:45:03,286][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:45:03,784][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:45:04,282][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:45:04,782][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:45:05,281][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:45:05,782][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:45:06,290][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:45:06,783][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:45:07,280][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:45:07,777][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:45:08,273][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:45:08,769][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:45:09,263][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:45:09,759][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:45:10,256][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:45:10,751][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:45:11,243][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:45:11,736][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:45:12,227][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:45:12,722][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:45:13,220][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:45:13,716][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:45:14,212][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:45:14,709][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:45:15,205][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:45:15,702][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:45:16,199][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:45:16,696][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:45:17,192][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:45:17,689][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:45:18,184][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:45:18,678][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:45:19,174][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:45:19,666][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:45:20,165][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:45:20,659][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:45:21,158][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:45:21,649][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:45:22,144][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:45:22,640][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:45:23,136][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:45:23,631][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:45:24,124][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:45:24,616][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:45:25,110][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:45:25,603][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:45:26,099][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:45:26,596][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:45:27,098][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:45:27,589][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:45:28,086][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:45:28,583][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:45:29,077][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:45:29,567][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9694 tokens. +[2026-03-26 09:45:30,494][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.90%, Current % of VRAM taken: 60.33%, Block Peak % of device VRAM: 61.86%, ΔTime: 00:00:32 +[2026-03-26 09:45:31,253][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:45:31,256][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:45:31,257][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:45:32,639][__main__][INFO] - Iteration 761 took 50s (27.48% Gen, 69.78% Train). Generation: 13s, Training: 35s. Estimated remaining time: 30h 12m 31s. Estimated total time: 41h 57m 15s. Time estimates for 10 more iterations: 8m 23s, 100 more iterations: 1h 23m 54s, 500 more iterations: 6h 59m 32s. +[2026-03-26 09:45:32,641][__main__][INFO] - Starting iteration 761. +[2026-03-26 09:45:33,040][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 76 and human policies 1. +[2026-03-26 09:45:33,040][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:45:52,623][__main__][INFO] - Number of regex retries in iteration 761: 0 +[2026-03-26 09:45:52,625][__main__][INFO] - agents played in iteration 761 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:45:53,419][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:45:53,443][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:45:53,468][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:45:53,492][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:45:53,493][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:45:53,493][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:45:54,395][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:45:54,853][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:45:55,356][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:45:55,854][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:45:56,356][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:45:56,854][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:45:57,352][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:45:57,861][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:45:58,361][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:45:58,860][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:45:59,358][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:45:59,856][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:46:00,353][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:46:00,848][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:46:01,347][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:46:01,845][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:46:02,349][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:46:02,848][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:46:03,348][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:46:03,849][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:46:04,356][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:46:04,858][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:46:05,352][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:46:05,850][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:46:06,350][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:46:06,849][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:46:07,347][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:46:07,841][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:46:08,339][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:46:08,840][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:46:09,339][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:46:09,839][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:46:10,339][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:46:10,832][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:46:11,327][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:46:11,825][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:46:12,336][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:46:12,843][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:46:13,343][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:46:13,844][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:46:14,345][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:46:14,844][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:46:15,345][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:46:15,845][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:46:16,344][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:46:16,842][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:46:17,341][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:46:17,840][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:46:18,340][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:46:18,837][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:46:19,337][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:46:19,834][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:46:20,335][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:46:20,837][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:46:21,345][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:46:21,844][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:46:22,344][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:46:22,844][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:46:23,343][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:46:23,840][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:46:24,339][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:46:24,839][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:46:25,337][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:46:25,840][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:46:26,345][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9875 tokens. +[2026-03-26 09:46:27,306][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.52%, Block Peak % of device VRAM: 62.14%, ΔTime: 00:00:32 +[2026-03-26 09:46:28,078][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:46:28,080][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:46:28,082][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:46:28,878][__main__][INFO] - Iteration 762 took 55s (35.07% Gen, 63.50% Train). Generation: 19s, Training: 35s. Estimated remaining time: 34h 46m 15s. Estimated total time: 46h 31m 55s. Time estimates for 10 more iterations: 9m 18s, 100 more iterations: 1h 33m 3s, 500 more iterations: 7h 45m 19s. +[2026-03-26 09:46:28,880][__main__][INFO] - Starting iteration 762. +[2026-03-26 09:46:29,292][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 76 and human policies 1. +[2026-03-26 09:46:29,292][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:46:43,673][__main__][INFO] - Number of regex retries in iteration 762: 0 +[2026-03-26 09:46:43,674][__main__][INFO] - agents played in iteration 762 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:46:44,452][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:46:44,478][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:46:44,503][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:46:44,528][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:46:44,528][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:46:44,529][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:46:45,427][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:46:45,881][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:46:46,386][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:46:46,886][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:46:47,388][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:46:47,887][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:46:48,388][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:46:48,888][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:46:49,388][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:46:49,887][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:46:50,386][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:46:50,884][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:46:51,376][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:46:51,883][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:46:52,377][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:46:52,874][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:46:53,368][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:46:53,867][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:46:54,366][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:46:54,860][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:46:55,358][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:46:55,858][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:46:56,356][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:46:56,849][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:46:57,351][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:46:57,852][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:46:58,351][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:46:58,850][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:46:59,351][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:46:59,847][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:47:00,342][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:47:00,844][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:47:01,342][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:47:01,840][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:47:02,338][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:47:02,836][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:47:03,333][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:47:03,830][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:47:04,327][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:47:04,825][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:47:05,322][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:47:05,823][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:47:06,324][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:47:06,825][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:47:07,325][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:47:07,825][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:47:08,326][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:47:08,827][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:47:09,327][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:47:09,825][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:47:10,327][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:47:10,828][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:47:11,322][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:47:11,826][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:47:12,323][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:47:12,822][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:47:13,319][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:47:13,819][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:47:14,320][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:47:14,825][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:47:15,325][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:47:15,831][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:47:16,333][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:47:16,836][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:47:17,341][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9830 tokens. +[2026-03-26 09:47:18,284][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.12%, Current % of VRAM taken: 60.55%, Block Peak % of device VRAM: 62.17%, ΔTime: 00:00:32 +[2026-03-26 09:47:19,075][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:47:19,078][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:47:19,079][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:47:19,824][__main__][INFO] - Iteration 763 took 50s (28.46% Gen, 70.06% Train). Generation: 14s, Training: 35s. Estimated remaining time: 30h 20m 9s. Estimated total time: 42h 6m 41s. Time estimates for 10 more iterations: 8m 25s, 100 more iterations: 1h 24m 13s, 500 more iterations: 7h 1m 6s. +[2026-03-26 09:47:19,826][__main__][INFO] - Starting iteration 763. +[2026-03-26 09:47:20,253][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 76 and human policies 1. +[2026-03-26 09:47:20,254][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:47:35,406][__main__][INFO] - Number of regex retries in iteration 763: 0 +[2026-03-26 09:47:35,407][__main__][INFO] - agents played in iteration 763 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:47:36,182][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:47:36,206][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:47:36,230][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:47:36,255][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:47:36,255][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:47:36,256][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:47:37,177][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:47:37,636][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:47:38,147][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:47:38,650][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:47:39,158][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:47:39,659][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:47:40,164][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:47:40,668][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:47:41,182][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:47:41,690][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:47:42,201][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:47:42,708][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:47:43,216][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:47:43,721][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:47:44,231][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:47:44,736][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:47:45,246][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:47:45,745][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:47:46,251][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:47:46,748][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:47:47,246][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:47:47,744][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:47:48,246][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:47:48,753][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:47:49,252][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:47:49,755][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:47:50,257][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:47:50,759][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:47:51,262][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:47:51,767][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:47:52,265][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:47:52,763][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:47:53,272][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:47:53,777][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:47:54,294][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:47:54,798][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:47:55,304][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:47:55,805][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:47:56,312][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:47:56,816][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:47:57,322][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:47:57,822][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:47:58,328][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:47:58,829][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:47:59,330][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:47:59,832][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:48:00,330][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:48:00,830][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:48:01,331][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:48:01,829][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:48:02,327][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:48:02,823][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:48:03,321][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:48:03,816][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:48:04,324][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:48:04,822][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:48:05,322][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:48:05,820][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:48:06,316][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:48:06,823][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:48:07,323][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:48:07,828][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:48:08,330][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:48:08,832][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:48:09,343][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9864 tokens. +[2026-03-26 09:48:10,287][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.98%, Current % of VRAM taken: 60.40%, Block Peak % of device VRAM: 62.09%, ΔTime: 00:00:33 +[2026-03-26 09:48:11,053][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:48:11,056][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:48:11,058][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:48:11,813][__main__][INFO] - Iteration 764 took 51s (29.39% Gen, 69.14% Train). Generation: 15s, Training: 35s. Estimated remaining time: 31h 10m 39s. Estimated total time: 42h 58m 2s. Time estimates for 10 more iterations: 8m 35s, 100 more iterations: 1h 25m 56s, 500 more iterations: 7h 9m 40s. +[2026-03-26 09:48:11,816][__main__][INFO] - Starting iteration 764. +[2026-03-26 09:48:12,229][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 76 and human policies 1. +[2026-03-26 09:48:12,229][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:48:36,283][__main__][INFO] - Number of regex retries in iteration 764: 0 +[2026-03-26 09:48:36,284][__main__][INFO] - agents played in iteration 764 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:48:37,109][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:48:37,138][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:48:37,165][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:48:37,190][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:48:37,191][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:48:37,192][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:48:38,122][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:48:38,586][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:48:39,096][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:48:39,603][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:48:40,109][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:48:40,613][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:48:41,117][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:48:41,622][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:48:42,125][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:48:42,628][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:48:43,132][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:48:43,635][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:48:44,135][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:48:44,637][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:48:45,140][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:48:45,644][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:48:46,150][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:48:46,657][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:48:47,172][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:48:47,682][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:48:48,185][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:48:48,696][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:48:49,202][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:48:49,706][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:48:50,213][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:48:50,709][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:48:51,205][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:48:51,702][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:48:52,201][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:48:52,699][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:48:53,198][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:48:53,701][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:48:54,204][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:48:54,707][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:48:55,209][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:48:55,709][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:48:56,204][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:48:56,710][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:48:57,213][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:48:57,714][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:48:58,219][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:48:58,723][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:48:59,226][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:48:59,729][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:49:00,228][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:49:00,741][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:49:01,243][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:49:01,741][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:49:02,237][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:49:02,743][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:49:03,242][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:49:03,741][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:49:04,238][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:49:04,743][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:49:05,239][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:49:05,737][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:49:06,239][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:49:06,740][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:49:07,237][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:49:07,729][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:49:08,221][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:49:08,725][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:49:09,220][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:49:09,720][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:49:10,219][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9798 tokens. +[2026-03-26 09:49:11,141][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.05%, Current % of VRAM taken: 60.48%, Block Peak % of device VRAM: 62.32%, ΔTime: 00:00:33 +[2026-03-26 09:49:11,895][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:49:11,898][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:49:11,899][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:49:12,626][__main__][INFO] - Iteration 765 took 1m 0s (39.83% Gen, 58.97% Train). Generation: 24s, Training: 35s. Estimated remaining time: 38h 31m 29s. Estimated total time: 50h 19m 53s. Time estimates for 10 more iterations: 10m 3s, 100 more iterations: 1h 40m 39s, 500 more iterations: 8h 23m 18s. +[2026-03-26 09:49:12,628][__main__][INFO] - Starting iteration 765. +[2026-03-26 09:49:13,028][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 76 and human policies 1. +[2026-03-26 09:49:13,029][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:49:21,999][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 09:49:33,769][__main__][INFO] - Number of regex retries in iteration 765: 1 +[2026-03-26 09:49:33,770][__main__][INFO] - agents played in iteration 765 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:49:34,553][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:49:34,577][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:49:34,600][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:49:34,624][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:49:34,625][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:49:34,626][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:49:35,541][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:49:36,001][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:49:36,506][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:49:37,008][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:49:37,508][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:49:38,004][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:49:38,502][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:49:39,005][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:49:39,508][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:49:40,005][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:49:40,501][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:49:41,004][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:49:41,507][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:49:42,013][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:49:42,515][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:49:43,011][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:49:43,507][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:49:44,003][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:49:44,498][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:49:44,998][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:49:45,499][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:49:45,999][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:49:46,496][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:49:47,002][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:49:47,499][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:49:47,992][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:49:48,490][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:49:48,989][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:49:49,488][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:49:49,991][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:49:50,489][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:49:50,991][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:49:51,494][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:49:51,994][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:49:52,501][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:49:53,013][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:49:53,521][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:49:54,022][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:49:54,525][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:49:55,022][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:49:55,522][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:49:56,021][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:49:56,520][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:49:57,013][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:49:57,507][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:49:58,000][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:49:58,497][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:49:58,991][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:49:59,513][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:50:00,011][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:50:00,508][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:50:01,006][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:50:01,503][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:50:02,000][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:50:02,501][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:50:03,000][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:50:03,508][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:50:04,006][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:50:04,502][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:50:05,002][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:50:05,496][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:50:05,991][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:50:06,488][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:50:06,980][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:50:07,475][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9667 tokens. +[2026-03-26 09:50:08,401][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.93%, Current % of VRAM taken: 60.36%, Block Peak % of device VRAM: 62.05%, ΔTime: 00:00:32 +[2026-03-26 09:50:09,157][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:50:09,159][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:50:09,161][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:50:09,890][__main__][INFO] - Iteration 766 took 56s (36.48% Gen, 62.24% Train). Generation: 20s, Training: 35s. Estimated remaining time: 35h 33m 46s. Estimated total time: 47h 23m 7s. Time estimates for 10 more iterations: 9m 28s, 100 more iterations: 1h 34m 46s, 500 more iterations: 7h 53m 51s. +[2026-03-26 09:50:09,892][__main__][INFO] - Starting iteration 766. +[2026-03-26 09:50:10,291][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 76 and human policies 1. +[2026-03-26 09:50:10,292][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:50:32,965][__main__][INFO] - Number of regex retries in iteration 766: 0 +[2026-03-26 09:50:32,965][__main__][INFO] - agents played in iteration 766 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:50:33,749][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:50:33,776][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:50:33,801][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:50:33,826][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.45%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:50:33,827][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:50:33,827][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:50:34,718][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:50:35,174][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:50:35,674][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:50:36,172][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:50:36,670][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:50:37,170][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:50:37,675][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:50:38,178][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:50:38,685][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:50:39,184][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:50:39,679][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:50:40,175][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:50:40,673][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:50:41,170][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:50:41,675][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:50:42,179][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:50:42,677][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:50:43,183][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:50:43,681][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:50:44,175][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:50:44,673][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:50:45,170][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:50:45,666][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:50:46,163][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:50:46,659][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:50:47,152][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:50:47,648][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:50:48,144][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:50:48,638][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:50:49,137][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:50:49,642][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:50:50,138][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:50:50,637][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:50:51,130][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:50:51,628][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:50:52,126][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:50:52,627][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:50:53,123][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:50:53,619][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:50:54,117][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:50:54,613][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:50:55,111][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:50:55,613][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:50:56,142][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:50:56,654][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:50:57,150][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:50:57,653][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:50:58,150][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:50:58,653][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:50:59,149][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:50:59,648][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:51:00,148][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:51:00,647][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:51:01,145][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:51:01,642][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:51:02,137][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:51:02,635][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:51:03,128][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:51:03,621][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:51:04,119][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:51:04,611][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:51:05,103][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:51:05,598][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:51:06,095][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:51:06,587][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9557 tokens. +[2026-03-26 09:51:07,504][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.94%, Current % of VRAM taken: 60.36%, Block Peak % of device VRAM: 62.07%, ΔTime: 00:00:32 +[2026-03-26 09:51:08,264][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:51:08,267][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:51:08,268][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:51:09,006][__main__][INFO] - Iteration 767 took 58s (38.62% Gen, 60.12% Train). Generation: 22s, Training: 35s. Estimated remaining time: 37h 5m 29s. Estimated total time: 48h 55m 49s. Time estimates for 10 more iterations: 9m 47s, 100 more iterations: 1h 37m 51s, 500 more iterations: 8h 9m 18s. +[2026-03-26 09:51:09,009][__main__][INFO] - Starting iteration 767. +[2026-03-26 09:51:09,424][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 76 and human policies 1. +[2026-03-26 09:51:09,425][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:51:13,893][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 09:51:25,012][__main__][INFO] - Number of regex retries in iteration 767: 1 +[2026-03-26 09:51:25,013][__main__][INFO] - agents played in iteration 767 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:51:25,842][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:51:25,866][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:51:25,890][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:51:25,913][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:51:25,914][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:51:25,914][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:51:26,794][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:51:27,243][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:51:27,745][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:51:28,249][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:51:28,752][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:51:29,251][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:51:29,757][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:51:30,264][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:51:30,767][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:51:31,265][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:51:31,768][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:51:32,268][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:51:32,767][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:51:33,264][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:51:33,767][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:51:34,264][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:51:34,759][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:51:35,252][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:51:35,747][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:51:36,240][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:51:36,733][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:51:37,234][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:51:37,731][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:51:38,228][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:51:38,724][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:51:39,224][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:51:39,726][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:51:40,234][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:51:40,740][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:51:41,242][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:51:41,741][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:51:42,266][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:51:42,767][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:51:43,268][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:51:43,771][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:51:44,270][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:51:44,770][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:51:45,270][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:51:45,768][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:51:46,268][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:51:46,768][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:51:47,267][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:51:47,768][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:51:48,264][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:51:48,765][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:51:49,269][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:51:49,766][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:51:50,267][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:51:50,767][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:51:51,261][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:51:51,754][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:51:52,246][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:51:52,740][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:51:53,236][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:51:53,729][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:51:54,227][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:51:54,723][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:51:55,224][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:51:55,728][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:51:56,227][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:51:56,725][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:51:57,223][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:51:57,723][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:51:58,231][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:51:58,730][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9693 tokens. +[2026-03-26 09:51:59,645][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.09%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.18%, ΔTime: 00:00:32 +[2026-03-26 09:52:00,393][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:52:00,395][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:52:00,397][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:52:01,126][__main__][INFO] - Iteration 768 took 51s (30.15% Gen, 68.44% Train). Generation: 15s, Training: 35s. Estimated remaining time: 31h 13m 57s. Estimated total time: 43h 5m 10s. Time estimates for 10 more iterations: 8m 37s, 100 more iterations: 1h 26m 10s, 500 more iterations: 7h 10m 51s. +[2026-03-26 09:52:01,129][__main__][INFO] - Starting iteration 768. +[2026-03-26 09:52:01,541][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 76 and human policies 1. +[2026-03-26 09:52:01,541][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:52:05,632][mllm.models.large_language_model_local][WARNING] - Response Proposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 09:52:15,482][__main__][INFO] - Number of regex retries in iteration 768: 1 +[2026-03-26 09:52:15,482][__main__][INFO] - agents played in iteration 768 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:52:16,279][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:52:16,303][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:52:16,327][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:52:16,351][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.51%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:52:16,352][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:52:16,352][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:52:17,224][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:52:17,677][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:52:18,192][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:52:18,688][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:52:19,189][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:52:19,685][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:52:20,180][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:52:20,674][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:52:21,171][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:52:21,681][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:52:22,181][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:52:22,683][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:52:23,182][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:52:23,681][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:52:24,180][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:52:24,683][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:52:25,185][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:52:25,684][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:52:26,185][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:52:26,685][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:52:27,189][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:52:27,688][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:52:28,190][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:52:28,702][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:52:29,205][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:52:29,711][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:52:30,213][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:52:30,717][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:52:31,222][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:52:31,733][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:52:32,238][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:52:32,744][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:52:33,249][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:52:33,746][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:52:34,249][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:52:34,745][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:52:35,244][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:52:35,740][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:52:36,235][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:52:36,741][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:52:37,237][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:52:37,736][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:52:38,232][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:52:38,730][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:52:39,223][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:52:39,722][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:52:40,218][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:52:40,714][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:52:41,213][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:52:41,713][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:52:42,212][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:52:42,713][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:52:43,223][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:52:43,724][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:52:44,225][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:52:44,726][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:52:45,231][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:52:45,732][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:52:46,228][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:52:46,724][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:52:47,223][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:52:47,718][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:52:48,219][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:52:48,715][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:52:49,213][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9713 tokens. +[2026-03-26 09:52:50,136][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 61.96%, ΔTime: 00:00:32 +[2026-03-26 09:52:50,884][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:52:50,886][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:52:50,888][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:52:51,679][__main__][INFO] - Iteration 769 took 50s (27.80% Gen, 70.61% Train). Generation: 13s, Training: 35s. Estimated remaining time: 29h 54m 53s. Estimated total time: 41h 46m 56s. Time estimates for 10 more iterations: 8m 21s, 100 more iterations: 1h 23m 33s, 500 more iterations: 6h 57m 49s. +[2026-03-26 09:52:51,681][__main__][INFO] - Starting iteration 769. +[2026-03-26 09:52:52,098][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 76 and human policies 1. +[2026-03-26 09:52:52,098][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:53:07,374][__main__][INFO] - Number of regex retries in iteration 769: 0 +[2026-03-26 09:53:07,375][__main__][INFO] - agents played in iteration 769 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:53:08,146][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:53:08,170][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:53:08,194][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:53:08,217][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:53:08,218][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:53:08,218][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:53:09,103][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:53:09,556][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:53:10,058][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:53:10,556][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:53:11,057][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:53:11,563][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:53:12,065][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:53:12,564][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:53:13,064][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:53:13,558][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:53:14,051][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:53:14,545][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:53:15,039][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:53:15,542][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:53:16,048][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:53:16,547][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:53:17,073][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:53:17,565][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:53:18,059][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:53:18,551][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:53:19,044][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:53:19,545][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:53:20,052][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:53:20,556][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:53:21,060][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:53:21,553][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:53:22,049][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:53:22,552][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:53:23,048][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:53:23,549][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:53:24,052][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:53:24,550][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:53:25,046][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:53:25,550][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:53:26,059][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:53:26,558][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:53:27,056][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:53:27,552][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:53:28,047][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:53:28,544][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:53:29,039][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:53:29,534][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:53:30,032][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:53:30,525][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:53:31,021][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:53:31,519][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:53:32,012][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:53:32,504][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:53:33,000][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:53:33,499][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:53:33,992][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:53:34,485][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:53:34,978][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:53:35,474][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:53:35,969][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:53:36,466][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:53:36,966][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:53:37,469][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:53:37,969][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:53:38,471][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:53:38,974][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:53:39,475][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:53:39,975][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:53:40,475][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:53:40,977][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9585 tokens. +[2026-03-26 09:53:41,910][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.16%, Current % of VRAM taken: 60.59%, Block Peak % of device VRAM: 62.01%, ΔTime: 00:00:32 +[2026-03-26 09:53:42,661][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:53:42,663][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:53:42,665][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:53:43,403][__main__][INFO] - Iteration 770 took 51s (29.77% Gen, 68.78% Train). Generation: 15s, Training: 35s. Estimated remaining time: 30h 52m 24s. Estimated total time: 42h 45m 19s. Time estimates for 10 more iterations: 8m 33s, 100 more iterations: 1h 25m 30s, 500 more iterations: 7h 7m 33s. +[2026-03-26 09:53:43,406][__main__][INFO] - Starting iteration 770. +[2026-03-26 09:53:43,802][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 76 and human policies 1. +[2026-03-26 09:53:43,803][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:54:03,081][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 09:54:03,981][__main__][INFO] - Number of regex retries in iteration 770: 1 +[2026-03-26 09:54:03,982][__main__][INFO] - agents played in iteration 770 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:54:04,754][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:54:04,779][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:54:04,803][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:54:04,827][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.52%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:54:04,828][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:54:04,829][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:54:05,705][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:54:06,162][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:54:06,673][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:54:07,171][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:54:07,674][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:54:08,173][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:54:08,672][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:54:09,170][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:54:09,668][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:54:10,160][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:54:10,652][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:54:11,143][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:54:12,123][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:54:12,689][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:54:13,207][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:54:13,704][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:54:14,202][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:54:14,705][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:54:15,203][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:54:15,703][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:54:16,197][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:54:16,701][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:54:17,214][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:54:17,711][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:54:18,206][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:54:18,699][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:54:19,196][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:54:19,691][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:54:20,188][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:54:20,693][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:54:21,195][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:54:21,695][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:54:22,188][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:54:22,686][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:54:23,187][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:54:23,684][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:54:24,183][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:54:24,694][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:54:25,193][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:54:25,690][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:54:26,188][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:54:26,694][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:54:27,195][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:54:27,689][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:54:28,190][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:54:28,686][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:54:29,187][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:54:29,682][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:54:30,181][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:54:30,674][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:54:31,169][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:54:31,664][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:54:32,162][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:54:32,657][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:54:33,153][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:54:33,648][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:54:34,141][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:54:34,639][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:54:35,137][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:54:35,636][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:54:36,136][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:54:36,634][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:54:37,134][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:54:37,632][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:54:38,131][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9630 tokens. +[2026-03-26 09:54:39,081][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.11%, Current % of VRAM taken: 60.53%, Block Peak % of device VRAM: 62.00%, ΔTime: 00:00:33 +[2026-03-26 09:54:39,831][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:54:39,833][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:54:39,835][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:54:41,121][__main__][INFO] - Iteration 771 took 57s (35.20% Gen, 62.55% Train). Generation: 20s, Training: 35s. Estimated remaining time: 35h 52m 4s. Estimated total time: 47h 45m 56s. Time estimates for 10 more iterations: 9m 33s, 100 more iterations: 1h 35m 31s, 500 more iterations: 7h 57m 39s. +[2026-03-26 09:54:41,123][__main__][INFO] - Starting iteration 771. +[2026-03-26 09:54:41,527][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 77 and human policies 1. +[2026-03-26 09:54:41,527][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:55:04,748][__main__][INFO] - Number of regex retries in iteration 771: 0 +[2026-03-26 09:55:04,748][__main__][INFO] - agents played in iteration 771 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:55:05,520][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:55:05,549][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:55:05,575][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:55:05,600][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.50%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:55:05,601][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:55:05,602][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:55:06,516][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:55:06,970][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:55:07,471][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:55:07,974][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:55:08,467][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:55:08,963][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:55:09,459][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:55:09,955][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:55:10,450][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:55:10,947][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:55:11,445][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:55:11,937][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:55:12,430][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:55:12,925][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:55:13,422][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:55:13,923][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:55:14,420][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:55:14,926][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:55:15,428][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:55:15,924][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:55:16,424][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:55:16,945][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:55:17,445][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:55:17,945][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:55:18,446][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:55:18,946][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:55:19,450][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:55:19,951][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:55:20,452][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:55:20,952][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:55:21,451][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:55:21,951][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:55:22,446][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:55:22,944][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:55:23,436][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:55:23,933][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:55:24,429][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:55:24,922][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:55:25,416][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:55:25,908][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:55:26,401][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:55:26,897][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:55:27,397][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:55:27,897][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:55:28,396][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:55:28,894][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:55:29,396][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:55:29,895][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:55:30,394][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:55:30,891][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:55:31,388][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:55:31,886][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:55:32,384][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:55:32,885][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:55:33,377][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:55:33,874][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:55:34,377][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:55:34,876][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:55:35,368][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:55:35,863][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:55:36,361][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:55:36,854][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:55:37,349][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:55:37,842][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:55:38,334][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9540 tokens. +[2026-03-26 09:55:39,269][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 12.93%, Current % of VRAM taken: 59.35%, Block Peak % of device VRAM: 62.02%, ΔTime: 00:00:32 +[2026-03-26 09:55:40,020][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:55:40,022][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:55:40,024][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:55:40,750][__main__][INFO] - Iteration 772 took 59s (39.21% Gen, 59.56% Train). Generation: 23s, Training: 35s. Estimated remaining time: 37h 26m 20s. Estimated total time: 49h 21m 12s. Time estimates for 10 more iterations: 9m 52s, 100 more iterations: 1h 38m 42s, 500 more iterations: 8h 13m 32s. +[2026-03-26 09:55:40,754][__main__][INFO] - Starting iteration 772. +[2026-03-26 09:55:41,156][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 77 and human policies 1. +[2026-03-26 09:55:41,157][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:55:49,645][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 09:55:59,233][__main__][INFO] - Number of regex retries in iteration 772: 1 +[2026-03-26 09:55:59,234][__main__][INFO] - agents played in iteration 772 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:56:00,000][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:56:00,024][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:56:00,049][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:56:00,074][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 52.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:56:00,075][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:56:00,076][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:56:00,976][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:56:01,432][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:56:01,942][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:56:02,445][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:56:02,947][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:56:03,449][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:56:03,957][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:56:04,461][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:56:04,963][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:56:05,465][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:56:05,966][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:56:06,464][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:56:06,957][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:56:07,447][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:56:07,945][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:56:08,441][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:56:08,939][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:56:09,439][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:56:09,936][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:56:10,448][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:56:10,950][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:56:11,452][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:56:11,951][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:56:12,451][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:56:12,951][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:56:13,446][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:56:13,940][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:56:14,435][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:56:14,937][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:56:15,442][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:56:15,937][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:56:16,431][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:56:16,924][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:56:17,418][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:56:17,913][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:56:18,408][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:56:18,905][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:56:19,403][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:56:19,895][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:56:20,389][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:56:20,887][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:56:21,383][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:56:21,875][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:56:22,371][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:56:22,867][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:56:23,364][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:56:23,856][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:56:24,346][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:56:24,837][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:56:25,335][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:56:25,835][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:56:26,333][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:56:26,830][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:56:27,328][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:56:27,827][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:56:28,326][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:56:28,824][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:56:29,325][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:56:29,830][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:56:30,330][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:56:30,831][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:56:31,333][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:56:31,834][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:56:32,333][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:56:32,831][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9731 tokens. +[2026-03-26 09:56:33,761][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.08%, Current % of VRAM taken: 60.51%, Block Peak % of device VRAM: 62.15%, ΔTime: 00:00:32 +[2026-03-26 09:56:34,529][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:56:34,532][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:56:34,533][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:56:35,388][__main__][INFO] - Iteration 773 took 54s (33.33% Gen, 65.09% Train). Generation: 18s, Training: 35s. Estimated remaining time: 33h 15m 51s. Estimated total time: 45h 11m 38s. Time estimates for 10 more iterations: 9m 2s, 100 more iterations: 1h 30m 23s, 500 more iterations: 7h 31m 56s. +[2026-03-26 09:56:35,391][__main__][INFO] - Starting iteration 773. +[2026-03-26 09:56:35,790][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 77 and human policies 1. +[2026-03-26 09:56:35,790][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:56:40,087][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 09:56:50,257][__main__][INFO] - Number of regex retries in iteration 773: 1 +[2026-03-26 09:56:50,258][__main__][INFO] - agents played in iteration 773 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:56:51,022][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:56:51,048][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:56:51,072][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:56:51,096][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.49%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:56:51,098][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:56:51,098][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:56:51,993][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:56:52,454][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:56:52,963][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:56:53,465][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:56:53,963][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:56:54,462][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:56:54,967][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:56:55,471][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:56:55,975][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:56:56,474][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:56:56,968][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:56:57,462][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:56:57,959][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:56:58,453][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:56:58,951][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:56:59,447][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:56:59,939][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:57:00,430][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:57:00,929][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:57:01,427][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:57:01,930][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:57:02,430][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:57:02,932][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:57:03,427][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:57:03,925][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:57:04,425][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:57:04,917][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:57:05,412][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:57:05,905][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:57:06,406][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:57:06,907][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:57:07,402][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:57:07,907][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:57:08,406][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:57:08,899][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:57:09,391][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:57:09,883][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:57:10,375][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:57:10,866][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:57:11,360][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:57:11,853][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:57:12,346][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:57:12,837][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:57:13,333][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:57:13,830][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:57:14,330][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:57:14,825][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:57:15,318][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:57:15,810][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:57:16,316][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:57:16,816][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:57:17,318][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:57:17,822][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:57:18,325][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:57:18,824][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:57:19,323][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:57:19,823][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:57:20,317][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:57:20,814][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:57:21,306][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:57:21,797][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:57:22,287][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:57:22,777][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:57:23,267][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:57:23,758][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9620 tokens. +[2026-03-26 09:57:24,687][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 11.89%, Current % of VRAM taken: 58.32%, Block Peak % of device VRAM: 62.15%, ΔTime: 00:00:32 +[2026-03-26 09:57:25,437][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:57:25,440][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:57:25,441][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:57:26,162][__main__][INFO] - Iteration 774 took 50s (28.72% Gen, 69.84% Train). Generation: 14s, Training: 35s. Estimated remaining time: 30h 2m 3s. Estimated total time: 41h 58m 40s. Time estimates for 10 more iterations: 8m 23s, 100 more iterations: 1h 23m 57s, 500 more iterations: 6h 59m 46s. +[2026-03-26 09:57:26,164][__main__][INFO] - Starting iteration 774. +[2026-03-26 09:57:26,566][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 77 and human policies 1. +[2026-03-26 09:57:26,567][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:57:41,827][__main__][INFO] - Number of regex retries in iteration 774: 0 +[2026-03-26 09:57:41,828][__main__][INFO] - agents played in iteration 774 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:57:42,589][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 51.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:57:42,613][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 51.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:57:42,637][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 51.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:57:42,660][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 51.48%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:57:42,661][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:57:42,662][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:57:43,528][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:57:43,974][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:57:44,469][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:57:44,963][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:57:45,455][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:57:45,952][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:57:46,451][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:57:46,949][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:57:47,452][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:57:47,959][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:57:48,459][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:57:48,965][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:57:49,471][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:57:49,974][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:57:50,473][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:57:50,970][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:57:51,474][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:57:52,002][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:57:52,496][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:57:53,016][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:57:53,516][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:57:54,016][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:57:54,513][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:57:55,018][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:57:55,519][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:57:56,024][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:57:56,525][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:57:57,020][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:57:57,516][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:57:58,017][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:57:58,516][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:57:59,013][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:57:59,509][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:58:00,003][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:58:00,496][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:58:00,990][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:58:01,487][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:58:01,983][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:58:02,485][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:58:02,985][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:58:03,481][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:58:03,976][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:58:04,467][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:58:04,966][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:58:05,464][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:58:05,961][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:58:06,453][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:58:06,952][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:58:07,448][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:58:07,942][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:58:08,438][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:58:08,928][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:58:09,423][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:58:09,927][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:58:10,424][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:58:10,925][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:58:11,418][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:58:11,914][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:58:12,410][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:58:12,907][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:58:13,404][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:58:13,902][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:58:14,399][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:58:14,897][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:58:15,393][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9570 tokens. +[2026-03-26 09:58:16,324][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.99%, Current % of VRAM taken: 60.42%, Block Peak % of device VRAM: 62.00%, ΔTime: 00:00:32 +[2026-03-26 09:58:17,086][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:58:17,089][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:58:17,090][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:58:17,955][__main__][INFO] - Iteration 775 took 51s (29.70% Gen, 68.62% Train). Generation: 15s, Training: 35s. Estimated remaining time: 30h 51m 59s. Estimated total time: 42h 49m 28s. Time estimates for 10 more iterations: 8m 33s, 100 more iterations: 1h 25m 38s, 500 more iterations: 7h 8m 14s. +[2026-03-26 09:58:17,958][__main__][INFO] - Starting iteration 775. +[2026-03-26 09:58:18,359][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 77 and human policies 1. +[2026-03-26 09:58:18,360][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:58:33,145][__main__][INFO] - Number of regex retries in iteration 775: 0 +[2026-03-26 09:58:33,146][__main__][INFO] - agents played in iteration 775 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:58:33,919][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:58:33,944][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:58:33,970][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:58:33,995][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.47%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:58:33,996][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:58:33,997][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:58:34,879][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:58:35,330][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:58:35,833][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:58:36,327][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:58:36,820][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:58:37,318][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:58:37,818][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:58:38,311][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:58:38,816][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:58:39,314][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:58:39,808][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:58:40,302][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:58:40,800][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:58:41,294][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:58:41,786][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:58:42,279][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:58:42,787][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:58:43,283][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:58:43,781][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:58:44,282][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:58:44,777][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:58:45,271][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:58:45,766][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:58:46,269][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:58:46,764][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:58:47,269][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:58:47,769][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:58:48,268][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:58:48,768][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:58:49,267][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:58:49,765][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:58:50,266][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:58:50,765][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:58:51,262][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:58:51,754][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:58:52,251][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:58:52,748][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:58:53,244][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:58:53,737][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:58:54,236][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:58:54,730][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:58:55,229][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:58:55,729][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:58:56,233][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:58:56,736][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:58:57,239][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:58:57,740][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:58:58,242][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:58:58,744][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:58:59,245][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:58:59,739][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:59:00,231][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:59:00,725][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:59:01,218][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:59:01,710][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:59:02,206][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:59:02,715][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:59:03,217][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:59:03,718][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:59:04,216][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:59:04,714][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:59:05,211][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:59:05,709][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:59:06,211][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:59:06,715][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9626 tokens. +[2026-03-26 09:59:07,675][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.14%, Current % of VRAM taken: 60.56%, Block Peak % of device VRAM: 61.99%, ΔTime: 00:00:32 +[2026-03-26 09:59:08,435][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 09:59:08,441][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 09:59:08,443][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 09:59:09,309][__main__][INFO] - Iteration 776 took 50s (29.02% Gen, 69.28% Train). Generation: 14s, Training: 35s. Estimated remaining time: 30h 29m 12s. Estimated total time: 42h 27m 33s. Time estimates for 10 more iterations: 8m 29s, 100 more iterations: 1h 24m 55s, 500 more iterations: 7h 4m 35s. +[2026-03-26 09:59:09,312][__main__][INFO] - Starting iteration 776. +[2026-03-26 09:59:09,713][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 77 and human policies 1. +[2026-03-26 09:59:09,714][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 09:59:14,503][mllm.models.large_language_model_local][WARNING] - Response Proposal: 0 hats, 10 books, 0 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 +[2026-03-26 09:59:23,713][__main__][INFO] - Number of regex retries in iteration 776: 1 +[2026-03-26 09:59:23,714][__main__][INFO] - agents played in iteration 776 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 09:59:24,497][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:59:24,522][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:59:24,546][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:59:24,571][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.53%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 09:59:24,571][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 09:59:24,572][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 09:59:25,453][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 09:59:25,906][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 09:59:26,408][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 09:59:26,907][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 09:59:27,408][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 09:59:27,906][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 09:59:28,408][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 09:59:28,909][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 09:59:29,408][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 09:59:29,907][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 09:59:30,411][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 09:59:30,909][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 09:59:31,409][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 09:59:31,908][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 09:59:32,410][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 09:59:32,916][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 09:59:33,417][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 09:59:33,918][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 09:59:34,414][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 09:59:34,910][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 09:59:35,409][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 09:59:35,906][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 09:59:36,409][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 09:59:36,915][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 09:59:37,415][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 09:59:37,917][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 09:59:38,420][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 09:59:38,926][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 09:59:39,436][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 09:59:41,552][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 09:59:42,053][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 09:59:42,552][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 09:59:43,051][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 09:59:43,550][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 09:59:44,048][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 09:59:44,547][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 09:59:45,043][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 09:59:45,542][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 09:59:46,041][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 09:59:46,537][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 09:59:47,032][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 09:59:47,531][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 09:59:48,029][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 09:59:48,527][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 09:59:49,025][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 09:59:49,524][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 09:59:50,023][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 09:59:50,522][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 09:59:51,020][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 09:59:51,521][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 09:59:52,029][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 09:59:52,528][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 09:59:53,025][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 09:59:53,528][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 09:59:54,026][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 09:59:54,525][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 09:59:55,025][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 09:59:55,524][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 09:59:56,022][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 09:59:56,521][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 09:59:57,020][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 09:59:57,518][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 09:59:58,016][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 09:59:58,513][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 09:59:59,004][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9834 tokens. +[2026-03-26 09:59:59,917][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 13.87%, Current % of VRAM taken: 60.30%, Block Peak % of device VRAM: 62.07%, ΔTime: 00:00:34 +[2026-03-26 10:00:00,671][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 10:00:00,673][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 10:00:00,675][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 10:00:01,481][__main__][INFO] - Iteration 777 took 51s (27.04% Gen, 71.40% Train). Generation: 14s, Training: 36s. Estimated remaining time: 31h 9m 11s. Estimated total time: 43h 8m 24s. Time estimates for 10 more iterations: 8m 37s, 100 more iterations: 1h 26m 16s, 500 more iterations: 7h 11m 24s. +[2026-03-26 10:00:01,483][__main__][INFO] - Starting iteration 777. +[2026-03-26 10:00:01,885][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 77 and human policies 1. +[2026-03-26 10:00:01,885][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 10:00:17,931][__main__][INFO] - Number of regex retries in iteration 777: 0 +[2026-03-26 10:00:17,932][__main__][INFO] - agents played in iteration 777 are Bob_buffer, Alice, Bob, Alice_buffer +[2026-03-26 10:00:18,702][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 10:00:18,726][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 10:00:18,750][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 10:00:18,773][mllm.training.trainer_ad_align][INFO] - For task: Get advantages with critic gradient accumulation, ΔVRAM % (total): 0.00%, Current % of VRAM taken: 53.44%, Block Peak % of device VRAM: 19.52%, ΔTime: 00:00:00 +[2026-03-26 10:00:18,774][mllm.training.trainer_ad_align][INFO] - Sharing advantage alignment data. +[2026-03-26 10:00:18,774][mllm.training.trainer_ad_align][INFO] - Receiving advantage packets. +[2026-03-26 10:00:19,662][mllm.training.trainer_common][INFO] - Processing mini-batch 0 of 64 +[2026-03-26 10:00:20,115][mllm.training.trainer_common][INFO] - Processing mini-batch 1 of 64 +[2026-03-26 10:00:20,617][mllm.training.trainer_common][INFO] - Processing mini-batch 2 of 64 +[2026-03-26 10:00:21,113][mllm.training.trainer_common][INFO] - Processing mini-batch 3 of 64 +[2026-03-26 10:00:21,607][mllm.training.trainer_common][INFO] - Processing mini-batch 4 of 64 +[2026-03-26 10:00:22,099][mllm.training.trainer_common][INFO] - Processing mini-batch 5 of 64 +[2026-03-26 10:00:22,589][mllm.training.trainer_common][INFO] - Processing mini-batch 6 of 64 +[2026-03-26 10:00:23,085][mllm.training.trainer_common][INFO] - Processing mini-batch 7 of 64 +[2026-03-26 10:00:23,582][mllm.training.trainer_common][INFO] - Processing mini-batch 8 of 64 +[2026-03-26 10:00:24,074][mllm.training.trainer_common][INFO] - Processing mini-batch 9 of 64 +[2026-03-26 10:00:24,573][mllm.training.trainer_common][INFO] - Processing mini-batch 10 of 64 +[2026-03-26 10:00:25,066][mllm.training.trainer_common][INFO] - Processing mini-batch 11 of 64 +[2026-03-26 10:00:25,568][mllm.training.trainer_common][INFO] - Processing mini-batch 12 of 64 +[2026-03-26 10:00:26,061][mllm.training.trainer_common][INFO] - Processing mini-batch 13 of 64 +[2026-03-26 10:00:26,554][mllm.training.trainer_common][INFO] - Processing mini-batch 14 of 64 +[2026-03-26 10:00:27,056][mllm.training.trainer_common][INFO] - Processing mini-batch 15 of 64 +[2026-03-26 10:00:27,555][mllm.training.trainer_common][INFO] - Processing mini-batch 16 of 64 +[2026-03-26 10:00:28,056][mllm.training.trainer_common][INFO] - Processing mini-batch 17 of 64 +[2026-03-26 10:00:28,550][mllm.training.trainer_common][INFO] - Processing mini-batch 18 of 64 +[2026-03-26 10:00:29,048][mllm.training.trainer_common][INFO] - Processing mini-batch 19 of 64 +[2026-03-26 10:00:29,544][mllm.training.trainer_common][INFO] - Processing mini-batch 20 of 64 +[2026-03-26 10:00:30,044][mllm.training.trainer_common][INFO] - Processing mini-batch 21 of 64 +[2026-03-26 10:00:30,544][mllm.training.trainer_common][INFO] - Processing mini-batch 22 of 64 +[2026-03-26 10:00:31,039][mllm.training.trainer_common][INFO] - Processing mini-batch 23 of 64 +[2026-03-26 10:00:31,532][mllm.training.trainer_common][INFO] - Processing mini-batch 24 of 64 +[2026-03-26 10:00:32,032][mllm.training.trainer_common][INFO] - Processing mini-batch 25 of 64 +[2026-03-26 10:00:32,526][mllm.training.trainer_common][INFO] - Processing mini-batch 26 of 64 +[2026-03-26 10:00:33,019][mllm.training.trainer_common][INFO] - Processing mini-batch 27 of 64 +[2026-03-26 10:00:33,512][mllm.training.trainer_common][INFO] - Processing mini-batch 28 of 64 +[2026-03-26 10:00:34,005][mllm.training.trainer_common][INFO] - Processing mini-batch 29 of 64 +[2026-03-26 10:00:34,506][mllm.training.trainer_common][INFO] - Processing mini-batch 30 of 64 +[2026-03-26 10:00:35,000][mllm.training.trainer_common][INFO] - Processing mini-batch 31 of 64 +[2026-03-26 10:00:35,492][mllm.training.trainer_common][INFO] - Processing mini-batch 32 of 64 +[2026-03-26 10:00:35,989][mllm.training.trainer_common][INFO] - Processing mini-batch 33 of 64 +[2026-03-26 10:00:36,484][mllm.training.trainer_common][INFO] - Processing mini-batch 34 of 64 +[2026-03-26 10:00:36,978][mllm.training.trainer_common][INFO] - Processing mini-batch 35 of 64 +[2026-03-26 10:00:37,476][mllm.training.trainer_common][INFO] - Processing mini-batch 36 of 64 +[2026-03-26 10:00:37,969][mllm.training.trainer_common][INFO] - Processing mini-batch 37 of 64 +[2026-03-26 10:00:38,461][mllm.training.trainer_common][INFO] - Processing mini-batch 38 of 64 +[2026-03-26 10:00:38,954][mllm.training.trainer_common][INFO] - Processing mini-batch 39 of 64 +[2026-03-26 10:00:39,449][mllm.training.trainer_common][INFO] - Processing mini-batch 40 of 64 +[2026-03-26 10:00:39,942][mllm.training.trainer_common][INFO] - Processing mini-batch 41 of 64 +[2026-03-26 10:00:40,436][mllm.training.trainer_common][INFO] - Processing mini-batch 42 of 64 +[2026-03-26 10:00:40,932][mllm.training.trainer_common][INFO] - Processing mini-batch 43 of 64 +[2026-03-26 10:00:41,425][mllm.training.trainer_common][INFO] - Processing mini-batch 44 of 64 +[2026-03-26 10:00:41,922][mllm.training.trainer_common][INFO] - Processing mini-batch 45 of 64 +[2026-03-26 10:00:42,414][mllm.training.trainer_common][INFO] - Processing mini-batch 46 of 64 +[2026-03-26 10:00:42,905][mllm.training.trainer_common][INFO] - Processing mini-batch 47 of 64 +[2026-03-26 10:00:43,403][mllm.training.trainer_common][INFO] - Processing mini-batch 48 of 64 +[2026-03-26 10:00:43,894][mllm.training.trainer_common][INFO] - Processing mini-batch 49 of 64 +[2026-03-26 10:00:44,384][mllm.training.trainer_common][INFO] - Processing mini-batch 50 of 64 +[2026-03-26 10:00:44,879][mllm.training.trainer_common][INFO] - Processing mini-batch 51 of 64 +[2026-03-26 10:00:45,369][mllm.training.trainer_common][INFO] - Processing mini-batch 52 of 64 +[2026-03-26 10:00:45,870][mllm.training.trainer_common][INFO] - Processing mini-batch 53 of 64 +[2026-03-26 10:00:46,361][mllm.training.trainer_common][INFO] - Processing mini-batch 54 of 64 +[2026-03-26 10:00:46,856][mllm.training.trainer_common][INFO] - Processing mini-batch 55 of 64 +[2026-03-26 10:00:47,349][mllm.training.trainer_common][INFO] - Processing mini-batch 56 of 64 +[2026-03-26 10:00:47,848][mllm.training.trainer_common][INFO] - Processing mini-batch 57 of 64 +[2026-03-26 10:00:48,342][mllm.training.trainer_common][INFO] - Processing mini-batch 58 of 64 +[2026-03-26 10:00:48,839][mllm.training.trainer_common][INFO] - Processing mini-batch 59 of 64 +[2026-03-26 10:00:49,334][mllm.training.trainer_common][INFO] - Processing mini-batch 60 of 64 +[2026-03-26 10:00:49,832][mllm.training.trainer_common][INFO] - Processing mini-batch 61 of 64 +[2026-03-26 10:00:50,327][mllm.training.trainer_common][INFO] - Processing mini-batch 62 of 64 +[2026-03-26 10:00:50,819][mllm.training.trainer_common][INFO] - Processing mini-batch 63 of 64 +[2026-03-26 10:00:51,319][mllm.training.trainer_common][INFO] - Accumulated the policy gradient loss for 9495 tokens. +[2026-03-26 10:00:52,235][mllm.training.trainer_common][INFO] - For task: Apply reinforce step, ΔVRAM % (total): 14.07%, Current % of VRAM taken: 60.50%, Block Peak % of device VRAM: 61.95%, ΔTime: 00:00:32 +[2026-03-26 10:00:52,984][mllm.training.trainer_common][INFO] - Saved main optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/policy_optimizer_state.pt +[2026-03-26 10:00:52,987][mllm.training.trainer_common][INFO] - Saved critic optimizer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/critic_optimizer_state.pt +[2026-03-26 10:00:52,988][mllm.training.trainer_common][INFO] - Saved trainer state to /scratch/muqeeth/llm_negotiation/2026_03/split_no_comm_vanilla_ad_align_beta2_seed42/seed_42/agent_trainer/trainer_annealing_state.pkl +[2026-03-26 10:00:53,719][__main__][INFO] - Iteration 778 took 51s (30.96% Gen, 67.63% Train). Generation: 16s, Training: 35s. Estimated remaining time: 31h 11m 40s. Estimated total time: 43h 11m 45s. Time estimates for 10 more iterations: 8m 38s, 100 more iterations: 1h 26m 23s, 500 more iterations: 7h 11m 57s. +[2026-03-26 10:00:53,722][__main__][INFO] - Starting iteration 778. +[2026-03-26 10:00:54,123][__main__][INFO] - Inference policies count is regular policies 2 and buffer policies 77 and human policies 1. +[2026-03-26 10:00:54,124][__main__][INFO] - Hard coded buffer agents are set to False with prob 0 +[2026-03-26 10:01:01,164][mllm.models.large_language_model_local][WARNING] - Response Propposal: 10 hats, 0 books, 10 balls did not match regex: (?i)Proposal:\s*((?:\s*(?P(10|[0-9]))\s*(?Phats?|books?|balls?)\s*,?)+), retry 1/3 diff --git a/src_code_for_reproducibility/__pycache__/__init__.cpython-312.pyc b/src_code_for_reproducibility/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..1e0ce601325acab2a3935c9b22e10e27fcccc961 Binary files /dev/null and b/src_code_for_reproducibility/__pycache__/__init__.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/chat_utils/apply_template.py b/src_code_for_reproducibility/chat_utils/apply_template.py new file mode 100644 index 0000000000000000000000000000000000000000..6bbdc32dbb1df0407ff24ae90395dba0d162bf7d --- /dev/null +++ b/src_code_for_reproducibility/chat_utils/apply_template.py @@ -0,0 +1,89 @@ +""" +File: mllm/chat_utils/apply_template.py +Summary: Applies tokenizer-specific chat templates and stitches chat token IDs. +""" + +import torch + +from mllm.chat_utils.chat_turn import ChatTurn +from mllm.chat_utils.template_specific import ( + custom_gemma3_template, + custom_llama3_template, + custom_qwen2_template, + custom_qwen3_template, + gemma3_assistant_postfix, + qwen2_assistant_postfix, + qwen3_assistant_postfix, +) + + +def get_custom_chat_template(tokenizer) -> str: + """ + Get the chat template for the tokenizer. + """ + if "qwen2" in tokenizer.name_or_path.lower(): + return custom_qwen2_template + elif "llama" in tokenizer.name_or_path.lower(): + return custom_llama3_template + elif "qwen3" in tokenizer.name_or_path.lower(): + return custom_qwen3_template + elif "gemma" in tokenizer.name_or_path.lower(): + return custom_gemma3_template + else: + raise ValueError(f"Tokenizer {tokenizer.name_or_path} not supported") + + +def get_custom_assistant_postfix(tokenizer) -> torch.Tensor: + """ + Get the custom assistant postfix for the tokenizer. + """ + if "qwen2" in tokenizer.name_or_path.lower(): + return qwen2_assistant_postfix + elif "qwen3" in tokenizer.name_or_path.lower(): + return qwen3_assistant_postfix + elif "gemma" in tokenizer.name_or_path.lower(): + return gemma3_assistant_postfix + return torch.tensor([], dtype=torch.long) + + +def tokenize_chats(chats: list[ChatTurn], tokenizer, enable_thinking) -> None: + """ + Set the chat_template_token_ids for each chat turn. + We rely on tokenizer-side templates because engine-provided cached tokens are not exposed yet. + """ + custom_template = get_custom_chat_template(tokenizer) + custom_assistant_postfix: torch.Tensor = get_custom_assistant_postfix(tokenizer) + for i, chat in enumerate(chats): + if chat.chat_template_token_ids is None: + if chat.role == "user": + next_chat = chats[i + 1] if i + 1 < len(chats) else None + add_generation_prompt = True + if next_chat and next_chat.role == "user": + add_generation_prompt = False + encoded_chat = tokenizer.apply_chat_template( + [chat], + return_tensors="pt", + chat_template=custom_template, + add_generation_prompt=add_generation_prompt, + add_system_prompt=True if i == 0 else False, + enable_thinking=enable_thinking, + ).flatten() + previous_chat = chats[i - 1] if i > 0 else None + if previous_chat and previous_chat.role == "assistant": + encoded_chat = torch.cat([custom_assistant_postfix, encoded_chat]) + elif chat.role == "assistant": + encoded_chat = chat.out_token_ids + chat.chat_template_token_ids = encoded_chat + + +def chat_turns_to_token_ids( + chats: list[ChatTurn], tokenizer, enable_thinking +) -> list[int]: + """ + Tokenize the chat turns and set the chat_template_token_ids for each chat turn. + """ + tokenize_chats(chats=chats, tokenizer=tokenizer, enable_thinking=enable_thinking) + token_ids = [] + for chat in chats: + token_ids.append(chat.chat_template_token_ids) + return torch.cat(token_ids) diff --git a/src_code_for_reproducibility/chat_utils/chat_turn.py b/src_code_for_reproducibility/chat_utils/chat_turn.py new file mode 100644 index 0000000000000000000000000000000000000000..cfc0d9422a6070c86b1da8abce17ad28816fb2eb --- /dev/null +++ b/src_code_for_reproducibility/chat_utils/chat_turn.py @@ -0,0 +1,32 @@ +""" +File: mllm/chat_utils/chat_turn.py +Summary: Defines the ChatTurn schema plus helpers for serialization and validation. +""" + +from __future__ import annotations + +import json +from dataclasses import dataclass +from pathlib import Path +from typing import Any, List, Literal, Optional, Tuple + +import jsonschema +import torch +from pydantic import BaseModel, ConfigDict, Field, model_validator + +AgentId = str + + +class ChatTurn(BaseModel): + model_config = ConfigDict(arbitrary_types_allowed=True) # needed for torch tensors + + role: str = Field(pattern="^(user|assistant)$") + agent_id: AgentId # ID of the agent with which the chat occured + content: str + reasoning_content: str | None = None + chat_template_token_ids: torch.LongTensor | None = None # Token ids of chat template format. For example, token ids of "{content}"" + out_token_ids: torch.LongTensor | None = ( + None # tokens generated from inference engine + ) + log_probs: torch.FloatTensor | None = None + is_state_end: bool = False # indicates whether this chat turn marks the end of a state in the trajectory diff --git a/src_code_for_reproducibility/chat_utils/template_specific.py b/src_code_for_reproducibility/chat_utils/template_specific.py new file mode 100644 index 0000000000000000000000000000000000000000..c22328455c55f0b0a02439efdacf6b09234d7981 --- /dev/null +++ b/src_code_for_reproducibility/chat_utils/template_specific.py @@ -0,0 +1,114 @@ +""" +File: mllm/chat_utils/template_specific.py +Summary: Stores chat template variants and assistant postfix tensors per tokenizer. +""" + +import huggingface_hub +import torch +from transformers import AutoTokenizer + +custom_llama3_template = """ +{%- if add_system_prompt %} + {{- '<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|>' }} +{%- endif %} +{%- for message in messages %} + {{- '<|start_header_id|>' + message['role'] + '<|end_header_id|>\n\n' + message['content'] | trim + '<|eot_id|>' }} +{%- endfor %} + +{%- if add_generation_prompt %} + {{- '<|start_header_id|>' + 'assistant' + '<|end_header_id|>\n\n' }} +{%- endif %} +""" + +qwen2_assistant_postfix = ( + AutoTokenizer.from_pretrained("Qwen/Qwen2.5-7B-Instruct") + .encode("\n", return_tensors="pt") + .flatten() +) +qwen3_assistant_postfix = ( + AutoTokenizer.from_pretrained("Qwen/Qwen3-8B") + .encode("\n", return_tensors="pt") + .flatten() +) +gemma3_assistant_postfix = ( + AutoTokenizer.from_pretrained("google/gemma-3-4b-it") + .encode("\n", return_tensors="pt") + .flatten() +) +custom_qwen2_template = """ +{%- if add_system_prompt %} + {{- '<|im_start|>system\nYou are Qwen, created by Alibaba Cloud. You are a helpful assistant.<|im_end|>\n' }} +{%- endif %} +{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %} +{%- for message in messages %} + {%- if message.content is string %} + {%- set content = message.content %} + {%- else %} + {%- set content = '' %} + {%- endif %} + {%- if (message.role == "user") %} + {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }} + {%- elif message.role == "assistant" %} + {%- set reasoning_content = '' %} + {%- if message.reasoning_content is string %} + {%- set reasoning_content = message.reasoning_content %} + {%- else %} + {%- if '' in content %} + {%- set reasoning_content = content.split('')[0].rstrip('\n').split('')[-1].lstrip('\n') %} + {%- set content = content.split('')[-1].lstrip('\n') %} + {%- endif %} + {%- endif %} + {%- if loop.index0 > ns.last_query_index %} + {%- if reasoning_content %} + {{- '<|im_start|>' + message.role + '\n\n' + reasoning_content.strip('\n') + '\n\n\n' + content.lstrip('\n') }} + {%- else %} + {{- '<|im_start|>' + message.role + '\n' + content }} + {%- endif %} + {%- else %} + {{- '<|im_start|>' + message.role + '\n' + content }} + {%- endif %} + {{- '<|im_end|>\n' }} + {%- endif %} +{%- endfor %} +{%- if add_generation_prompt %} + {{- '<|im_start|>assistant\n' }} +{%- endif %} +""" + +custom_qwen3_template = """ +{%- for message in messages %} + {%- if message.content is string %} + {%- set content = message.content %} + {%- else %} + {%- set content = '' %} + {%- endif %} + {%- if (message.role == "user") %} + {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }} + {%- elif message.role == "assistant" %} + {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }} + {%- endif %} +{%- endfor %} +{%- if add_generation_prompt %} + {{- '<|im_start|>assistant\n' }} + {%- if enable_thinking is defined and enable_thinking is false %} + {{- '\n\n\n\n' }} + {%- endif %} +{%- endif %} +""" + +custom_gemma3_template = """ +{%- if add_system_prompt %} +{{- bos_token -}} +{%- endif %} +{%- for message in messages -%} +{%- if message['role'] == 'assistant' -%} +{%- set role = 'model' -%} +{%- else -%} +{%- set role = message['role'] -%} +{%- endif -%} +{{ '' + role + '\n' + message['content'] | trim + '\n' }} +{%- endfor -%} +{%- if add_generation_prompt -%} +{{ 'model\n' }} +{%- endif -%} +""" diff --git a/src_code_for_reproducibility/markov_games/__init__.py b/src_code_for_reproducibility/markov_games/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..2a7015344d8ac9b53f2660d4e837f709908213db --- /dev/null +++ b/src_code_for_reproducibility/markov_games/__init__.py @@ -0,0 +1,4 @@ +""" +File: mllm/markov_games/__init__.py +Summary: Makes Markov-game subpackages importable from the top-level namespace. +""" diff --git a/src_code_for_reproducibility/markov_games/agent.py b/src_code_for_reproducibility/markov_games/agent.py new file mode 100644 index 0000000000000000000000000000000000000000..56406ae2695ce97ad7fa4fc436908904ee11be9f --- /dev/null +++ b/src_code_for_reproducibility/markov_games/agent.py @@ -0,0 +1,72 @@ +""" +File: mllm/markov_games/agent.py +Summary: Declares the base Agent interface connecting simulations to policy calls. +""" + +from abc import ABC, abstractmethod +from collections.abc import Callable +from typing import Any, Tuple + +from numpy.random import default_rng + +from mllm.markov_games.rollout_tree import AgentActLog + + +class Agent(ABC): + """Abstract policy wrapper that bridges simulations with arbitrary backends.""" + + @abstractmethod + def __init__( + self, + seed: int, + agent_id: str, + agent_name: str, + agent_policy: Callable[[list[dict]], str], + *args, + **kwargs, + ): + """ + Initialize the agent state and seed its RNG. + + Subclasses typically store extra handles (tokenizers, inference clients, etc.) + but they should always call ``super().__init__`` so sampling remains reproducible. + """ + self.seed = seed + self.agent_id = agent_id + self.agent_name = agent_name + self.policy = policy + self.rng = default_rng(self.seed) + raise NotImplementedError + + async def act(self, observation) -> Tuple[Any, AgentActLog]: + """ + Produce the next action (and associated chat log) given an environment observation. + + Implementations can iterate with rejection sampling, multi-call deliberation, etc. + Returns both the chosen action and an `AgentActLog` describing how it was produced. + """ + raise NotImplementedError + + def get_safe_copy(self): + """ + Return a deep copy whose future calls do not mutate the original agent. + + Needed for branch exploration/reruns with alternative actions. + """ + raise NotImplementedError + + def reset(self): + """Reset any internal state between rollouts.""" + raise NotImplementedError + + def render(self): + """Optional human-readable visualization of the agent (CLI/UI).""" + raise NotImplementedError + + def close(self): + """Release any external resources (network sockets, subprocesses, etc.).""" + raise NotImplementedError + + def get_agent_info(self): + """Return diagnostic metadata to embed inside rollout logs.""" + raise NotImplementedError diff --git a/src_code_for_reproducibility/markov_games/alternative_actions_runner.py b/src_code_for_reproducibility/markov_games/alternative_actions_runner.py new file mode 100644 index 0000000000000000000000000000000000000000..d5165a2552019aefdf281c2bd41e50d204713921 --- /dev/null +++ b/src_code_for_reproducibility/markov_games/alternative_actions_runner.py @@ -0,0 +1,146 @@ +""" +File: mllm/markov_games/alternative_actions_runner.py +Summary: Generates rollout branches by replaying trajectories with unilateral action changes. +""" + +import asyncio +import copy +import json +import os.path +from typing import Any, Tuple + +from mllm.markov_games.markov_game import AgentAndActionSafeCopy, MarkovGame +from mllm.markov_games.rollout_tree import ( + AgentActLog, + RolloutTreeBranchNode, + RolloutTreeNode, + RolloutTreeRootNode, + StepLog, +) + +AgentId = str + + +async def run_with_unilateral_alt_action( + markov_game: MarkovGame, + agent_id: AgentId, + time_step: int, + branch_node: RolloutTreeBranchNode, + max_depth: int, +): + """ + Roll out a counterfactual branch where ``agent_id`` deviates unilaterally. + + Starting from ``branch_node`` (which already contains the main trajectory), + we replay the simulation with the deviating agent's action while freezing + all other agents/actions, then continue for ``max_depth`` steps. + """ + + # Generate alternative action and take a step + await markov_game.set_action_of_agent(agent_id) + terminated: bool = markov_game.take_simulation_step() + step_log = markov_game.get_step_log() + first_alternative_node = RolloutTreeNode( + step_log=step_log, + time_step=time_step, + ) + + # Generate rest of trajectory up to max depth + time_step += 1 + counter = 1 + previous_node = first_alternative_node + while not terminated and counter <= max_depth: + terminated, step_log = await markov_game.step() + current_node = RolloutTreeNode(step_log=step_log, time_step=time_step) + previous_node.child = current_node + previous_node = current_node + counter += 1 + time_step += 1 + + if branch_node.branches == None: + branch_node.branches = {agent_id: [first_alternative_node]} + else: + agent_branches = branch_node.branches.get(agent_id, []) + agent_branches.append(first_alternative_node) + branch_node.branches[agent_id] = agent_branches + + +async def AlternativeActionsRunner( + markov_game: MarkovGame, + output_folder: str, + nb_alternative_actions: int, + max_depth: int, + branch_only_on_new_round: bool = False, +): + """ + Generate a rollout tree containing the main path plus unilateral deviation branches. + + For each timestep we: + 1. Cache agent actions without side effects. + 2. Advance the main trajectory. + 3. Spawn ``nb_alternative_actions`` asynchronous deviations per agent, + each replaying up to ``max_depth`` steps from the cached pre-action state. + The resulting branches feed advantage-alignment estimators. + """ + + tasks = [] + time_step = 0 + terminated = False + root = RolloutTreeRootNode(id=markov_game.get_id(), crn_id=markov_game.get_crn_id()) + previous_node = root + + while not terminated: + mg_before_action = markov_game.get_safe_copy() + + # Get safe copies for main branch + agent_action_safe_copies: dict[ + AgentId, AgentAndActionSafeCopy + ] = await markov_game.get_actions_of_agents_without_side_effects() + + markov_game.set_actions_of_agents_manually(agent_action_safe_copies) + terminated = markov_game.take_simulation_step() + main_node = RolloutTreeNode( + step_log=markov_game.get_step_log(), time_step=time_step + ) + branch_node = RolloutTreeBranchNode(main_child=main_node) + previous_node.child = branch_node + previous_node = main_node + + # Get alternative branches by generating new unilateral actions + for agent_id in markov_game.agent_ids: + for _ in range(nb_alternative_actions): + # Get safe copies for branches + branch_agent_action_safe_copies: dict[ + AgentId, AgentAndActionSafeCopy + ] = { + agent_id: AgentAndActionSafeCopy( + action=copy.deepcopy(agent_action_safe_copy.action), + action_info=copy.deepcopy(agent_action_safe_copy.action_info), + agent_after_action=agent_action_safe_copy.agent_after_action.get_safe_copy(), + ) + for agent_id, agent_action_safe_copy in agent_action_safe_copies.items() + } + mg_branch: MarkovGame = mg_before_action.get_safe_copy() + other_agent_id = [id for id in mg_branch.agent_ids if id != agent_id][0] + mg_branch.set_action_and_agent_after_action_manually( + agent_id=other_agent_id, + agent_action_safe_copy=branch_agent_action_safe_copies[ + other_agent_id + ], + ) + task = asyncio.create_task( + run_with_unilateral_alt_action( + markov_game=mg_branch, + time_step=time_step, + agent_id=agent_id, + branch_node=branch_node, + max_depth=max_depth, + ) + ) + tasks.append(task) + time_step += 1 + + # wait for all branches to complete + await asyncio.gather(*tasks) + + return root diff --git a/src_code_for_reproducibility/markov_games/group_timesteps.py b/src_code_for_reproducibility/markov_games/group_timesteps.py new file mode 100644 index 0000000000000000000000000000000000000000..48b5882a632ba858787befaac306195af959b376 --- /dev/null +++ b/src_code_for_reproducibility/markov_games/group_timesteps.py @@ -0,0 +1,133 @@ +""" +File: mllm/markov_games/group_timesteps.py +Summary: Provides timestep-grouping utilities for rollout trees and training. +""" + +import copy +from typing import Callable + +from mllm.markov_games.markov_game import MarkovGame +from mllm.markov_games.rollout_tree import ( + AgentActLog, + RolloutTreeBranchNode, + RolloutTreeNode, + RolloutTreeRootNode, + StepLog, +) +from mllm.markov_games.simulation import SimulationStepLog + +AgentId = str + + +def group_time_steps( + rollout_tree: RolloutTreeRootNode, + accumulation_stop_condition: Callable[[StepLog], bool], +) -> RolloutTreeRootNode: + """ + During generation, we create rollout trees according to the real time steps. + However, during training, we might want to treat groups of time steps as a single time step. + As a concrete example, take Trust-and-Split. At each round, say we have X time steps of communication and then one time step for the split. + Then the communication actions will not get any reward, and the split action will get the reward. During REINFORCE training, with discounting, this + can cause training instability. We could instead treat every action in the round as being part of a single action, and give it the reward of the split action. + This method helps to do this sort of grouping. + It accumulates actions until the accumulation_stop_condition is met, and then creates a new node with the accumulated actions. + It then recursively calls itself on the child node. + Details: + - The reward for the group is the reward of the last time step in the group. + - The simulation log for the group is the simulation log of the last time step in the group. + - The state end for the group becomes the first state end in the group. + - The agent info for the group is the agent info of the last time step in the group. + """ + + def group_step_logs(step_logs: list[StepLog]) -> StepLog: + """ + Concatenate per-agent chat turns across steps; keep only the first is_state_end. + """ + last_sim_log = step_logs[-1].simulation_step_log + agent_ids = {aid for s in step_logs for aid in s.action_logs.keys()} + grouped_logs: dict[AgentId, AgentActLog] = {} + for aid in agent_ids: + turns = [] + for s in step_logs: + act = s.action_logs.get(aid) + if act and act.chat_turns: + turns.extend(copy.deepcopy(act.chat_turns)) + disable_is_state_end = False + # Only the first state_end should be True, the rest should be False + for t in turns: + if t.is_state_end: + if disable_is_state_end: + t.is_state_end = False + else: + disable_is_state_end = True + continue + grouped_logs[aid] = AgentActLog( + chat_turns=turns, info=step_logs[-1].action_logs[aid].info + ) + return StepLog(action_logs=grouped_logs, simulation_step_log=last_sim_log) + + def group_time_steps_rec( + current_node: RolloutTreeNode | RolloutTreeBranchNode, + group_time_step: int, + accumulation_step_logs: list[StepLog], + ) -> RolloutTreeNode | RolloutTreeBranchNode: + """ + Groups time steps. Recursion is used to handle branches. + """ + assert isinstance(current_node, RolloutTreeNode) or isinstance( + current_node, RolloutTreeBranchNode + ), "Current node must be a tree node or a branch node. Is of type: " + str( + type(current_node) + ) + first_group_node = None + current_group_node = None + while current_node is not None: + if isinstance(current_node, RolloutTreeBranchNode): + raise Exception( + "Grouping timesteps by round is not supported for branching trajectories yet." + ) + + # Accumulate + accumulation_step_logs.append(current_node.step_log) + if accumulation_stop_condition(current_node.step_log): + grouped_step_logs = group_step_logs(accumulation_step_logs) + accumulation_step_logs = [] + new_group_node = RolloutTreeNode( + step_log=grouped_step_logs, time_step=group_time_step, child=None + ) + if first_group_node == None: + first_group_node = new_group_node + group_time_step += 1 + if current_group_node is not None: + current_group_node.child = new_group_node + current_group_node = new_group_node + current_node = current_node.child + return first_group_node + + node = group_time_steps_rec( + current_node=rollout_tree.child, group_time_step=0, accumulation_step_logs=[] + ) + return RolloutTreeRootNode( + id=rollout_tree.id, + crn_id=rollout_tree.crn_id, + child=node, + agent_ids=rollout_tree.agent_ids, + ) + + +def stop_when_round_ends(step_log: StepLog) -> bool: + """ + Simplest stop condition. Will return True if step log is the last time step of a round. + This will throw an error if this information is not available in the simulation info. + """ + assert ( + "is_last_timestep_in_round" in step_log.simulation_step_log.info.keys() + ), "To group by round, is_last_timestep_in_round must be set in the info of your simulation step log at each time step." + return step_log.simulation_step_log.info["is_last_timestep_in_round"] + + +def group_by_round(rollout_tree: RolloutTreeRootNode) -> RolloutTreeRootNode: + """ + Groups time steps by round. + """ + return group_time_steps(rollout_tree, stop_when_round_ends) diff --git a/src_code_for_reproducibility/markov_games/linear_runner.py b/src_code_for_reproducibility/markov_games/linear_runner.py new file mode 100644 index 0000000000000000000000000000000000000000..e3e14a3eda72cf4e620db5ab8ed0d3f7d552e9fe --- /dev/null +++ b/src_code_for_reproducibility/markov_games/linear_runner.py @@ -0,0 +1,42 @@ +""" +File: mllm/markov_games/linear_runner.py +Summary: Simulates a single unbranched Markov-game rollout and records it. +""" + +import asyncio +import json +import os.path + +from mllm.markov_games.markov_game import MarkovGame +from mllm.markov_games.rollout_tree import RolloutTreeNode, RolloutTreeRootNode + + +async def LinearRunner( + markov_game: MarkovGame, output_folder: str +) -> RolloutTreeRootNode: + """ + Generate a single main-path rollout (no branching) for the provided Markov game. + + Parameters + ---------- + markov_game: + Initialized ``MarkovGame`` with agents + simulation ready to step. + output_folder: + Unused placeholder in the legacy API (kept for compatibility). + """ + time_step = 0 + terminated = False + root = RolloutTreeRootNode( + id=markov_game.get_id(), + crn_id=markov_game.get_crn_id(), + agent_ids=markov_game.get_agent_ids(), + ) + previous_node = root + while not terminated: + terminated, step_log = await markov_game.step() + current_node = RolloutTreeNode(step_log=step_log, time_step=time_step) + previous_node.child = current_node + previous_node = current_node + time_step += 1 + + return root diff --git a/src_code_for_reproducibility/markov_games/markov_game.py b/src_code_for_reproducibility/markov_games/markov_game.py new file mode 100644 index 0000000000000000000000000000000000000000..7964fd69d24f617c76e36f852491b1e6141f6c48 --- /dev/null +++ b/src_code_for_reproducibility/markov_games/markov_game.py @@ -0,0 +1,217 @@ +""" +File: mllm/markov_games/markov_game.py +Summary: Defines the MarkovGame base class plus shared simulation interfaces. +""" + +import asyncio +import copy +import json +import os +from dataclasses import dataclass +from typing import Any, List, Literal, Optional, Tuple + +from transformers.models.idefics2 import Idefics2Config + +from mllm.markov_games.agent import Agent +from mllm.markov_games.rollout_tree import AgentActLog, StepLog +from mllm.markov_games.simulation import Simulation + +AgentId = str + + +@dataclass +class AgentAndActionSafeCopy: + """Snapshot of an agent, its action, and metadata used for branch replay.""" + + action: Any + action_info: AgentActLog + agent_after_action: type[Agent] + + +class MarkovGame(object): + def __init__( + self, + id: int, + agents: dict[AgentId, type[Agent]], + simulation: type[Simulation], + crn_id: int, + ): + """ + Initialize the Markov game wrapper. + + Parameters + ---------- + id: + Unique rollout identifier (logged into rollout trees). + agents: + Mapping of agent_id -> Agent instance. + simulation: + Environment implementing the ``Simulation`` interface (IPD, TAS, etc.). + crn_id: + Identifier for the common random number stream used by this rollout. + """ + self.agents = agents + self.agent_ids = self.agents.keys() + self.simulation = simulation + self.simulation_step_log = None + self.agent_step_logs = {agent_id: None for agent_id in self.agent_ids} + self.actions = {} + self.id = id + self.crn_id = crn_id + + def get_id(self) -> str: + return self.id + + def get_crn_id(self) -> int: + return self.crn_id + + def get_agent_ids(self) -> List[AgentId]: + return list(self.agent_ids) + + async def get_action_of_agent_without_side_effects( + self, agent_id: AgentId + ) -> Tuple[Any, AgentActLog]: + """ + Safe function to get an action of an agent without modifying the agent or the simulation. + """ + agent = self.agents[agent_id] + agent_before_action = agent.get_safe_copy() + obs = self.simulation.get_obs_agent(agent_id) + action, action_info = await agent.act(observation=obs) + self.agents[agent_id] = agent_before_action + agent_after_action = agent.get_safe_copy() + return AgentAndActionSafeCopy(action, action_info, agent_after_action) + + async def get_actions_of_agents_without_side_effects( + self, + ) -> dict[AgentId, AgentAndActionSafeCopy]: + """ + Safe function to get an action of an agent without modifying the agent or the simulation. + """ + tasks = [] + for agent_id in self.agent_ids: + task = asyncio.create_task( + self.get_action_of_agent_without_side_effects(agent_id) + ) + tasks.append(task) + agent_and_action_safe_copies: list[ + AgentAndActionSafeCopy + ] = await asyncio.gather(*tasks) + return { + agent_id: agent_and_action_safe_copy + for agent_id, agent_and_action_safe_copy in zip( + self.agent_ids, agent_and_action_safe_copies + ) + } + + def set_action_and_agent_after_action_manually( + self, + agent_id: AgentId, + agent_action_safe_copy: AgentAndActionSafeCopy, + ): + """ + Set the action and the agent after action manually. + """ + self.actions[agent_id] = agent_action_safe_copy.action + self.agent_step_logs[agent_id] = agent_action_safe_copy.action_info + self.agents[agent_id] = agent_action_safe_copy.agent_after_action + + def set_actions_of_agents_manually( + self, actions: dict[AgentId, AgentAndActionSafeCopy] + ): + """ + Set the actions of agents manually. + """ + for agent_id, agent_action_safe_copy in actions.items(): + self.set_action_and_agent_after_action_manually( + agent_id, agent_action_safe_copy + ) + + async def set_action_of_agent(self, agent_id: AgentId): + """ + Query a single agent for its next action and store the result locally. + """ + agent = self.agents[agent_id] + obs = self.simulation.get_obs_agent(agent_id) + action, action_info = await agent.act(observation=obs) + self.actions[agent_id] = action + self.agent_step_logs[agent_id] = action_info + + async def set_actions(self): + """ + Query every agent concurrently and populate the cached actions/logs. + """ + # background_tasks = set() + tasks = [] + for agent_id in self.agent_ids: + task = asyncio.create_task(self.set_action_of_agent(agent_id)) + tasks.append(task) + await asyncio.gather(*tasks) + + def take_simulation_step(self): + """ + Advance the simulation by one step using the cached actions. + """ + terminated, self.simulation_step_log = self.simulation.step(self.actions) + return terminated + + def get_step_log(self) -> StepLog: + """ + Package the most recent simulation step and agent logs into a StepLog. + """ + if self.simulation_step_log is None: + raise RuntimeError( + "Simulation step log is empty; call take_simulation_step() first." + ) + missing_logs = [ + agent_id for agent_id, log in self.agent_step_logs.items() if log is None + ] + if missing_logs: + raise RuntimeError( + f"Agent action logs missing for: {', '.join(missing_logs)}. " + "Ensure set_actions() ran before requesting the step log." + ) + step_log = StepLog( + simulation_step_log=self.simulation_step_log, + action_logs=self.agent_step_logs, + ) + return step_log + + async def step(self) -> Tuple[bool, StepLog]: + """ + Convenience step that collects actions, advances the simulation, and returns the log. + """ + await self.set_actions() + terminated = self.take_simulation_step() + step_log = self.get_step_log() + return terminated, step_log + + def get_safe_copy(self): + """ + Create a shallow copy of the game with deep-copied agents/simulation for branching. + """ + + new_markov_game = copy.copy(self) + new_simulation = self.simulation.get_safe_copy() + new_agents = { + agent_id: agent.get_safe_copy() for agent_id, agent in self.agents.items() + } + + # Reassign copied components + new_markov_game.simulation = new_simulation + new_markov_game.agents = new_agents + + # IMPORTANT: ensure agent_ids references the new agents dict, not the original + new_markov_game.agent_ids = new_markov_game.agents.keys() + + # Deep-copy step data to avoid correlation + new_markov_game.simulation_step_log = copy.deepcopy(self.simulation_step_log) + new_markov_game.actions = copy.deepcopy(self.actions) + # Rebuild logs to align exactly with new agent ids + old_agent_step_logs = copy.deepcopy(self.agent_step_logs) + new_markov_game.agent_step_logs = { + agent_id: old_agent_step_logs.get(agent_id) + for agent_id in new_markov_game.agent_ids + } + + return new_markov_game diff --git a/src_code_for_reproducibility/markov_games/mg_utils.py b/src_code_for_reproducibility/markov_games/mg_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..4fc406cd1f0cba593daad1108de2746b6a1d7678 --- /dev/null +++ b/src_code_for_reproducibility/markov_games/mg_utils.py @@ -0,0 +1,97 @@ +""" +File: mllm/markov_games/mg_utils.py +Summary: Holds miscellaneous helpers shared across Markov-game modules. +""" + +import asyncio +import copy +from collections.abc import Callable +from dataclasses import dataclass + +from mllm.markov_games.ipd.ipd_agent import IPDAgent +from mllm.markov_games.ipd.Ipd_hard_coded_agents import ( + AlwaysCooperateIPDAgent, + AlwaysDefectIPDAgent, +) +from mllm.markov_games.ipd.ipd_simulation import IPD +from mllm.markov_games.markov_game import MarkovGame +from mllm.markov_games.negotiation.dond_agent import DealNoDealAgent +from mllm.markov_games.negotiation.dond_simulation import DealNoDealSimulation +from mllm.markov_games.negotiation.nego_hard_coded_policies import ( + HardCodedNegoGreedyPolicy, + HardCodedNegoWelfareMaximizingPolicy, +) +from mllm.markov_games.negotiation.no_press_nego_agent import NoPressAgent +from mllm.markov_games.negotiation.no_press_nego_simulation import NoPressSimulation +from mllm.markov_games.negotiation.tas_rps_agent import TrustAndSplitRPSAgent +from mllm.markov_games.negotiation.tas_rps_simulation import TrustAndSplitRPSSimulation +from mllm.markov_games.rollout_tree import ( + AgentActLog, + RolloutTreeBranchNode, + RolloutTreeNode, + RolloutTreeRootNode, + StepLog, +) +from mllm.markov_games.simulation import SimulationStepLog + +AgentId = str + + +@dataclass +class AgentConfig: + """Configuration blob describing one agent in a Markov game spec.""" + + agent_id: str + agent_name: str + agent_class_name: str + policy_id: str + init_kwargs: dict + + +@dataclass +class MarkovGameConfig: + """Top-level config that ties together simulation settings and agent configs.""" + + id: int + seed: int + simulation_class_name: str + simulation_init_args: dict + agent_configs: list[AgentConfig] + + +def init_markov_game_components( + config: MarkovGameConfig, policies: dict[str, Callable[[list[dict]], str]] +): + """ + Materialize Agents and the Simulation described by ``config`` and return a MarkovGame. + + `policies` is a mapping of policy_id -> callable retrieved from the hosting trainer. + """ + agents = {} + agent_names = [] + for agent_config in config.agent_configs: + agent_id = agent_config.agent_id + agent_name = agent_config.agent_name + agent_class = eval(agent_config.agent_class_name) + agent = agent_class( + seed=config.seed, + agent_id=agent_id, + agent_name=agent_name, + policy=policies[agent_config.policy_id], + **agent_config.init_kwargs, + ) + agents[agent_id] = agent + agent_names.append(agent_name) + simulation = eval(config.simulation_class_name)( + seed=config.seed, + agent_ids=list(agents.keys()), + agent_names=agent_names, + **config.simulation_init_args, + ) + markov_game = MarkovGame( + id=config.id, + crn_id=config.seed, + agents=agents, + simulation=simulation, + ) + return markov_game diff --git a/src_code_for_reproducibility/markov_games/negotiation/dond_simulation.py b/src_code_for_reproducibility/markov_games/negotiation/dond_simulation.py new file mode 100644 index 0000000000000000000000000000000000000000..d941fff41dd68d0d675dcffc31720ba20a7efb4f --- /dev/null +++ b/src_code_for_reproducibility/markov_games/negotiation/dond_simulation.py @@ -0,0 +1,176 @@ +""" +File: mllm/markov_games/negotiation/dond_simulation.py +Summary: Simulates Deal-or-No-Deal negotiation games and logs rollouts. +""" + +import copy +from dataclasses import dataclass +from typing import Any, Dict, List, Tuple + +from numpy.random import default_rng + +from mllm.markov_games.negotiation.nego_simulation import ( + NegotiationObs, + NegotiationSimulation, + NegotiationState, + Split, +) +from mllm.markov_games.rollout_tree import SimulationStepLog +from mllm.utils.get_coagent_id import get_coagent_id + +AgentId = str + + +@dataclass +class DealNoDealState(NegotiationState): + """NegotiationState with per-agent value tables and item taxonomy.""" + + item_types: List[str] + values: Dict[AgentId, Dict[str, int]] + + +@dataclass +class DealNoDealObs(NegotiationObs): + """Observation that reveals own values and (lagged) opponent values.""" + + my_values: Dict[str, int] + item_types: List[str] + previous_values_coagent: Dict[str, int] | None + + +def random_partition_integer(rng, total: int, parts: int) -> List[int]: + """Sample non-negative integers summing to ``total`` across ``parts`` buckets.""" + if parts <= 0: + return [] + if total <= 0: + return [0 for _ in range(parts)] + cuts = sorted(rng.integers(0, total + 1, size=parts - 1).tolist()) + vals = [] + prev = 0 + for c in cuts + [total]: + vals.append(c - prev) + prev = c + return vals + + +class DealNoDealSimulation(NegotiationSimulation): + """NegotiationSimulation variant implementing the Rubinstein-style Deal-or-No-Deal.""" + + def __init__( + self, + item_types: List[str] = ["books", "hats", "balls"], + *args, + **kwargs, + ): + super().__init__(item_types=item_types, *args, **kwargs) + self.reset() + + def _other(self, agent_id: AgentId) -> AgentId: + return get_coagent_id(self.agent_ids, agent_id) + + def _sample_stock(self) -> Dict[str, int]: + # total items between 5 and 7 + total_items = int(self.rng.integers(5, 8)) + # nonnegative per-type counts summing to total_items + parts = random_partition_integer(self.rng, total_items, len(self.item_types)) + # allow zeros per type + return {t: int(c) for t, c in zip(self.item_types, parts)} + + def _sample_values_pair(self) -> Dict[AgentId, Dict[str, int]]: + # Each agent has integer non-negative values that sum to 10 + # Each item type valued by at least one agent + # Some item type valued by both agents + while True: + vals_a = random_partition_integer(self.rng, 10, len(self.item_types)) + vals_b = random_partition_integer(self.rng, 10, len(self.item_types)) + a = {t: int(v) for t, v in zip(self.item_types, vals_a)} + b = {t: int(v) for t, v in zip(self.item_types, vals_b)} + # each item valued by at least one + ok1 = all((a[t] > 0) or (b[t] > 0) for t in self.item_types) + # some item valued by both + ok2 = any((a[t] > 0) and (b[t] > 0) for t in self.item_types) + if ok1 and ok2: + return {self.agent_ids[0]: a, self.agent_ids[1]: b} + + def _is_valid_allocation( + self, allocation: Dict[str, int], stock: Dict[str, int] + ) -> bool: + for t in self.item_types: + v = allocation.get(t) + if v is None: + return False + if not isinstance(v, int): + return False + if v < 0 or v > int(stock.get(t, 0)): + return False + return True + + def set_new_round_of_variant(self): + # Keep same values, resample stock + self.state.quantities = self._sample_stock() + + def get_info_of_variant( + self, state: NegotiationState, actions: Dict[AgentId, Any] + ) -> Dict[str, Any]: + return { + "quantities": copy.deepcopy(state.quantities), + "values": copy.deepcopy(state.values), + "splits": copy.deepcopy(state.splits), + } + + def get_rewards(self, splits: Dict[AgentId, Split]) -> Dict[AgentId, float]: + """ + Returns the rewards for each agent. + """ + split_a = splits[self.agent_ids[0]].items_given_to_self + split_b = splits[self.agent_ids[1]].items_given_to_self + rewards = {self.agent_ids[0]: 0, self.agent_ids[1]: 0} + for t in self.item_types: + # If not complementary, return 0! + if not split_a[t] + split_b[t] == self.state.quantities[t]: + return {self.agent_ids[0]: 0, self.agent_ids[1]: 0} + rewards[self.agent_ids[0]] += ( + split_a[t] * self.state.values[self.agent_ids[0]][t] + ) + rewards[self.agent_ids[1]] += ( + split_b[t] * self.state.values[self.agent_ids[1]][t] + ) + return rewards + + def get_obs(self): + return {agent_id: self.get_obs_agent(agent_id) for agent_id in self.agent_ids} + + def get_obs_agent(self, agent_id): + other_id = self._other(agent_id) + obs = DealNoDealObs( + round_nb=self.state.round_nb, + last_message=self.state.last_message, + current_agent=self.state.current_agent, + quantities=copy.deepcopy(self.state.quantities), + value=0.0, # unused in DOND + other_agent_split=None, # not meaningful until split + split_phase=self.state.split_phase, + quota_messages_per_agent_per_round=self.quota_messages_per_agent_per_round, + my_values=copy.deepcopy(self.state.values[agent_id]), + item_types=list(self.item_types), + previous_values_coagent=copy.deepcopy(self.state.values.get(other_id, {})), + ) + return obs + + def reset(self): + start_agent = self.agent_ids[self._starting_agent_index] + stock = self._sample_stock() + values = self._sample_values_pair() + self.state = DealNoDealState( + round_nb=0, + last_message="", + current_agent=start_agent, + quantities=stock, + values=values, + previous_values=None, + splits={aid: None for aid in self.agent_ids}, + nb_messages_sent={aid: 0 for aid in self.agent_ids}, + split_phase=False, + item_types=list(self.item_types), + ) + return self.get_obs() diff --git a/src_code_for_reproducibility/markov_games/negotiation/nego_agent.py b/src_code_for_reproducibility/markov_games/negotiation/nego_agent.py new file mode 100644 index 0000000000000000000000000000000000000000..af30f90968c7d3a5739700d1d2ba6c906194573f --- /dev/null +++ b/src_code_for_reproducibility/markov_games/negotiation/nego_agent.py @@ -0,0 +1,261 @@ +""" +File: mllm/markov_games/negotiation/nego_agent.py +Summary: General-purpose negotiation agent coordinating prompts and actions. +""" + +import copy +from abc import abstractmethod +from collections.abc import Callable +from dataclasses import dataclass +from typing import Any, Dict, List, Tuple + +import numpy as np + +from mllm.markov_games.agent import Agent +from mllm.markov_games.negotiation.nego_simulation import Message, NegotiationObs, Split +from mllm.markov_games.rollout_tree import AgentActLog, ChatTurn + + +@dataclass +class NegotiationAgentState: + """Lightweight container tracking round progression and message history.""" + + round_nb: int + nb_messages_sent_this_round: int + chat_counter: int + chat_history: List[ChatTurn] + + +class NegotiationAgent(Agent): + """Base agent that manages prompt scaffolding and regex validation for variants.""" + + def __init__( + self, + seed: int, + agent_id: str, + agent_name: str, + policy: Callable[[List[Dict]], str], + goal: str, + exploration_prompts: List[str] = [], + exploration_prompt_probs: List[float] = [], + ): + self.seed = seed + self.agent_id = agent_id + self.agent_name = agent_name + self.policy = policy + self.goal = goal + self.exploration_prompts_toggled = len(exploration_prompts) > 0 + if self.exploration_prompts_toggled: + exploration_prompts = copy.deepcopy(exploration_prompts) + exploration_prompts.append(None) + self.exploration_prompts = exploration_prompts + self.exploration_prompt_probs = np.array(exploration_prompt_probs) + assert self.exploration_prompt_probs.sum() <= 1 + assert np.all(self.exploration_prompt_probs >= 0) + self.exploration_prompt_probs = np.append( + self.exploration_prompt_probs, 1 - self.exploration_prompt_probs.sum() + ) + self.state = NegotiationAgentState( + round_nb=0, nb_messages_sent_this_round=0, chat_counter=0, chat_history=[] + ) + + # Implemented in variants + self.intro_prompt = "" + self.new_round_prompt = "" + self.last_round_prompt = "" + self.send_split_prompt = "" + self.wait_for_message_prompt = "" + self.last_message_prompt = "" + self.send_message_prompt = "" + + @abstractmethod + def get_message_regex(self, observation: NegotiationObs) -> str: + """Return the regex that outgoing chat messages must satisfy.""" + pass + + @abstractmethod + def get_split_regex(self, observation: NegotiationObs) -> str: + """Return the regex that final split proposals must satisfy.""" + pass + + @abstractmethod + def get_split_action( + self, policy_output: str, observation: NegotiationObs + ) -> Split: + """Convert raw LLM output into the ``Split`` structure required by simulations.""" + pass + + async def act(self, observation: NegotiationObs) -> Tuple[Any, AgentActLog]: + """ + Assemble the appropriate prompt, query the policy, and return message or split. + + This handles intro text, new-round reminders, quota tracking, and post-processing + (regex enforcement + ChatTurn logging) so subclasses only customize prompts/regexes. + """ + + def dict_to_str(d: dict) -> str: + return ", ".join(f"{v} {k}" for k, v in d.items()) + + def dict_to_eq_str(d: dict) -> str: + return ", ".join(f"{k}={v}" for k, v in d.items()) + + is_our_turn = observation.current_agent == self.agent_id + action: Any = None + round_nb = observation.round_nb + + prompt_parts: List[str] = [] + obs_ctx = vars(observation) + obs_ctx_formmated = obs_ctx.copy() + for key in obs_ctx_formmated: + if isinstance(obs_ctx_formmated[key], dict) and "value" not in key: + obs_ctx_formmated[key] = dict_to_str(obs_ctx_formmated[key]) + elif isinstance(obs_ctx_formmated[key], dict) and "value" in key: + obs_ctx_formmated[key] = dict_to_eq_str(obs_ctx_formmated[key]) + + ####################################### + # build user prompt + ####################################### + + # First-ever call + is_intro = round_nb == 0 and self.state.chat_counter == 0 + if is_intro: + prompt_parts.append( + self.intro_prompt.format( + goal=self.goal, agent=self.agent_name, **obs_ctx_formmated + ) + ) + + # New round + is_new_round = round_nb > self.state.round_nb + if is_new_round or is_intro: + self.state.nb_messages_sent_this_round = 0 + if not is_intro: + prompt_parts.append(self.last_round_prompt.format(**obs_ctx_formmated)) + prompt_parts.append(self.new_round_prompt.format(**obs_ctx_formmated)) + if self.exploration_prompts_toggled: + exploration_prompt = self.exploration_prompts[ + np.random.choice( + len(self.exploration_prompts), p=self.exploration_prompt_probs + ) + ] + if exploration_prompt is not None: + prompt_parts.append(exploration_prompt) + self.state.round_nb = round_nb + + # Wait for message + if not is_our_turn and not observation.split_phase: + prompt_parts.append( + self.wait_for_message_prompt.format(**obs_ctx_formmated) + ) + + # Get last message + if is_our_turn and not is_new_round and not is_intro: + prompt_parts.append(self.last_message_prompt.format(**obs_ctx_formmated)) + + # Prompt to send message + must_send_message = not observation.split_phase and is_our_turn + if must_send_message: + prompt_parts.append(self.send_message_prompt.format(**obs_ctx_formmated)) + + # Prompt to give split + must_send_split = not must_send_message and observation.split_phase + if must_send_split: + var_names = ["x", "y", "z", "w"] # Extend as needed + items_str = ", ".join( + [ + f"{var_names[i]} {item}" + for i, item in enumerate(obs_ctx["quantities"].keys()) + ] + ) + ranges_str = ", ".join( + [ + f"{var_names[i]}: 0-{obs_ctx['quantities'][item]} (integer)" + for i, item in enumerate(obs_ctx["quantities"].keys()) + ] + ) + proposal_style = f"Proposal: {items_str} where {ranges_str}." + proposal_style2 = ( + f" {items_str} where {ranges_str}." + ) + prompt_parts.append( + self.send_split_prompt.format( + proposal_style=proposal_style, + proposal_style2=proposal_style2, + **obs_ctx_formmated, + ) + ) + + # Append one ChatTurn with is_state_end=True + user_prompt = "\n".join(prompt_parts) + self.state.chat_history.append( + ChatTurn( + agent_id=self.agent_id, + role="user", + content=user_prompt, + is_state_end=True, + ) + ) + + ####################################### + # Get policy action + ####################################### + + # Query policy for the appropriate format + if must_send_message: + return_regex = self.get_message_regex(observation) + policy_output = await self.policy( + state=self.state.chat_history, + agent_id=self.agent_id, + regex=return_regex, + ) + self.state.chat_history.append( + ChatTurn( + agent_id=self.agent_id, + role="assistant", + content=policy_output.content, + reasoning_content=policy_output.reasoning_content, + log_probs=policy_output.log_probs, + out_token_ids=policy_output.out_token_ids, + is_state_end=False, + ) + ) + action = Message(message=policy_output.content) + self.state.nb_messages_sent_this_round += 1 + + elif must_send_split: + return_regex = self.get_split_regex(observation) + policy_output = await self.policy( + state=self.state.chat_history, + agent_id=self.agent_id, + regex=return_regex, + ) + self.state.chat_history.append( + ChatTurn( + agent_id=self.agent_id, + role="assistant", + content=policy_output.content, + reasoning_content=policy_output.reasoning_content, + log_probs=policy_output.log_probs, + out_token_ids=policy_output.out_token_ids, + is_state_end=False, + ) + ) + action = self.get_split_action(policy_output.content, observation) + else: + action = None + + agent_step_log = AgentActLog( + chat_turns=self.state.chat_history[self.state.chat_counter :], info=None + ) + self.state.chat_counter = len(self.state.chat_history) + return action, agent_step_log + + def get_safe_copy(self): + agent_copy = copy.copy(self) + agent_copy.state = copy.deepcopy(self.state) + return agent_copy + + def reset(self): + self.state = NegotiationAgentState( + round_nb=0, nb_messages_sent_this_round=0, chat_counter=0, chat_history=[] + ) diff --git a/src_code_for_reproducibility/markov_games/negotiation/nego_simulation.py b/src_code_for_reproducibility/markov_games/negotiation/nego_simulation.py new file mode 100644 index 0000000000000000000000000000000000000000..2172aa345aba163ed8d84b477f54c3fbde5ce249 --- /dev/null +++ b/src_code_for_reproducibility/markov_games/negotiation/nego_simulation.py @@ -0,0 +1,252 @@ +""" +File: mllm/markov_games/negotiation/nego_simulation.py +Summary: Simulation harness for general negotiation environments. +""" + +import copy +from abc import abstractmethod +from dataclasses import dataclass +from typing import Any, Dict, List, Tuple + +from numpy.random import default_rng + +from mllm.markov_games.rollout_tree import SimulationStepLog +from mllm.markov_games.simulation import Simulation +from mllm.utils.get_coagent_id import get_coagent_id + +AgentId = str + + +@dataclass +class Split: + """Structured proposal describing how many units of each item an agent keeps.""" + + items_given_to_self: Dict[str, int] + + +@dataclass +class Message: + """Single chat utterance exchanged during the negotiation phase.""" + + message: str + + +@dataclass # gets extended by variants +class NegotiationState: + """Full simulator state snapshot shared by all negotiation variants.""" + + round_nb: int + last_message: str + current_agent: AgentId + quantities: Dict[str, int] + values: Dict[AgentId, Dict[str, float]] + splits: Dict[AgentId, Split | None] + nb_messages_sent: Dict[AgentId, int] + previous_values: Dict[AgentId, Dict[str, float]] | None + previous_splits: Dict[AgentId, Dict[str, int] | None] | None + previous_points: Dict[AgentId, float] | None + previous_quantities: Dict[str, int] | None + split_phase: bool + + +@dataclass # gets extended by variants +class NegotiationObs: + """Observation presented to agents each turn (base fields; variants extend).""" + + round_nb: int + last_message: str + quota_messages_per_agent_per_round: int + current_agent: AgentId + other_agent: str + quantities: Dict[str, int] + item_types: List[str] + value: Dict[str, int] + split_phase: bool + last_split_agent: Dict[str, int] | None + last_value_agent: Dict[str, int] | None + last_points_agent: float | None + last_split_coagent: Dict[str, int] | None + last_value_coagent: Dict[str, int] | None + last_points_coagent: float | None + last_quantities: Dict[str, int] | None + + +def compute_tas_style_rewards( + agent_ids: List[AgentId], + values: Dict[AgentId, float], + splits: Dict[AgentId, Split], + quantities: Dict[str, int], +) -> Dict[AgentId, float]: + """ + TAS-like reward computation: if sum of proposed coins exceeds max_coins, + allocate proportionally. Otherwise, use proposed amounts directly. + Rewards are quantity_kept * per-coin value for each agent. + """ + a0, a1 = agent_ids[0], agent_ids[1] + r0, r1 = 0.0, 0.0 + + for item in quantities: + max_item = quantities[item] + item_to_self_0 = int( + (splits[a0].items_given_to_self.get(item, 0)) + if splits[a0] is not None + else 0 + ) + item_to_self_1 = int( + (splits[a1].items_given_to_self.get(item, 0)) + if splits[a1] is not None + else 0 + ) + denom = max(int(max_item), item_to_self_0 + item_to_self_1) + q0 = float(max_item) * float(item_to_self_0) / float(denom) + q1 = float(max_item) * float(item_to_self_1) / float(denom) + if type(values[a0]) is not dict: + r0 += q0 * float(values[a0]) + r1 += q1 * float(values[a1]) + else: + r0 += q0 * float(values[a0][item]) + r1 += q1 * float(values[a1][item]) + return {a0: r0, a1: r1} + + +class NegotiationSimulation(Simulation): + def __init__( + self, + agent_ids: List[AgentId], + agent_names: List[str], + seed: int, + nb_of_rounds: int, + quota_messages_per_agent_per_round: int, + item_types: List[str] | None = None, + ): + self.seed = seed + self.rng = default_rng(self.seed) + self.agent_ids = list(agent_ids) + self.agent_names = agent_names + self.agent_id_to_name = { + agent_id: agent_name for agent_id, agent_name in zip(agent_ids, agent_names) + } + self.nb_of_rounds = int(nb_of_rounds) + self.quota_messages_per_agent_per_round = int( + quota_messages_per_agent_per_round + ) + if item_types is not None: + self.item_types = [item.lower() for item in item_types] + else: + self.item_types = ["coins"] + self.state: NegotiationState | None = None + self._starting_agent_index = self.rng.choice([0, 1]) + self.reset() + + def _other(self, agent_id: AgentId) -> AgentId: + return get_coagent_id(self.agent_ids, agent_id) + + @abstractmethod + def set_new_round_of_variant(self): + """Variant hook: sample new private values / stock before each round.""" + pass + + @abstractmethod + def get_info_of_variant( + self, state: NegotiationState, actions: Dict[AgentId, Any] + ) -> Dict[str, Any]: + """Variant hook: populate SimulationStepLog.info with custom diagnostics.""" + pass + + def step(self, actions: Any) -> Tuple[bool, SimulationStepLog]: + """ + Returns terminated, step_log + """ + assert self.state is not None + current_agent = self.state.current_agent + a0, a1 = self.agent_ids[0], self.agent_ids[1] + action = actions.get(current_agent) + + # Split phase: require both splits in the same timestep + if self.state.split_phase: + action_a0 = actions.get(a0) + action_a1 = actions.get(a1) + have_both_splits = isinstance(action_a0, Split) and isinstance( + action_a1, Split + ) + if not have_both_splits: + rewards = {agent_id: 0.0 for agent_id in self.agent_ids} + return False, SimulationStepLog( + rewards=rewards, info={"type": "waiting_for_splits"} + ) + + # Record splits + self.state.splits[a0] = action_a0 + self.state.splits[a1] = action_a1 + + # Compute rewards and end round + rewards = self.get_rewards(self.state.splits) + + # Info + info = self.get_info_of_variant(self.state, actions) + + # Prepare next round + # Alternate starting agent + self.state.round_nb += 1 + self._starting_agent_index = 1 - self._starting_agent_index + self.state.current_agent = self.agent_ids[self._starting_agent_index] + self.state.previous_values = copy.deepcopy(self.state.values) + self.state.previous_splits = copy.deepcopy(self.state.splits) + self.state.previous_quantities = copy.deepcopy(self.state.quantities) + self.state.previous_points = copy.deepcopy(rewards) + self.state.last_message = "" + self.set_new_round_of_variant() # variant specific + self.state.splits = {agent_id: None for agent_id in self.agent_ids} + self.state.nb_messages_sent = {agent_id: 0 for agent_id in self.agent_ids} + is_last_timestep_in_round = True + done = self.state.round_nb >= self.nb_of_rounds + + # Message phase: roll the conversation forward a single turn. + elif isinstance(action, Message): + self.state.last_message = action.message + self.state.nb_messages_sent[current_agent] += 1 + + # Move turn to other agent + self.state.current_agent = self._other(current_agent) + + # If both agents have reached their message quota, enter split phase + if all( + self.state.nb_messages_sent[agent_id] + >= self.quota_messages_per_agent_per_round + for agent_id in self.agent_ids + ): + self.state.split_phase = True + is_last_timestep_in_round = False + done = False + rewards = {agent_id: 0.0 for agent_id in self.agent_ids} + info = {"type": "message"} + + info[ + "is_last_timestep_in_round" + ] = is_last_timestep_in_round # Used later to group round timesteps if needed + return done, SimulationStepLog(rewards=rewards, info=info) + + def get_obs(self): + """Returns all agent observations in dict""" + return {agent_id: self.get_obs_agent(agent_id) for agent_id in self.agent_ids} + + @abstractmethod + def get_rewards(self, splits: Dict[AgentId, Split]) -> Dict[AgentId, float]: + pass + + @abstractmethod + def get_obs_agent(self, agent_id): + pass + + def get_state(self): + return self.state + + def get_safe_copy(self): + """Return a safe copy of the simulation.""" + simulation_copy = copy.copy(self) + simulation_copy.state = copy.deepcopy(self.state) + return simulation_copy + + @abstractmethod + def reset(self) -> dict[AgentId, NegotiationObs]: + pass diff --git a/src_code_for_reproducibility/markov_games/negotiation/negotiation_statistics.py b/src_code_for_reproducibility/markov_games/negotiation/negotiation_statistics.py new file mode 100644 index 0000000000000000000000000000000000000000..f0c49b6cd995c05fe7ff4ad2efc7d7082035ac2e --- /dev/null +++ b/src_code_for_reproducibility/markov_games/negotiation/negotiation_statistics.py @@ -0,0 +1,249 @@ +""" +File: mllm/markov_games/negotiation/negotiation_statistics.py +Summary: Aggregates and reports statistics for negotiation experiments. +""" + +from __future__ import annotations + +from typing import Callable, Dict, List, Tuple + +from mllm.markov_games.negotiation.nego_simulation import Split +from mllm.markov_games.rollout_tree import SimulationStepLog + + +def avg_reward(sl: SimulationStepLog) -> List[Tuple[str, float]]: + """Average (per-step) reward for each agent and overall. + + What it computes: + - Returns the raw reward for every (non-buffer) agent at the current + simulation step. + - Adds an aggregate key ``all_agents`` which is the simple arithmetic + mean across the agents present in ``sl.rewards``. + + Rationale / motivation: + Monitoring the reward stream at each step helps: + * Diagnose reward shaping issues (e.g., unintended negative drift). + * Provide a fairness snapshot (are rewards systematically skewed?). + * Supply a ubiquitous baseline metric used by other higher‑level + summaries (efficiency, surplus allocation, etc.). + + Return shape: + { agent_id: float, ..., "all_agents": float } + If any agent id contains the substring "buffer" we treat this step as + an implementation artifact (e.g., rollout buffer) and return ``None`` + to avoid polluting aggregates. + """ + for aid in sl.rewards.keys(): + if "buffer" in str(aid) and "live" not in str(aid): + return None + # One value per agent at each step + rewards_dict = {f"reward-{aid}": float(v) for aid, v in (sl.rewards or {}).items()} + return [(key, value) for key, value in rewards_dict.items() if value is not None] + + +def split_efficiency(sl: SimulationStepLog) -> List[Tuple[str, float]] | None: + """Final‑round allocation efficiency relative to an upper bound. + + What it computes (only on the last timestep of a negotiation round): + - Uses ``info['values']`` (per‑agent per‑item valuations) and + ``info['quantities']`` (available item counts) to form a greedy + *upper bound* on achievable total reward: allocate each unit of an + item to the single agent who values that item most. + - Compares the actually realized sum of rewards at that final + timestep to this constructed maximum. + - Emits a single scalar under key ``"all_agents"`` equal to + achieved / theoretical_max. + + Motivation: + Efficiency (a core welfare notion) distinguishes between coordination + failures (low efficiency) versus strategic distributional disputes + (high efficiency but uneven splits). Tracking this per round helps + evaluate whether models learn to identify and realize joint surplus. + + Notes / caveats: + - Only defined for 2+ non‑buffer agents; if a buffer agent is present + returns ``None`` to exclude spurious steps. + - Requires the environment to have populated ``values`` and + ``quantities``; otherwise returns ``None``. + - This is an optimistic bound (not necessarily reachable under + protocol constraints) but is simple, fast, and comparable across + runs. + """ + info = sl.info or {} + if not info or not info.get("is_last_timestep_in_round"): + return None + quantities = info.get("quantities") or {} + values = info.get("values") or {} + if not values or not quantities: + return None + agent_ids = list(sl.rewards.keys()) + if type(values[agent_ids[0]]) is dict: + item_keys = list(values.values())[0].keys() + max_vals, max_quantities = [], [] + for item in item_keys: + max_val = max(float(agent_vals[item]) for agent_vals in values.values()) + max_vals.append(max_val) + max_quantities.append(quantities[item]) + else: + max_vals = [max(float(v) for v in values.values())] + max_quantities = [quantities[item] for item in quantities.keys()] + for aid in sl.rewards.keys(): + if "buffer" in str(aid) and "live" not in str(aid): + return None + achieved = sum(float(v) for v in sl.rewards.values()) + max_reward = sum(d * v for d, v in zip(max_quantities, max_vals)) + # Efficiency is a global metric; emit same value for a special key "all" + return [("split_efficiency", achieved / max_reward)] + + +def _extract_items_from_split(raw_split: Dict) -> Dict[str, float] | None: + """Return a mapping item->proposal amount from a split structure. + + Supports both generic negotiation splits with nested structure + { 'items_given_to_self': {item: qty, ...}} + and TAS coin-only variants which may already be a flat mapping {'coins': qty}. + """ + + if raw_split is None: + return {} + elif isinstance(raw_split, Split): + return {k: float(v) for k, v in raw_split.items_given_to_self.items()} + elif isinstance(raw_split, dict): + if "items_given_to_self" in raw_split and isinstance( + raw_split["items_given_to_self"], dict + ): + return {k: float(v) for k, v in raw_split["items_given_to_self"].items()} + # Fallback: assume already flat mapping of items + elif hasattr(raw_split, "items_given_to_self"): + return {k: float(v) for k, v in raw_split["items_given_to_self"].items()} + return { + k: float(v) for k, v in raw_split.items() if isinstance(v, (int, float)) + } + return {} + + +def _average_proposal_relative_value( + sl: SimulationStepLog, + metric_name: str, + comparator: Callable[[float, float], bool], + opposite_comparator: Callable[[float, float], bool], +) -> Dict[str, float | None] | None: + """Shared implementation for proposal size conditioned on relative value. + + Parameters: + comparator: returns True when agent_0's value relation (e.g. < or >) + to agent_1 holds for an item and we should collect agent_0's + proposed quantity for that item. + opposite_comparator: inverse relation used to collect agent_1's items. + + Behavior: + - Executes only on final timestep of a round (where the definitive + proposal / allocation is known via ``info['splits']``). + - For each item, classifies which agent's value satisfies the chosen + relation and records that agent's proposed quantity from the split. + - Averages (mean) across all qualifying items per agent; if no items + qualify for an agent returns ``None`` for that agent id. + - Adds ``all_agents`` mean across the numeric (non-None) agent values. + + Why this matters: + Distinguishing how much an agent *asks for* when it subjectively + values items more (or less) than its counterpart reveals patterns of + opportunism vs. concession. This is especially useful when raw reward + differences are subtle but allocation *intent* differs. + """ + info = sl.info or {} + if not info or not info.get("is_last_timestep_in_round"): + return None + quantities = info.get("quantities") or {} + splits = info.get("splits") or {} + values = info.get("values") or {} + agent_ids: List[str] = list(sl.rewards.keys()) + if len(agent_ids) != 2: + return None # Only defined for 2-agent case. + for aid in agent_ids: + if "buffer" in str(aid) and "live" not in str(aid): + return None + # Extract per-agent item proposals robustly + split_items = {aid: _extract_items_from_split(splits.get(aid)) for aid in agent_ids} + agent_0_vals: List[float] = [] + agent_1_vals: List[float] = [] + for item in quantities.keys(): + # Values may be either a float (same for all items) or dict per item + v0_raw = values[agent_ids[0]] + v1_raw = values[agent_ids[1]] + v0 = float(v0_raw[item]) if isinstance(v0_raw, dict) else float(v0_raw) + v1 = float(v1_raw[item]) if isinstance(v1_raw, dict) else float(v1_raw) + if comparator(v0, v1): + agent_0_vals.append(split_items[agent_ids[0]].get(item, 0.0)) + elif opposite_comparator(v0, v1): + agent_1_vals.append(split_items[agent_ids[1]].get(item, 0.0)) + out: Dict[str, float | None] = {} + out[f"{metric_name}-{agent_ids[0]}"] = ( + sum(agent_0_vals) / len(agent_0_vals) if agent_0_vals else None + ) + out[f"{metric_name}-{agent_ids[1]}"] = ( + sum(agent_1_vals) / len(agent_1_vals) if agent_1_vals else None + ) + + return [(key, value) for key, value in out.items() if value is not None] + + +def average_proposal_when_agent_values_item_lower( + sl: SimulationStepLog, +) -> List[Tuple[str, float | None]] | None: + """Mean quantity an agent proposes for items it values *less* than opponent. + + Interpretation: + A higher value implies the agent still claims (or is allocated) a + notable share of items where it has a comparative *disadvantage* in + valuation, signaling either strategic over-claiming or protocol-driven + egalitarian splits. Conversely, very low numbers can indicate + efficient specialization or excessive concession. + + Returns: + Mapping { agent_id: float | None, "all_agents": float | None } where + None indicates no qualifying items for that agent in the round. + """ + return _average_proposal_relative_value( + sl, + "average_proposal_when_agent_values_item_lower", + lambda a, b: a < b, + lambda a, b: a > b, + ) + + +def average_proposal_when_agent_values_item_higher( + sl: SimulationStepLog, +) -> List[Tuple[str, float | None]] | None: + """Mean quantity an agent proposes for items it values *more* than opponent. + + Interpretation: + Captures how aggressively an agent claims items where it holds a + comparative *advantage*. Elevated values can reflect rational + specialization (efficient exploitation of comparative advantage) or + potentially unfair grabs if paired with low concession in the lower + valuation metric. Comparing this with the 'lower' counterpart helps + profile negotiation style (cooperative vs. exploitative). + + Returns: + Mapping { agent_id: float | None, "all_agents": float | None } where + None indicates no qualifying items. + """ + return _average_proposal_relative_value( + sl, + "average_proposal_when_agent_values_item_higher", + lambda a, b: a > b, + lambda a, b: a < b, + ) + + +# Explicit list of metric functions exported for rendering. Helper functions +# starting with '_' are intentionally excluded. Update this list when adding +# new public statistics so render.py can rely on it instead of introspecting +# every callable in the module. +stat_functs: list[Callable[[SimulationStepLog], List[Tuple[str, float]]]] = [ + avg_reward, + average_proposal_when_agent_values_item_lower, + average_proposal_when_agent_values_item_higher, + split_efficiency, +] diff --git a/src_code_for_reproducibility/markov_games/negotiation/no_press_nego_simulation.py b/src_code_for_reproducibility/markov_games/negotiation/no_press_nego_simulation.py new file mode 100644 index 0000000000000000000000000000000000000000..e4b6cfeed0821a3ebdebdaa5abf4a8f09a69723c --- /dev/null +++ b/src_code_for_reproducibility/markov_games/negotiation/no_press_nego_simulation.py @@ -0,0 +1,182 @@ +""" +File: mllm/markov_games/negotiation/no_press_nego_simulation.py +Summary: Simulation driver for no-press negotiation scenarios. +""" + +import copy +from collections import defaultdict +from dataclasses import dataclass +from typing import Any, Dict, List, Literal, Tuple + +from mllm.markov_games.negotiation.nego_simulation import ( + NegotiationObs, + NegotiationSimulation, + NegotiationState, + Split, + compute_tas_style_rewards, +) + +AgentId = str + + +@dataclass +class NoPressState(NegotiationState): + """NegotiationState alias used to clarify we run in always-split phase.""" + + pass + + +@dataclass +class NoPressObs(NegotiationObs): + """Observation that includes both agents' values (since there is no messaging).""" + + other_value: Dict[str, float] + + +class NoPressSimulation(NegotiationSimulation): + def __init__( + self, + game_type: Literal["10-1-exclusive", "10-1-ties", "1-to-20"] = "1-to-20", + same_round_value: bool = True, + atleast_one_conflict: bool = False, + *args, + **kwargs, + ): + self.game_type = game_type + self.same_round_value = same_round_value + self.atleast_one_conflict = atleast_one_conflict + super().__init__(*args, **kwargs) + + def _sample_values(self) -> Dict[AgentId, dict]: + """Sample per-item valuations according to the configured template.""" + values = defaultdict(dict) + if self.state is None: + item_types = self.item_types + else: + item_types = list(self.state.quantities.keys()) + while True: + for item in item_types: + if self.game_type == "10-1-exclusive": + v = int(self.rng.choice([1, 10])) + values[self.agent_ids[0]][item] = v + values[self.agent_ids[1]][item] = 10 if v == 1 else 1 + elif self.game_type == "10-1-ties": + for aid in self.agent_ids: + values[aid][item] = int(self.rng.choice([1, 10])) + elif self.game_type == "1-to-20": + for aid in self.agent_ids: + values[aid][item] = int(self.rng.integers(1, 21)) + if self.atleast_one_conflict: + has_conflict = False + for item in item_types: + agent_values_for_item = [ + values[aid][item] for aid in self.agent_ids + ] + if len(set(agent_values_for_item)) > 1: + has_conflict = True + break + if not has_conflict: + continue + agent_values = [sum(v.values()) for v in values.values()] + if len(set(agent_values)) == 1 or not self.same_round_value: + break + return values + + def _sample_quantities(self) -> Dict[str, int]: + """No-press setups use symmetric 10-unit stocks for every item.""" + return {item.lower(): 10 for item in self.item_types} + + def set_new_round_of_variant(self): + """Refresh quantities/values and jump directly into the simultaneous split.""" + self.state.quantities = self._sample_quantities() + self.state.values = self._sample_values() + self.state.split_phase = True + + def get_info_of_variant( + self, state: NegotiationState, actions: Dict[AgentId, Any] + ) -> Dict[str, Any]: + """Surface quantities/values/splits so statistics modules can read them.""" + return { + "quantities": copy.deepcopy(state.quantities), + "values": copy.deepcopy(state.values), + "splits": copy.deepcopy(state.splits), + } + + def get_rewards(self, splits: Dict[AgentId, Split]) -> Dict[AgentId, float]: + """Reuse TAS reward logic because the split arbitration is identical.""" + return compute_tas_style_rewards( + self.agent_ids, self.state.values, splits, self.state.quantities + ) + + def get_obs(self): + return {agent_id: self.get_obs_agent(agent_id) for agent_id in self.agent_ids} + + def get_obs_agent(self, agent_id): + other_id = self._other(agent_id) + last_value_coagent = ( + None + if self.state.previous_values is None + else self.state.previous_values.get(other_id) + ) + last_points_coagent = ( + None + if self.state.previous_points is None + else round(self.state.previous_points.get(other_id), 1) + ) + last_value_agent = ( + None + if self.state.previous_values is None + else self.state.previous_values.get(agent_id) + ) + last_points_agent = ( + None + if self.state.previous_points is None + else round(self.state.previous_points.get(agent_id), 1) + ) + last_split_coagent = None + last_split_agent = None + if self.state.previous_splits is not None: + last_split_coagent = self.state.previous_splits[ + other_id + ].items_given_to_self + last_split_agent = self.state.previous_splits[agent_id].items_given_to_self + obs = NoPressObs( + round_nb=self.state.round_nb, + last_message="", + quota_messages_per_agent_per_round=self.quota_messages_per_agent_per_round, + current_agent=self.state.current_agent, + other_agent=self.agent_id_to_name[other_id], + quantities=self.state.quantities, + item_types=self.item_types, + value=self.state.values[agent_id], + split_phase=self.state.split_phase, + last_split_agent=last_split_agent, + last_value_agent=last_value_agent, + last_points_agent=last_points_agent, + last_split_coagent=last_split_coagent, + last_value_coagent=last_value_coagent, + last_points_coagent=last_points_coagent, + other_value=self.state.values[other_id], + last_quantities=self.state.previous_quantities, + ) + return obs + + def reset(self): + start_agent = self.agent_ids[self._starting_agent_index] + quantities = self._sample_quantities() + values = self._sample_values() + self.state = NoPressState( + round_nb=0, + last_message="", + current_agent=start_agent, + quantities=quantities, + values=values, + previous_values=None, + splits={aid: None for aid in self.agent_ids}, + nb_messages_sent={aid: 0 for aid in self.agent_ids}, + split_phase=True, + previous_splits=None, + previous_points=None, + previous_quantities=None, + ) + return self.get_obs() diff --git a/src_code_for_reproducibility/markov_games/negotiation/tas_agent.py b/src_code_for_reproducibility/markov_games/negotiation/tas_agent.py new file mode 100644 index 0000000000000000000000000000000000000000..9e0ba17b1fcf2be321966898af7a024c31af9e08 --- /dev/null +++ b/src_code_for_reproducibility/markov_games/negotiation/tas_agent.py @@ -0,0 +1,118 @@ +""" +File: mllm/markov_games/negotiation/tas_agent.py +Summary: Agent implementation for Take-and-Split negotiations. +""" + +from mllm.markov_games.negotiation.nego_agent import NegotiationAgent +from mllm.markov_games.negotiation.nego_simulation import Split +from mllm.markov_games.negotiation.tas_simulation import TrustAndSplitObs + + +class TrustAndSplitAgent(NegotiationAgent): + """Prompt/template wrapper for the classic multi-item Take-and-Split benchmark.""" + + def __init__(self, num_message_chars, *args, **kwargs): + self.num_message_chars = num_message_chars + super().__init__(*args, **kwargs) + self.intro_prompt = ( + "Welcome to an iterated game. You are {agent}. The other agent is {other_agent}.\n" + "Setup:\n" + "1. The game has multiple independent rounds.\n" + "2. In each round, there are multiple items to split between the two agents.\n" + "3. Both agents are assigned a per-item value between 1 and 20 (inclusive) in each round.\n" + "4. You can only observe your own per-item values.\n" + "5. Because assignments are random, both agents are equally likely to have same expected per-item value.\n" + "\n" + "Protocol:\n" + "1. At the start of the round, one agent begins the conversation. The starting role alternates each round.\n" + "2. Agents exchange a short chat ({quota_messages_per_agent_per_round} messages per round per agent) to negotiate how to split the item.\n" + " - Use this chat to communicate your private per-item value to make informed proposals.\n" + "3. After the chat, both agents simultaneously propose the amount of each item they will keep.\n" + "4. If the total sum of proposals is less than or equal to the item quantity, both agents receive their proposed amounts.\n" + "5. If the total sum of proposals exceeds the item quantity, they are allocated proportionally.\n" + "6. Your points for the round = (amount you receive per item) x (your per-item value for that round), added across all items.\n" + "7. Points are accumulated across rounds.\n" + "Your goal: {goal}\n" + ) + self.new_round_prompt = ( + "A New Round Begins\n" + "The items to split are {quantities}.\n" + "Your per-item values are {value}." + ) + self.last_round_prompt = ( + "Last Round Summary:\n" + " - Items to split: {last_quantities}\n" + " - Your per-item values: {last_value_agent}\n" + " - {other_agent}'s per-item values: {last_value_coagent}\n" + " - You proposed: {last_split_agent}\n" + " - You earned: {last_points_agent} points\n" + " - {other_agent} proposed: {last_split_coagent}\n" + " - {other_agent} earned: {last_points_coagent} points\n" + " - Round Complete.\n" + ) + self.send_split_prompt = ( + "Message quota is finished for this round.\n" + "{other_agent} has finalized their proposal.\n" + "Submit your finalization now\n" + "Respond with {proposal_style2}" + ) + # self.wait_for_message_prompt = "Wait for {other_agent} to send a message..." + self.wait_for_message_prompt = "" + self.last_message_prompt = "{other_agent} said: {last_message}" + # self.send_message_prompt = ( + # f"Send your message now (max {self.num_message_chars} chars)." + # ) + self.send_message_prompt = f"Send your message now in ... (<={self.num_message_chars} chars)." + + def get_message_regex(self, observation: TrustAndSplitObs) -> str: + """Constrain chat to bounded XML tags for stable parsing.""" + return rf"[\s\S]{{0,{self.num_message_chars}}}" + + # def get_message_regex(self, observation: TrustAndSplitObs) -> str: + # return rf"(?s).{{0,{self.num_message_chars}}}" + + def get_split_regex(self, observation: TrustAndSplitObs) -> str: + """Allow natural-language item names while still returning machine-parsable XML.""" + items = list(observation.quantities.keys()) + # Accept both singular and plural forms + item_pattern = "|".join( + [f"{item[:-1]}s?" if item.endswith("s") else f"{item}s?" for item in items] + ) + regex = rf"(?i) ?((?:\s*(?P(10|[0-9]))\s*(?P{item_pattern})\s*,?)+) ?" + return regex + + def get_split_action( + self, policy_output: str, observation: TrustAndSplitObs + ) -> Split: + """Convert human-readable allocation text back into canonical item IDs.""" + items = list(observation.quantities.keys()) + import re as _re + + split_regex = self.get_split_regex(observation) + items_given_to_self = {item: 0 for item in items} + m = _re.match(split_regex, policy_output.strip()) + if m: + # Find all (number, item) pairs + item_pattern = "|".join( + [ + f"{item[:-1]}s?" if item.endswith("s") else f"{item}s?" + for item in items + ] + ) + inner_regex = rf"(?i)(10|[0-9])\s*({item_pattern})" + + def normalize_item_name(item_str): + for orig in items: + if item_str.lower() == orig.lower(): + return orig + if orig.endswith("s") and item_str.lower() == orig[:-1].lower(): + return orig + if ( + not orig.endswith("s") + and item_str.lower() == orig.lower() + "s" + ): + return orig + + for num, item in _re.findall(inner_regex, m.group(1)): + items_given_to_self[normalize_item_name(item)] = int(num) + return Split(items_given_to_self=items_given_to_self) diff --git a/src_code_for_reproducibility/markov_games/negotiation/tas_rps_simulation.py b/src_code_for_reproducibility/markov_games/negotiation/tas_rps_simulation.py new file mode 100644 index 0000000000000000000000000000000000000000..ac4b67f5a04a916af94768031983555fbb33cb36 --- /dev/null +++ b/src_code_for_reproducibility/markov_games/negotiation/tas_rps_simulation.py @@ -0,0 +1,257 @@ +""" +File: mllm/markov_games/negotiation/tas_rps_simulation.py +Summary: Simulation for TAS Rock-Paper-Scissors blended scenarios. +""" + +import copy +from dataclasses import dataclass +from typing import Any, Dict, List, Literal, Tuple + +from mllm.markov_games.negotiation.nego_simulation import ( + Message, + NegotiationObs, + NegotiationSimulation, + NegotiationState, + Split, + compute_tas_style_rewards, +) +from mllm.markov_games.rollout_tree import SimulationStepLog + +AgentId = str + + +def _get_rps_winner( + hand1: Literal["rock", "paper", "scissors"], + hand2: Literal["rock", "paper", "scissors"], +) -> Literal["rock", "paper", "scissors"]: + """Determine winner of rock-paper-scissors between two hands.""" + if hand1 == hand2: + raise ValueError("Hands should be different") + if ( + (hand1 == "rock" and hand2 == "scissors") + or (hand1 == "paper" and hand2 == "rock") + or (hand1 == "scissors" and hand2 == "paper") + ): + return hand1 + else: + return hand2 + + +@dataclass +class TrustAndSplitRPSState(NegotiationState): + """Negotiation state augmented with the current and previous RPS hands.""" + + hands: Dict[ + AgentId, Literal["rock", "paper", "scissors"] + ] # rock, paper, or scissors + previous_hands: Dict[AgentId, Literal["rock", "paper", "scissors"]] | None + + +@dataclass +class TrustAndSplitRPSObs(NegotiationObs): + """Agent-facing observation enriched with last-hand metadata.""" + + hand: Literal["rock", "paper", "scissors"] + last_hand_agent: Literal["rock", "paper", "scissors"] | None + last_hand_coagent: Literal["rock", "paper", "scissors"] | None + last_hand_value_coagent: Literal["upper", "lower"] | None + + +class TrustAndSplitRPSSimulation(NegotiationSimulation): + """Negotiation variant that splices TAS splitting with RPS-determined stakes.""" + + def __init__( + self, + alternating_hands: bool = False, + alternating_mix_ratio: float = None, + *args, + **kwargs, + ): + self.alternating_hands = alternating_hands + self.alternating_mix_ratio = alternating_mix_ratio + super().__init__(*args, **kwargs) + if self.alternating_mix_ratio is not None: + if self.rng.random() < self.alternating_mix_ratio: + self.alternating_hands = True + else: + self.alternating_hands = False + + def _sample_hands_and_values( + self, + alternate_hands: bool = False, + ) -> Tuple[Dict[AgentId, str], Dict[AgentId, float]]: + """ + Sample a rock-paper-scissors hand for each agent plus the per-hand value. + + When ``alternate_hands`` is True we deliberately flip the previous round's + winner/loser roles to create nonstationary payoffs; otherwise we draw + uniformly without replacement. + """ + hands = ["rock", "paper", "scissors"] + if alternate_hands: + previous_hands = list(self.state.previous_hands.values()) + hand1, hand2 = self.rng.choice(hands, size=2, replace=False) + winner = _get_rps_winner(hand1, hand2) + loser = hand1 if winner == hand2 else hand2 + previous_winner = _get_rps_winner(previous_hands[0], previous_hands[1]) + agent_hands, values = {}, {} + for agent_id in self.agent_ids: + if self.state.previous_hands[agent_id] == previous_winner: + agent_hands[agent_id] = loser + values[agent_id] = 1.0 + else: + agent_hands[agent_id] = winner + values[agent_id] = 10.0 + return agent_hands, values + else: + # Assign different hands to each agent + hand1, hand2 = self.rng.choice(hands, size=2, replace=False) + + agent_hands = {self.agent_ids[0]: hand1, self.agent_ids[1]: hand2} + + # Determine winner and assign values + winner = _get_rps_winner(hand1, hand2) + values = {} + for agent_id in self.agent_ids: + if agent_hands[agent_id] == winner: + values[agent_id] = 10.0 # Winner gets value 10 + else: + values[agent_id] = 1.0 # Loser gets value 1 + + return agent_hands, values + + def set_new_round_of_variant(self): + """Refresh hands/values and reset round-specific state.""" + self.state.previous_hands = copy.deepcopy(self.state.hands) + new_hands, new_values = self._sample_hands_and_values( + alternate_hands=self.alternating_hands + ) + self.state.hands = new_hands + self.state.values = new_values + # Quantities are constant in TAS + self.state.quantities = {"coins": 10} + self.state.split_phase = False + + def get_info_of_variant( + self, state: NegotiationState, actions: Dict[AgentId, Any] + ) -> Dict[str, Any]: + """Expose variant-specific tensors for downstream logging/analysis.""" + return { + "quantities": copy.deepcopy(state.quantities), + "hands": copy.deepcopy(state.hands), + "values": copy.deepcopy(state.values), + "previous_hands": copy.deepcopy(state.previous_hands), + "previous_values": copy.deepcopy(state.previous_values), + "splits": copy.deepcopy(state.splits), + } + + def get_rewards(self, splits: Dict[AgentId, Split]) -> Dict[AgentId, float]: + """Delegates to TAS reward helper because the payout rule is identical.""" + return compute_tas_style_rewards( + self.agent_ids, self.state.values, splits, self.state.quantities + ) + + def get_obs_agent(self, agent_id): + """Return a full Trust-and-Split observation for ``agent_id``.""" + other_id = self._other(agent_id) + last_value_coagent = ( + None + if self.state.previous_values is None + else self.state.previous_values.get(other_id) + ) + last_hand_coagent = ( + None + if self.state.previous_hands is None + else self.state.previous_hands.get(other_id) + ) + last_points_coagent = ( + None + if self.state.previous_points is None + else round(self.state.previous_points.get(other_id), 1) + ) + last_value_agent = ( + None + if self.state.previous_values is None + else self.state.previous_values.get(agent_id) + ) + last_hand_agent = ( + None + if self.state.previous_hands is None + else self.state.previous_hands.get(agent_id) + ) + last_points_agent = ( + None + if self.state.previous_points is None + else round(self.state.previous_points.get(agent_id), 1) + ) + last_split_coagent = None + last_split_agent = None + if self.state.previous_splits is not None: + last_split_coagent = self.state.previous_splits[ + other_id + ].items_given_to_self["coins"] + last_split_agent = self.state.previous_splits[agent_id].items_given_to_self[ + "coins" + ] + if last_hand_agent is None or last_hand_coagent is None: + last_hand_value_coagent = None + else: + winner = _get_rps_winner(last_hand_agent, last_hand_coagent) + last_hand_value_coagent = ( + "upper" if winner == last_hand_coagent else "lower" + ) + obs = TrustAndSplitRPSObs( + round_nb=self.state.round_nb, + last_message=self.state.last_message, + quota_messages_per_agent_per_round=self.quota_messages_per_agent_per_round, + current_agent=self.state.current_agent, + other_agent=self.agent_id_to_name[other_id], + quantities={"coins": 10}, + item_types=self.item_types, + value=self.state.values[agent_id], + split_phase=self.state.split_phase, + last_split_agent=last_split_agent, + last_value_agent=last_value_agent, + last_points_agent=last_points_agent, + last_split_coagent=last_split_coagent, + last_value_coagent=last_value_coagent, + last_points_coagent=last_points_coagent, + hand=self.state.hands[agent_id], + last_hand_coagent=last_hand_coagent, + last_hand_agent=last_hand_agent, + last_quantities=self.state.previous_quantities, + last_hand_value_coagent=last_hand_value_coagent, + ) + return obs + + def get_state(self): + return self.state + + def get_safe_copy(self): + """Return a safe copy of the simulation.""" + simulation_copy = copy.copy(self) + simulation_copy.state = copy.deepcopy(self.state) + return simulation_copy + + def reset(self): + """Initialize and return initial observations""" + # Decide starting agent alternating across resets for determinism + start_agent = self.agent_ids[self._starting_agent_index] + hands, values = self._sample_hands_and_values() + self.state = TrustAndSplitRPSState( + round_nb=0, + last_message="", + current_agent=start_agent, + quantities={"coins": 10}, + values=values, + splits={aid: None for aid in self.agent_ids}, + nb_messages_sent={aid: 0 for aid in self.agent_ids}, + previous_values=None, + previous_splits=None, + previous_points=None, + split_phase=False, + hands=hands, + previous_hands=None, + previous_quantities=None, + ) + return self.get_obs() diff --git a/src_code_for_reproducibility/markov_games/rollout_tree.py b/src_code_for_reproducibility/markov_games/rollout_tree.py new file mode 100644 index 0000000000000000000000000000000000000000..c9feb0e92f3bcf19255d80c6ff2dcd9a045c6c21 --- /dev/null +++ b/src_code_for_reproducibility/markov_games/rollout_tree.py @@ -0,0 +1,95 @@ +""" +File: mllm/markov_games/rollout_tree.py +Summary: Defines rollout tree data structures and serialization helpers. +""" + +from __future__ import annotations + +import json +from dataclasses import dataclass +from pathlib import Path +from typing import Any, List, Literal, Optional, Tuple + +import jsonschema +from pydantic import BaseModel, Field, model_validator + +from mllm.chat_utils.chat_turn import ChatTurn + +AgentId = str + + +class SimulationStepLog(BaseModel): + """Minimal snapshot of environment-side rewards and auxiliary info.""" + + rewards: dict[AgentId, float] + info: Any = None + + +class AgentActLog(BaseModel): + """LLM-side provenance for an action (chat turns + metadata).""" + + chat_turns: list[ChatTurn] | None + info: Any = None + + @model_validator(mode="after") + def _exactly_one_state_end(self): + """ + This method is used to enforce that for each AgentActLog, there is exactly one ChatTurn which is a state end. + """ + if self.chat_turns != []: + n = sum(1 for t in self.chat_turns if t.is_state_end) + if n != 1: + raise ValueError( + f"AgentActLog must have exactly one ChatTurn with is_state_end=True; got {self.chat_turns}." + ) + return self + else: + return self + + +class StepLog(BaseModel): + action_logs: dict[AgentId, AgentActLog] + simulation_step_log: SimulationStepLog + + +# BranchType = Literal["unilateral_deviation", "common_deviation"] # might not be necessary +# class BranchNodeInfo(BaseModel): +# branch_id: str +# branch_for: AgentId +# branch_type: BranchType + + +class RolloutTreeNode(BaseModel): + """Single timestep of the main trajectory (or a branch) plus linkage.""" + + step_log: StepLog + time_step: int + child: RolloutTreeNode | RolloutTreeBranchNode | None = None + + +class RolloutTreeBranchNode(BaseModel): + """ + First item of the tuple indicates which agent "called" for an alternative branch. + """ + + main_child: RolloutTreeNode + branches: dict[AgentId, list[RolloutTreeNode]] | None = None + + +class RolloutTreeRootNode(BaseModel): + """Entry point for serialized rollouts (main path plus optional branches).""" + + id: int + crn_id: int # ID of the rng used to generate this rollout tree + child: RolloutTreeNode | RolloutTreeBranchNode | None = None + agent_ids: List[AgentId] = Field(min_length=1) + + +# class RolloutTreeLeafNode(BaseModel): +# step_log: StepLog +# time_step: int + + +# Necessary for self-referential stuff in pydantic +RolloutTreeBranchNode.model_rebuild() +RolloutTreeNode.model_rebuild() diff --git a/src_code_for_reproducibility/markov_games/run_markov_games.py b/src_code_for_reproducibility/markov_games/run_markov_games.py new file mode 100644 index 0000000000000000000000000000000000000000..86b7f2ef0ac2e19afdd61a8a32342aaf8120e7de --- /dev/null +++ b/src_code_for_reproducibility/markov_games/run_markov_games.py @@ -0,0 +1,35 @@ +""" +File: mllm/markov_games/run_markov_games.py +Summary: CLI entry point for running configured Markov-game experiments. +""" + +import asyncio +from collections.abc import Callable +from dataclasses import dataclass + +from torch._C import ClassType + +from mllm.markov_games.markov_game import MarkovGame +from mllm.markov_games.rollout_tree import RolloutTreeRootNode + + +async def run_markov_games( + runner: Callable[[MarkovGame], RolloutTreeRootNode], + runner_kwargs: dict, + output_folder: str, + markov_games: list[MarkovGame], +) -> list[RolloutTreeRootNode]: + """ + Kick off multiple Markov game rollouts concurrently and return their trees. + + Parameters mirror the Hydra configs (runner callable + kwargs) so callers can + choose ``LinearRunner``, ``AlternativeActionsRunner`` or future variants. + """ + tasks = [] + for mg in markov_games: + tasks.append( + asyncio.create_task( + runner(markov_game=mg, output_folder=output_folder, **runner_kwargs) + ) + ) + return await asyncio.gather(*tasks) diff --git a/src_code_for_reproducibility/markov_games/simulation.py b/src_code_for_reproducibility/markov_games/simulation.py new file mode 100644 index 0000000000000000000000000000000000000000..dd0a2e61924f9a79aee3229ed8d7aa20827ae859 --- /dev/null +++ b/src_code_for_reproducibility/markov_games/simulation.py @@ -0,0 +1,94 @@ +""" +File: mllm/markov_games/simulation.py +Summary: Core simulation loop utilities and step logging for Markov games. +""" + +from abc import ABC, abstractmethod +from typing import Any, Tuple + +from numpy.random import default_rng + +from mllm.markov_games.rollout_tree import SimulationStepLog + + +class Simulation(ABC): + @abstractmethod + def __init__(self, seed: int, *args, **kwargs): + self.seed = seed + self.rng = default_rng(self.seed) + + @abstractmethod + def step(self, actions: Any) -> Tuple[bool, SimulationStepLog]: + """ + Advance the environment by one logical tick using ``actions``. + + Returns + ------- + terminated: bool + Whether the episode has finished. + SimulationStepLog + Reward/info bundle describing this transition. + """ + raise NotImplementedError + + def get_obs(self): + """Return a dict mapping agent_id -> observation for *all* agents.""" + raise NotImplementedError + + def get_obs_agent(self, agent_id): + """Return the observation for a single agent.""" + raise NotImplementedError + + def get_obs_size(self): + """Describe the observation tensor shape (useful for critic heads).""" + raise NotImplementedError + + def get_state(self): + """Return the privileged simulator state if available.""" + raise NotImplementedError + + def get_state_size(self): + """Describe the state tensor shape.""" + raise NotImplementedError + + def get_avail_actions(self): + """Return the global action mask/tensor if the space is discrete.""" + raise NotImplementedError + + def get_avail_agent_actions(self, agent_id): + """Return the available action mask for a given agent.""" + raise NotImplementedError + + def get_total_actions(self): + """Returns the total number of actions an agent could ever take. + + Implementations currently assume a discrete, one-dimensional action space per agent. + """ + raise NotImplementedError + + def get_safe_copy(self): + """ + Return copy of the simulator that shares no mutable state with the original. + """ + raise NotImplementedError + + def reset(self): + """Reset to the initial state and return the starting observations.""" + raise NotImplementedError + + def render(self): + """Optional human-facing visualization.""" + raise NotImplementedError + + def close(self): + """Release any owned resources (files, processes, etc.).""" + raise NotImplementedError + + # def seed(self): + # raise NotImplementedError + + def save_replay(self): + raise NotImplementedError + + def get_simulation_info(self): + raise NotImplementedError diff --git a/src_code_for_reproducibility/markov_games/statistics_runner.py b/src_code_for_reproducibility/markov_games/statistics_runner.py new file mode 100644 index 0000000000000000000000000000000000000000..e58131fc505806a758936978a46e4f8faefacad3 --- /dev/null +++ b/src_code_for_reproducibility/markov_games/statistics_runner.py @@ -0,0 +1,415 @@ +""" +File: mllm/markov_games/statistics_runner.py +Summary: Executes multiple rollouts to compute experiment statistics. +""" + +from __future__ import annotations + +import gc +import json +import pickle +from dataclasses import dataclass +from pathlib import Path +from typing import Any, Callable, Dict, Iterable, Iterator, List, Optional + +from basic_render import find_iteration_folders + +from mllm.markov_games.rollout_tree import ( + RolloutTreeBranchNode, + RolloutTreeNode, + RolloutTreeRootNode, + SimulationStepLog, +) + + +def _iterate_main_nodes(root: RolloutTreeRootNode) -> Iterator[RolloutTreeNode]: + """ + Iterate the main path nodes without materializing full path lists. + """ + current = root.child + while current is not None: + if isinstance(current, RolloutTreeNode): + yield current + current = current.child + elif isinstance(current, RolloutTreeBranchNode): + # Follow only the main child on the main trajectory + current = current.main_child + else: + break + + +def iterate_main_simulation_logs( + root: RolloutTreeRootNode, +) -> Iterator[SimulationStepLog]: + """Yield ``SimulationStepLog`` objects along the main (non-branch) path.""" + for node in _iterate_main_nodes(root): + yield node.step_log.simulation_step_log + + +def stream_rollout_files(iteration_folder: Path) -> Iterator[Path]: + """Iterate over every ``*.rt.pkl`` file under an iteration directory.""" + for p in iteration_folder.rglob("*.rt.pkl"): + if p.is_file(): + yield p + + +def load_root(path: Path) -> RolloutTreeRootNode: + """Load and validate a rollout tree from disk.""" + with open(path, "rb") as f: + data = pickle.load(f) + return RolloutTreeRootNode.model_validate(data) + + +@dataclass +class StatRecord: + """Convenience container for serialized stat rows.""" + + mgid: int + crn_id: Optional[int] + iteration: str + values: Dict[str, Any] + + +class StatComputer: + """ + Stateful stat computer that consumes SimulationStepLog instances + and produces final aggregated values for one rollout (mgid). + """ + + def update(self, sl: SimulationStepLog) -> None: # pragma: no cover - interface + raise NotImplementedError + + def finalize(self) -> Dict[str, Any]: # pragma: no cover - interface + raise NotImplementedError + + +def run_stats( + data_root: Path, + game_name: str, + make_computers: Callable[[], List[StatComputer]], + output_filename: Optional[str] = None, + output_format: str = "json", # "json" (dict of lists) or "jsonl" +) -> Path: + """ + Compute stats across all iteration_* folders under data_root. + Writes JSONL to data_root/statistics/. + """ + data_root = Path(data_root) + outdir = data_root / "statistics" + outdir.mkdir(parents=True, exist_ok=True) + # Choose extension by format + default_name = ( + f"{game_name}.stats.json" + if output_format == "json" + else f"{game_name}.stats.jsonl" + ) + outfile = outdir / ( + output_filename if output_filename is not None else default_name + ) + + # Rewrite file each run to keep it clean and small + if outfile.exists(): + outfile.unlink() + + iteration_folders = find_iteration_folders(str(data_root)) + + # If writing JSONL, stream directly; otherwise accumulate minimal records + if output_format == "jsonl": + with open(outfile, "w", encoding="utf-8") as w: + for iteration_folder in iteration_folders: + iteration_name = Path(iteration_folder).name + for pkl_path in stream_rollout_files(Path(iteration_folder)): + root = load_root(pkl_path) + + computers = make_computers() + for sl in iterate_main_simulation_logs(root): + for comp in computers: + try: + comp.update(sl) + except Exception: + continue + + values: Dict[str, Any] = {} + for comp in computers: + try: + values.update(comp.finalize()) + except Exception: + continue + + rec = { + "mgid": getattr(root, "id", None), + "crn_id": getattr(root, "crn_id", None), + "iteration": iteration_name, + "stats": values, + } + w.write(json.dumps(rec, ensure_ascii=False) + "\n") + + del root + del computers + gc.collect() + else: + # Aggregate to dict-of-lists for easier plotting + records: List[Dict[str, Any]] = [] + # Process in deterministic order + for iteration_folder in iteration_folders: + iteration_name = Path(iteration_folder).name + for pkl_path in stream_rollout_files(Path(iteration_folder)): + root = load_root(pkl_path) + + computers = make_computers() + for sl in iterate_main_simulation_logs(root): + for comp in computers: + try: + comp.update(sl) + except Exception: + continue + + values: Dict[str, Any] = {} + for comp in computers: + try: + values.update(comp.finalize()) + except Exception: + continue + + records.append( + { + "mgid": getattr(root, "id", None), + "crn_id": getattr(root, "crn_id", None), + "iteration": iteration_name, + "stats": values, + } + ) + + del root + del computers + gc.collect() + + # Build dict-of-lists with nested stats preserved + # Collect all stat keys and nested agent keys where needed + mgids: List[Any] = [] + crn_ids: List[Any] = [] + iterations_out: List[str] = [] + # stats_out is a nested structure mirroring keys but with lists + stats_out: Dict[str, Any] = {} + + # First pass to collect union of keys + stat_keys: set[str] = set() + nested_agent_keys: Dict[str, set[str]] = {} + for r in records: + stats = r.get("stats", {}) or {} + for k, v in stats.items(): + stat_keys.add(k) + if isinstance(v, dict): + nested = nested_agent_keys.setdefault(k, set()) + for ak in v.keys(): + nested.add(str(ak)) + + # Initialize structure + for k in stat_keys: + if k in nested_agent_keys: + stats_out[k] = {ak: [] for ak in sorted(nested_agent_keys[k])} + else: + stats_out[k] = [] + + # Fill lists + for r in records: + mgids.append(r.get("mgid")) + crn_ids.append(r.get("crn_id")) + iterations_out.append(r.get("iteration")) + stats = r.get("stats", {}) or {} + for k in stat_keys: + val = stats.get(k) + if isinstance(stats_out[k], dict): + # per-agent dict + agent_dict = val if isinstance(val, dict) else {} + for ak in stats_out[k].keys(): + stats_out[k][ak].append(agent_dict.get(ak)) + else: + stats_out[k].append(val) + + with open(outfile, "w", encoding="utf-8") as w: + json.dump( + { + "mgid": mgids, + "crn_id": crn_ids, + "iteration": iterations_out, + "stats": stats_out, + }, + w, + ensure_ascii=False, + ) + + return outfile + + +def run_stats_functional( + data_root: Path, + game_name: str, + metrics: Dict[str, Callable[[SimulationStepLog], Optional[Dict[str, float]]]], + output_filename: Optional[str] = None, + output_format: str = "json", +) -> Path: + """ + Functional variant where metrics is a dict of name -> f(SimulationStepLog) -> {agent_id: value}. + Aggregates per rollout by averaging over steps where a metric produced a value. + Writes a single consolidated file in data_root/statistics/. + """ + data_root = Path(data_root) + outdir = data_root / "statistics" + outdir.mkdir(parents=True, exist_ok=True) + default_name = ( + f"{game_name}.stats.json" + if output_format == "json" + else f"{game_name}.stats.jsonl" + ) + outfile = outdir / ( + output_filename if output_filename is not None else default_name + ) + + if outfile.exists(): + outfile.unlink() + + iteration_folders = find_iteration_folders(str(data_root)) + + def finalize_rollout( + agg: Dict[str, Dict[str, List[float]]] + ) -> Dict[str, Dict[str, float]]: + # avg per metric per agent + result: Dict[str, Dict[str, float]] = {} + for mname, agent_values in agg.items(): + result[mname] = {} + for aid, vals in agent_values.items(): + if not vals: + result[mname][aid] = None # keep alignment; could be None + else: + result[mname][aid] = sum(vals) / len(vals) + return result + + if output_format == "jsonl": + with open(outfile, "w", encoding="utf-8") as w: + for iteration_folder in iteration_folders: + iteration_name = Path(iteration_folder).name + for pkl_path in stream_rollout_files(Path(iteration_folder)): + root = load_root(pkl_path) + + # aggregator structure: metric -> agent_id -> list of values + agg: Dict[str, Dict[str, List[float]]] = { + m: {} for m in metrics.keys() + } + + for sl in iterate_main_simulation_logs(root): + for mname, fn in metrics.items(): + try: + vals = fn(sl) + except Exception: + vals = None + if not vals: + continue + for aid, v in vals.items(): + if v is None: + continue + lst = agg[mname].setdefault(str(aid), []) + try: + lst.append(float(v)) + except Exception: + continue + + values = finalize_rollout(agg) + rec = { + "mgid": getattr(root, "id", None), + "crn_id": getattr(root, "crn_id", None), + "iteration": iteration_name, + "stats": values, + } + w.write(json.dumps(rec, ensure_ascii=False) + "\n") + + del root + gc.collect() + else: + records: List[Dict[str, Any]] = [] + for iteration_folder in iteration_folders: + iteration_name = Path(iteration_folder).name + for pkl_path in stream_rollout_files(Path(iteration_folder)): + root = load_root(pkl_path) + + agg: Dict[str, Dict[str, List[float]]] = {m: {} for m in metrics.keys()} + for sl in iterate_main_simulation_logs(root): + for mname, fn in metrics.items(): + try: + vals = fn(sl) + except Exception: + vals = None + if not vals: + continue + for aid, v in vals.items(): + if v is None: + continue + lst = agg[mname].setdefault(str(aid), []) + try: + lst.append(float(v)) + except Exception: + continue + + values = finalize_rollout(agg) + records.append( + { + "mgid": getattr(root, "id", None), + "crn_id": getattr(root, "crn_id", None), + "iteration": iteration_name, + "stats": values, + } + ) + + del root + gc.collect() + + # Build dict-of-lists output + mgids: List[Any] = [] + crn_ids: List[Any] = [] + iterations_out: List[str] = [] + stats_out: Dict[str, Any] = {} + + stat_keys: set[str] = set() + nested_agent_keys: Dict[str, set[str]] = {} + for r in records: + stats = r.get("stats", {}) or {} + for k, v in stats.items(): + stat_keys.add(k) + if isinstance(v, dict): + nested = nested_agent_keys.setdefault(k, set()) + for ak in v.keys(): + nested.add(str(ak)) + + for k in stat_keys: + if k in nested_agent_keys: + stats_out[k] = {ak: [] for ak in sorted(nested_agent_keys[k])} + else: + stats_out[k] = [] + + for r in records: + mgids.append(r.get("mgid")) + crn_ids.append(r.get("crn_id")) + iterations_out.append(r.get("iteration")) + stats = r.get("stats", {}) or {} + for k in stat_keys: + val = stats.get(k) + if isinstance(stats_out[k], dict): + agent_dict = val if isinstance(val, dict) else {} + for ak in stats_out[k].keys(): + stats_out[k][ak].append(agent_dict.get(ak)) + else: + stats_out[k].append(val) + + with open(outfile, "w", encoding="utf-8") as w: + json.dump( + { + "mgid": mgids, + "crn_id": crn_ids, + "iteration": iterations_out, + "stats": stats_out, + }, + w, + ensure_ascii=False, + ) + + return outfile diff --git a/src_code_for_reproducibility/models/__init__.py b/src_code_for_reproducibility/models/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..46d40ee71acb4c10a596d4107d18fd3e890df610 --- /dev/null +++ b/src_code_for_reproducibility/models/__init__.py @@ -0,0 +1,4 @@ +""" +File: mllm/models/__init__.py +Summary: Exports model-layer utilities from the models package. +""" diff --git a/src_code_for_reproducibility/models/__pycache__/__init__.cpython-312.pyc b/src_code_for_reproducibility/models/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..a723185386598b4a1423dc07cd3022c07e2fdfdc Binary files /dev/null and b/src_code_for_reproducibility/models/__pycache__/__init__.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/models/__pycache__/adapter_training_wrapper.cpython-312.pyc b/src_code_for_reproducibility/models/__pycache__/adapter_training_wrapper.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..20bd4c3c5474187bdc2c3856bf591729509b1269 Binary files /dev/null and b/src_code_for_reproducibility/models/__pycache__/adapter_training_wrapper.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/models/__pycache__/human_policy.cpython-312.pyc b/src_code_for_reproducibility/models/__pycache__/human_policy.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..e8250e46883c9a18d29e02c6917ea0a528366fed Binary files /dev/null and b/src_code_for_reproducibility/models/__pycache__/human_policy.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/models/__pycache__/inference_backend.cpython-312.pyc b/src_code_for_reproducibility/models/__pycache__/inference_backend.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..405731ad14e4fa8cc427a89e8f23062f05f28482 Binary files /dev/null and b/src_code_for_reproducibility/models/__pycache__/inference_backend.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/models/__pycache__/inference_backend_dummy.cpython-312.pyc b/src_code_for_reproducibility/models/__pycache__/inference_backend_dummy.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..9b0fb935c37b2dce88065a0da5188c58fcb2b762 Binary files /dev/null and b/src_code_for_reproducibility/models/__pycache__/inference_backend_dummy.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/models/__pycache__/inference_backend_vllm.cpython-312.pyc b/src_code_for_reproducibility/models/__pycache__/inference_backend_vllm.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..555da444de4de381e84ad3e9999f1b84aab9dd37 Binary files /dev/null and b/src_code_for_reproducibility/models/__pycache__/inference_backend_vllm.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/models/__pycache__/large_language_model_api.cpython-312.pyc b/src_code_for_reproducibility/models/__pycache__/large_language_model_api.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..250bd3412ee0a5198f76e824ef265ae2d0bb5732 Binary files /dev/null and b/src_code_for_reproducibility/models/__pycache__/large_language_model_api.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/models/__pycache__/large_language_model_local.cpython-312.pyc b/src_code_for_reproducibility/models/__pycache__/large_language_model_local.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..0a2d8e4d055374deea50f4647881338c04779a13 Binary files /dev/null and b/src_code_for_reproducibility/models/__pycache__/large_language_model_local.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/models/__pycache__/scalar_critic.cpython-312.pyc b/src_code_for_reproducibility/models/__pycache__/scalar_critic.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..8a8922c3e90b6faf918756a152fe55910127d33b Binary files /dev/null and b/src_code_for_reproducibility/models/__pycache__/scalar_critic.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/models/adapter_training_wrapper.py b/src_code_for_reproducibility/models/adapter_training_wrapper.py new file mode 100644 index 0000000000000000000000000000000000000000..f99c768935cca8203f7a9482d25aeade0dee0d59 --- /dev/null +++ b/src_code_for_reproducibility/models/adapter_training_wrapper.py @@ -0,0 +1,104 @@ +""" +File: mllm/models/adapter_training_wrapper.py +Summary: Wraps a shared LLM with adapter-specific PEFT handling for training. +""" + +import logging +from typing import Union + +import torch +import torch.nn as nn +from peft import LoraConfig, get_peft_model + +logger = logging.getLogger(__name__) + + +class AdapterWrapper(nn.Module): + """ + A thin façade that + • keeps a reference to a *shared* PEFT-wrapped model, + • ensures `set_adapter(adapter)` is called on every forward, + • exposes only the parameters that should be trained for that adapter + (plus whatever extra modules you name). + """ + + def __init__( + self, + shared_llm: nn.Module, + adapter_id: str, + lora_config: dict, + path: Union[str, None] = None, + ): + super().__init__() + self.shared_llm = shared_llm + self.adapter_id = adapter_id + lora_config = LoraConfig(**lora_config) + # this modifies the shared llm in place, adding a lora adapter inside + self.shared_llm = get_peft_model( + model=shared_llm, + peft_config=lora_config, + adapter_name=adapter_id, + ) + self.shared_llm.train() + # Load external adapter weights if provided + loaded_from: str | None = None + if path: + try: + # Supports both local filesystem paths and HF Hub repo IDs + self.shared_llm.load_adapter( + is_trainable=True, + model_id=path, + adapter_name=adapter_id, + ) + loaded_from = path + except ( + Exception + ) as exc: # noqa: BLE001 - want to log any load failure context + logger.warning( + f"Adapter '{adapter_id}': failed to load from '{path}': {exc}" + ) + + if loaded_from: + logger.info( + f"Adapter '{adapter_id}': loaded initial weights from '{loaded_from}'." + ) + else: + logger.info( + f"Adapter '{adapter_id}': initialized with fresh weights (no initial weights found)." + ) + + def parameters(self, recurse: bool = True): + """ + "recurse" is just for pytorch compatibility + """ + self.shared_llm.set_adapter(self.adapter_id) + params = [p for p in self.shared_llm.parameters() if p.requires_grad] + + return params + + def get_base_model_logits(self, contexts): + """ + Run the base model (without adapter) in inference mode, without tracking gradients. + This is useful to get reference logits for KL-divergence computation. + """ + with torch.no_grad(): + with self.shared_llm.disable_adapter(): + return self.shared_llm(input_ids=contexts)[0] + + def forward(self, *args, **kwargs): + self.shared_llm.set_adapter(self.adapter_id) + return self.shared_llm(*args, **kwargs) + + def save_pretrained(self, save_path): + self.shared_llm.save_pretrained(save_path) + + def gradient_checkpointing_enable(self, *args, **kwargs): + self.shared_llm.gradient_checkpointing_enable(*args, **kwargs) + + @property + def dtype(self): + return self.shared_llm.dtype + + @property + def device(self): + return self.shared_llm.device diff --git a/src_code_for_reproducibility/models/human_policy.py b/src_code_for_reproducibility/models/human_policy.py new file mode 100644 index 0000000000000000000000000000000000000000..699c2d817c17abfdd0e144e99e14fb2b3ba06872 --- /dev/null +++ b/src_code_for_reproducibility/models/human_policy.py @@ -0,0 +1,260 @@ +""" +File: mllm/models/human_policy.py +Summary: Implements an interactive human-in-the-loop policy for experiments. +""" + +import asyncio +import os +import re +import shutil +import sys +from typing import Callable, Dict, List, Optional + +from mllm.markov_games.rollout_tree import ChatTurn + +try: + import rstr # For generating example strings from regex +except Exception: # pragma: no cover + rstr = None + + +def _clear_terminal() -> None: + """ + Clear the terminal screen in a cross-platform manner. + """ + if sys.stdout.isatty(): + os.system("cls" if os.name == "nt" else "clear") + + +def _terminal_width(default: int = 100) -> int: + try: + return shutil.get_terminal_size().columns + except Exception: + return default + + +def _horizontal_rule(char: str = "─") -> str: + width = max(20, _terminal_width() - 2) + return char * width + + +class _Style: + # ANSI colors (bright, readable) + RESET = "\033[0m" + BOLD = "\033[1m" + DIM = "\033[2m" + # Foreground colors + FG_BLUE = "\033[94m" # user/system headers + FG_GREEN = "\033[92m" # human response header + FG_YELLOW = "\033[93m" # notices + FG_RED = "\033[91m" # errors + FG_MAGENTA = "\033[95m" # regex + FG_CYAN = "\033[96m" # tips + + +def _render_chat(state) -> str: + """ + Render prior messages in a compact, readable terminal format. + + Expected message dict keys: {"role": str, "content": str, ...} + """ + lines: List[str] = [] + lines.append(_horizontal_rule()) + lines.append(f"{_Style.FG_BLUE}{_Style.BOLD} Conversation so far {_Style.RESET}") + lines.append(_horizontal_rule()) + for chat in state: + role = chat.role + content = str(chat.content).strip() + # Map roles to display names and colors/emojis + if role == "assistant": + header = f"{_Style.FG_GREEN}{_Style.BOLD}HUMAN--🧑‍💻{_Style.RESET}" + elif role == "user": + header = f"{_Style.FG_BLUE}{_Style.BOLD}USER--⚙️{_Style.RESET}" + else: + header = f"[{_Style.DIM}{role.upper()}{_Style.RESET}]" + lines.append(header) + # Indent content for readability + for line in content.splitlines() or [""]: + lines.append(f" {line}") + lines.append("") + lines.append(_horizontal_rule()) + return "\n".join(lines) + + +async def _async_input(prompt_text: str) -> str: + """Non-blocking input using a background thread.""" + return await asyncio.to_thread(input, prompt_text) + + +def _short_regex_example(regex: str, max_len: int = 30) -> Optional[str]: + """ + Try to produce a short example string that matches the regex. + We attempt multiple times and pick the first <= max_len. + """ + if rstr is None: + return None + try: + for _ in range(20): + candidate = rstr.xeger(regex) + if len(candidate) <= max_len: + return candidate + # Fallback to truncation (may break match, so don't return) + return None + except Exception: + return None + + +def _detect_input_type(regex: str | None) -> tuple[str, str, str]: + """ + Detect what type of input is expected based on the regex pattern. + Returns (input_type, start_tag, end_tag) + """ + if regex is None: + return "text", "", "" + + if "message_start" in regex and "message_end" in regex: + return "message", "<>", "<>" + elif "proposal_start" in regex and "proposal_end" in regex: + return "proposal", "<>", "<>" + else: + return "text", "", "" + + +async def human_policy(state, agent_id, regex: str | None = None) -> str: + """ + Async human-in-the-loop policy. + + - Displays prior conversation context in the terminal. + - Prompts the user for a response. + - If a regex is provided, validates and re-prompts until it matches. + - Automatically adds formatting tags based on expected input type. + + Args: + prompt: Chat history as a list of {role, content} dicts. + regex: Optional fullmatch validation pattern. + + Returns: + The user's validated response string. + """ + # Detect input type and formatting + input_type, start_tag, end_tag = _detect_input_type(regex) + + while True: + _clear_terminal() + print(_render_chat(state)) + + if regex: + example = _short_regex_example(regex, max_len=30) + print( + f"{_Style.FG_MAGENTA}{_Style.BOLD}Expected format (regex fullmatch):{_Style.RESET}" + ) + print(f" {_Style.FG_MAGENTA}{regex}{_Style.RESET}") + if example: + print( + f"{_Style.FG_CYAN}Example (random, <=30 chars):{_Style.RESET} {example}" + ) + print(_horizontal_rule(".")) + + # Custom prompt based on input type + if input_type == "message": + print( + f"{_Style.FG_YELLOW}Type your message content (formatting will be added automatically):{_Style.RESET}" + ) + elif input_type == "proposal": + print( + f"{_Style.FG_YELLOW}Type your proposal (number only, formatting will be added automatically):{_Style.RESET}" + ) + else: + print( + f"{_Style.FG_YELLOW}Type your response and press Enter.{_Style.RESET}" + ) + + print( + f"{_Style.DIM}Commands: /help to view commands, /refresh to re-render, /quit to abort{_Style.RESET}" + ) + else: + print( + f"{_Style.FG_YELLOW}Type your response and press Enter.{_Style.RESET} {_Style.DIM}(/help for commands){_Style.RESET}" + ) + + user_in = (await _async_input("> ")).rstrip("\n") + + # Commands + if user_in.strip().lower() in {"/help", "/h"}: + print(f"\n{_Style.FG_CYAN}{_Style.BOLD}Available commands:{_Style.RESET}") + print( + f" {_Style.FG_CYAN}/help{_Style.RESET} or {_Style.FG_CYAN}/h{_Style.RESET} Show this help" + ) + print( + f" {_Style.FG_CYAN}/refresh{_Style.RESET} or {_Style.FG_CYAN}/r{_Style.RESET} Re-render the conversation and prompt" + ) + print( + f" {_Style.FG_CYAN}/quit{_Style.RESET} or {_Style.FG_CYAN}/q{_Style.RESET} Abort the run (raises KeyboardInterrupt)" + ) + await asyncio.sleep(1.0) + continue + if user_in.strip().lower() in {"/refresh", "/r"}: + continue + if user_in.strip().lower() in {"/quit", "/q"}: + raise KeyboardInterrupt("Human aborted run from human_policy") + + # Add formatting tags if needed + if start_tag and end_tag: + formatted_input = f"{start_tag}{user_in}{end_tag}" + else: + formatted_input = user_in + + if regex is None: + return ChatTurn( + role="assistant", agent_id=agent_id, content=formatted_input + ) + + # Validate against regex (fullmatch) + try: + pattern = re.compile(regex) + except re.error as e: + # If regex is invalid, fall back to accepting any input + print( + f"{_Style.FG_RED}Warning:{_Style.RESET} Provided regex is invalid: {e}. Accepting input without validation." + ) + await asyncio.sleep(0.5) + return ChatTurn( + role="assistant", agent_id=agent_id, content=formatted_input + ) + + if pattern.fullmatch(formatted_input): + return ChatTurn( + role="assistant", agent_id=agent_id, content=formatted_input + ) + + # Show validation error and re-prompt + print("") + print( + f"{_Style.FG_RED}{_Style.BOLD}Input did not match the required format.{_Style.RESET} Please try again." + ) + + if input_type == "message": + print( + f"You entered: {_Style.FG_CYAN}{start_tag}{user_in}{end_tag}{_Style.RESET}" + ) + print(f"Just type the message content without tags.") + elif input_type == "proposal": + print( + f"You entered: {_Style.FG_CYAN}{start_tag}{user_in}{end_tag}{_Style.RESET}" + ) + print(f"Just type the number without tags.") + else: + print(f"Expected (regex):") + print(f" {_Style.FG_MAGENTA}{regex}{_Style.RESET}") + + print(_horizontal_rule(".")) + print(f"{_Style.FG_YELLOW}Press Enter to retry...{_Style.RESET}") + await _async_input("") + + +def get_human_policies() -> Dict[str, Callable[[List[Dict]], str]]: + """ + Expose the human policy in the same map shape used elsewhere. + """ + # Type hint says Callable[[List[Dict]], str] but we intentionally return the async callable. + return {"human_policy": human_policy} # type: ignore[return-value] diff --git a/src_code_for_reproducibility/models/inference_backend.py b/src_code_for_reproducibility/models/inference_backend.py new file mode 100644 index 0000000000000000000000000000000000000000..c204482170d5a4418870805b620295cab294fab6 --- /dev/null +++ b/src_code_for_reproducibility/models/inference_backend.py @@ -0,0 +1,44 @@ +""" +File: mllm/models/inference_backend.py +Summary: Declares the inference backend interface and shared dataclasses. +""" + +from abc import ABC, abstractmethod +from dataclasses import dataclass +from typing import Any, Optional + + +@dataclass +class LLMInferenceOutput: + content: str + reasoning_content: str | None = None + log_probs: list[float] | None = None + out_token_ids: list[int] | None = None + + +class LLMInferenceBackend(ABC): + @abstractmethod + def __init__(self, **kwargs): + ... + + @abstractmethod + def prepare_adapter( + self, adapter_id: str, weights_got_updated: bool = False + ) -> None: + """Ensure adapter is ready/loaded for next generation call.""" + + @abstractmethod + async def generate(self, prompt: list[dict], regex: Optional[str] = None) -> str: + ... + + @abstractmethod + def toggle_training_mode(self) -> None: + ... + + @abstractmethod + def toggle_eval_mode(self) -> None: + ... + + @abstractmethod + def shutdown(self) -> None: + ... diff --git a/src_code_for_reproducibility/models/inference_backend_dummy.py b/src_code_for_reproducibility/models/inference_backend_dummy.py new file mode 100644 index 0000000000000000000000000000000000000000..22dd123f5fbcf9a976282b0657097edc680c6ac3 --- /dev/null +++ b/src_code_for_reproducibility/models/inference_backend_dummy.py @@ -0,0 +1,59 @@ +""" +File: mllm/models/inference_backend_dummy.py +Summary: Stub inference backend that returns synthetic completions for tests. +""" + +import asyncio +from typing import Optional + +import rstr +from transformers import AutoTokenizer + +from mllm.models.inference_backend import LLMInferenceBackend, LLMInferenceOutput +from mllm.utils.short_id_gen import generate_short_id + + +class DummyInferenceBackend(LLMInferenceBackend): + def __init__( + self, + *args, + **kwargs, + ): + pass + + def prepare_adapter( + self, + adapter_id: Optional[str], + weights_got_updated: bool, + adapter_path: Optional[str] = None, + ) -> None: + pass + + async def toggle_training_mode(self) -> None: + await asyncio.sleep(0) + pass + + async def toggle_eval_mode(self) -> None: + await asyncio.sleep(0) + pass + + def shutdown(self) -> None: + pass + + async def generate( + self, + prompt_text: str, + regex: Optional[str] = None, + extract_thinking: bool = False, + ) -> LLMInferenceOutput: + if regex: + # Create random string that respects the regex + return LLMInferenceOutput( + content=rstr.xeger(regex), + reasoning_content="I don't think, I am a dummy backend.", + ) + else: + return LLMInferenceOutput( + content="I am a dummy backend without a regex.", + reasoning_content="I don't think, I am a dummy backend.", + ) diff --git a/src_code_for_reproducibility/models/inference_backend_vllm.py b/src_code_for_reproducibility/models/inference_backend_vllm.py new file mode 100644 index 0000000000000000000000000000000000000000..a4a7fc73287cb676ce56beea5de77cf03fc24555 --- /dev/null +++ b/src_code_for_reproducibility/models/inference_backend_vllm.py @@ -0,0 +1,111 @@ +""" +File: mllm/models/inference_backend_vllm.py +Summary: Connects to in-process vLLM instances for batched generation. +""" + +import asyncio +import re +from typing import Optional + +import torch +from transformers import AutoTokenizer +from vllm import AsyncEngineArgs, AsyncLLMEngine, SamplingParams +from vllm.inputs import TokensPrompt +from vllm.lora.request import LoRARequest +from vllm.sampling_params import GuidedDecodingParams, RequestOutputKind + +from mllm.models.inference_backend import LLMInferenceBackend, LLMInferenceOutput +from mllm.utils.short_id_gen import generate_short_id + + +class VLLMAsyncBackend(LLMInferenceBackend): + def __init__( + self, + model_name: str, + tokenizer: AutoTokenizer, + # adapter_paths: dict[str, str], + engine_init_kwargs: dict = {}, + sampling_params: dict = {}, + ): + self.model_name = model_name + self.vllm_adapter_ids = {} + ea = dict(model=model_name, **engine_init_kwargs) + self.engine = AsyncLLMEngine.from_engine_args(AsyncEngineArgs(**ea)) + + self.sampling_params = sampling_params + self.tokenizer = tokenizer + + def prepare_adapter( + self, + adapter_id: Optional[str], + adapter_path: Optional[str], + weights_got_updated: bool, + ) -> None: + if weights_got_updated: + self.vllm_adapter_ids[adapter_id] = generate_short_id() + self.current_lora_request = LoRARequest( + adapter_id, + self.vllm_adapter_ids[adapter_id], + adapter_path, + ) + + async def toggle_training_mode(self) -> None: + await self.engine.sleep(level=1) + + async def toggle_eval_mode(self) -> None: + await self.engine.wake_up() + + def shutdown(self) -> None: + # No explicit close call; engine stops when process exits. + pass + + async def generate( + self, + input_token_ids: list[int], + regex: Optional[str] = None, + extract_thinking: bool = False, + ) -> LLMInferenceOutput: + # Build SamplingParams correctly + guided = GuidedDecodingParams(regex=regex) if regex else None + sp = SamplingParams( + **self.sampling_params, + guided_decoding=guided, + output_kind=RequestOutputKind.FINAL_ONLY, + ) + + prompt = TokensPrompt(prompt_token_ids=input_token_ids) + request_id = f"req-{asyncio.get_running_loop().time()}" + result_generator = self.engine.generate( + prompt, + sp, # SamplingParams(...) + request_id, + lora_request=self.current_lora_request, + ) + + async for out in result_generator: # with FINAL_ONLY this runs once + res = out + + raw_text = res.outputs[0].text + out_token_ids = res.outputs[0].token_ids + log_probs = [ + logprob_dict[token_id].logprob + for token_id, logprob_dict in zip(out_token_ids, res.outputs[0].logprobs) + ] + log_probs = torch.tensor(log_probs) + out_token_ids = torch.tensor(out_token_ids, dtype=torch.long) + content = raw_text + reasoning_content = None + + if extract_thinking: + m = re.match( + r"^\n\n([\s\S]*?)\n\n(.*)$", raw_text, flags=re.DOTALL + ) + if m: + reasoning_content = m.group(1) + content = m.group(2) + return LLMInferenceOutput( + content=content, + reasoning_content=reasoning_content, + log_probs=log_probs, + out_token_ids=out_token_ids, + ) diff --git a/src_code_for_reproducibility/models/large_language_model_api.py b/src_code_for_reproducibility/models/large_language_model_api.py new file mode 100644 index 0000000000000000000000000000000000000000..d687c85d5386ec909ab140c7166886725d64f97d --- /dev/null +++ b/src_code_for_reproducibility/models/large_language_model_api.py @@ -0,0 +1,174 @@ +""" +File: mllm/models/large_language_model_api.py +Summary: Implements API-based large-language-model inference adapters. +""" + +from __future__ import annotations + +import asyncio +import copy +import os +import random +import re +from typing import Any, Callable, Dict, List, Optional, Sequence + +import backoff +from openai import AsyncOpenAI, OpenAIError + +from mllm.markov_games.rollout_tree import ChatTurn +from mllm.models.inference_backend import LLMInferenceOutput + +# Static list copied from the public OpenAI docs until a discovery endpoint is exposed. +reasoning_models = [ + "gpt-5-nano", + "gpt-5-mini", + "gpt-5", + "o1-mini", + "o1", + "o1-pro", + "o3-mini", + "o3", + "o3-pro", + "o4-mini", + "o4", + "o4-pro", +] + + +class LargeLanguageModelOpenAI: + """Tiny async wrapper for OpenAI Chat Completions.""" + + def __init__( + self, + llm_id: str = "", + model: str = "gpt-4.1-mini", + api_key: Optional[str] = None, + base_url: Optional[str] = None, + timeout_s: float = 300.0, + regex_max_attempts: int = 10, + sampling_params: Optional[Dict[str, Any]] = None, + init_kwargs: Optional[Dict[str, Any]] = None, + output_directory: Optional[str] = None, + ) -> None: + self.llm_id = llm_id + self.model = model + key = api_key or os.getenv("OPENAI_API_KEY") + if not key: + raise RuntimeError( + "Set OPENAI_API_KEY as global environment variable or pass api_key." + ) + client_kwargs: Dict[str, Any] = {"api_key": key, "timeout": timeout_s} + if base_url: + client_kwargs["base_url"] = base_url + self.client = AsyncOpenAI(**client_kwargs) + + # Sampling/default request params set at init + self.sampling_params = sampling_params + self.use_reasoning = model in reasoning_models + if self.use_reasoning: + self.sampling_params["reasoning"] = { + "effort": "low", + "summary": "detailed", + } + self.regex_max_attempts = max(1, int(regex_max_attempts)) + + def get_inference_policies(self) -> Dict[str, Callable]: + return { + self.llm_id: self.get_action, + } + + async def prepare_adapter_for_inference(self, *args: Any, **kwargs: Any) -> None: + await asyncio.sleep(0) + pass + + async def toggle_eval_mode(self, *args: Any, **kwargs: Any) -> None: + await asyncio.sleep(0) + pass + + async def toggle_training_mode(self, *args: Any, **kwargs: Any) -> None: + await asyncio.sleep(0) + pass + + async def export_adapters(self, *args: Any, **kwargs: Any) -> None: + await asyncio.sleep(0) + pass + + async def checkpoint_all_adapters(self, *args: Any, **kwargs: Any) -> None: + await asyncio.sleep(0) + pass + + def extract_output_from_response(self, resp: Response) -> LLMInferenceOutput: + if len(resp.output) > 1: + summary = resp.output[0].summary + if summary != []: + reasoning_content = summary[0].text + reasoning_content = f"OpenAI Reasoning Summary: {reasoning_content}" + else: + reasoning_content = None + content = resp.output[1].content[0].text + else: + reasoning_content = None + content = resp.output[0].content[0].text + + return LLMInferenceOutput( + content=content, + reasoning_content=reasoning_content, + ) + + @backoff.on_exception( + backoff.expo, Exception, max_time=10**10, max_tries=10**10 + ) + async def get_action( + self, + state: list[ChatTurn], + agent_id: str, + regex: Optional[str] = None, + ) -> LLMInferenceOutput: + # Remove any non-role/content keys from the prompt else openai will error. + prompt = [{"role": p.role, "content": p.content} for p in state] + + # if self.sleep_between_requests: + # await self.wait_random_time() + + # If regex is required, prime the model and validate client-side + if regex: + constraint_msg = { + "role": "user", + "content": ( + f"Output must match this regex exactly: {regex} \n" + "Return only the matching string, with no quotes or extra text." + ), + } + prompt = [constraint_msg, *prompt] + pattern = re.compile(regex) + for _ in range(self.regex_max_attempts): + resp = await self.client.responses.create( + model=self.model, + input=prompt, + **self.sampling_params, + ) + policy_output = self.extract_output_from_response(resp) + if pattern.fullmatch(policy_output.content): + return policy_output + prompt = [ + *prompt, + { + "role": "user", + "content": ( + f"Invalid response format. Expected format (regex): {regex}\n Please try again and provide ONLY a response that matches this regex." + ), + }, + ] + return policy_output + + # Simple, unconstrained generation + resp = await self.client.responses.create( + model=self.model, + input=prompt, + **self.sampling_params, + ) + policy_output = self.extract_output_from_response(resp) + return policy_output + + def shutdown(self) -> None: + self.client = None diff --git a/src_code_for_reproducibility/models/large_language_model_local.py b/src_code_for_reproducibility/models/large_language_model_local.py new file mode 100644 index 0000000000000000000000000000000000000000..4475b51d8fa895895df6d6dcaff4ed430cda74ca --- /dev/null +++ b/src_code_for_reproducibility/models/large_language_model_local.py @@ -0,0 +1,361 @@ +""" +File: mllm/models/large_language_model_local.py +Summary: Provides a local large language model wrapper over inference backends. +""" + +import logging +import os +import re +import sys +import uuid +from collections.abc import Callable +from copy import deepcopy +from datetime import datetime +from typing import Literal + +import httpx +import requests +import torch +import torch.nn as nn +from torch.optim import SGD, Adam, AdamW, RMSprop +from transformers import AutoModelForCausalLM, AutoTokenizer + +from mllm.chat_utils.apply_template import chat_turns_to_token_ids +from mllm.markov_games.rollout_tree import ChatTurn +from mllm.models.adapter_training_wrapper import AdapterWrapper +from mllm.models.inference_backend import LLMInferenceOutput +from mllm.models.inference_backend_dummy import DummyInferenceBackend +from mllm.models.inference_backend_vllm import VLLMAsyncBackend + +logger = logging.getLogger(__name__) +logger.addHandler(logging.StreamHandler(sys.stdout)) + +AdapterID = str +PolicyID = str + + +class LeanLocalLLM: + """ + Wrapper that manages local HuggingFace models, adapters, and inference backends. + """ + + def __init__( + self, + llm_id: str = "base_llm", + model_name: str = "Qwen/Qwen3-4B-Instruct-2507", + device: str = "cuda", + hf_kwargs: dict = {}, + adapter_configs: dict = {}, + output_directory: str = "./models/", + inference_backend: Literal["vllm", "dummy"] = "vllm", + inference_backend_sampling_params: dict = {}, + inference_backend_init_kwargs: dict = {}, + initial_adapter_paths: dict[str, str] | None = None, + initial_buffer_paths: list[str] | None = None, + enable_thinking: bool = None, + regex_max_attempts: int = -1, + max_thinking_characters: int = 0, + ): + self.inference_backend_name = inference_backend + self.output_directory = output_directory + self.llm_id = llm_id + self.device = torch.device(device) if device else torch.device("cuda") + self.model_name = model_name + self.adapter_configs = adapter_configs + self.adapter_ids = list(adapter_configs.keys()) + self.enable_thinking = enable_thinking + self.regex_max_attempts = regex_max_attempts + self.initial_buffer_paths = initial_buffer_paths + self.max_thinking_characters = max_thinking_characters + self.regex_retries_count = 0 + + # Optional user-specified initial adapter weight locations (local or HF Hub) + # Format: {adapter_id: path_or_repo_id} + self.initial_adapter_paths: dict[str, str] | None = initial_adapter_paths + + # Path management / imports + self.save_path = str(os.path.join(output_directory, model_name, "adapters")) + self.adapter_paths = { + adapter_id: os.path.join(self.save_path, adapter_id) + for adapter_id in self.adapter_ids + } + checkpoints_dir = os.path.join(self.output_directory, "checkpoints") + self.past_agent_adapter_paths = {} + if os.path.isdir(checkpoints_dir): + for dirname in os.listdir(checkpoints_dir): + dirpath = os.path.join(checkpoints_dir, dirname) + if os.path.isdir(dirpath): + self.past_agent_adapter_paths[f"{dirname}_buffer"] = os.path.join( + dirpath, "agent_adapter" + ) + logger.info( + f"Loaded {len(self.past_agent_adapter_paths)} past agent adapters from checkpoints directory." + ) + if self.initial_buffer_paths is not None: + previous_count = len(self.past_agent_adapter_paths) + for path in self.initial_buffer_paths: + if os.path.isdir(path): + for dirname in os.listdir(path): + dirpath = os.path.join(path, dirname) + if os.path.isdir(dirpath): + self.past_agent_adapter_paths[ + f"{dirname}_buffer" + ] = os.path.join(dirpath, "agent_adapter") + else: + logger.warning( + f"Initial buffer path {path} does not exist or is not a directory." + ) + logger.info( + f"Loaded {len(self.past_agent_adapter_paths) - previous_count} past agent adapters from user-specified initial buffer paths." + ) + self.past_agent_adapter_ids = list(self.past_agent_adapter_paths.keys()) + + # ID management for tracking adapter versions + self.adapter_train_ids = { + adapter_id: self.short_id_generator() for adapter_id in self.adapter_ids + } + # Initialize tokenizer + self.tokenizer = AutoTokenizer.from_pretrained(self.model_name) + # Setup padding token to be same as EOS token + self.tokenizer.pad_token_id = self.tokenizer.eos_token_id + self.tokenizer.pad_token = self.tokenizer.eos_token + + self.weights_got_updated: dict[AdapterID, bool] = { + adapter_id: False for adapter_id in self.adapter_ids + } + self.weights_got_updated.update( + {adapter_id: False for adapter_id in self.past_agent_adapter_ids} + ) + self.current_lora_request = None + self.currently_loaded_adapter_id = None + + # --------------------------------------------------------- + # Init HF model, peft adapters + # --------------------------------------------------------- + self.shared_hf_llm = AutoModelForCausalLM.from_pretrained( + pretrained_model_name_or_path=model_name, + **hf_kwargs, + ) + self.hf_adapters = {} + self.optimizers = {} + for adapter_id in self.adapter_ids: + # Prefer output-folder path if it exists; else fall back to user-specified initial path if provided + output_path = os.path.join(self.save_path, adapter_id) + chosen_path: str | None = None + if os.path.isdir(output_path) and os.listdir(output_path): + chosen_path = output_path + logger.info( + f"Initializing adapter '{adapter_id}': using existing weights from output folder '{chosen_path}'." + ) + elif ( + self.initial_adapter_paths and adapter_id in self.initial_adapter_paths + ): + chosen_path = self.initial_adapter_paths[adapter_id] + logger.info( + f"Initializing adapter '{adapter_id}': using provided initial path '{chosen_path}'." + ) + else: + logger.info( + f"Initializing adapter '{adapter_id}': no initial weights provided or found; starting from scratch." + ) + hf_adapter = AdapterWrapper( + shared_llm=self.shared_hf_llm, + adapter_id=adapter_id, + lora_config=adapter_configs[adapter_id], + path=chosen_path, + ).to(device) + self.hf_adapters[adapter_id] = hf_adapter + # Persist current state of all adapters (ensures remote loads are cached to disk) + self.export_adapters() + + # --------------------------------------------------------- + # Init inference inference_backend + # --------------------------------------------------------- + + if inference_backend == "vllm": + self.inference_backend = VLLMAsyncBackend( + model_name=self.model_name, + # adapter_paths=self.adapter_paths, + tokenizer=self.tokenizer, + engine_init_kwargs=inference_backend_init_kwargs, + sampling_params=inference_backend_sampling_params, + ) + elif inference_backend == "dummy": + self.inference_backend = DummyInferenceBackend() + else: + raise ValueError(f"Unknown inference_backend: {inference_backend}") + + def reset_regex_retries_count(self) -> None: + self.regex_retries_count = 0 + + def get_inference_policies(self) -> dict[PolicyID, Callable]: + """ + Build async policy callables keyed by adapter id for inference-only usage. + """ + policies = {} + for adapter_id in self.adapter_ids: + # define policy func + async def policy( + state: list[ChatTurn], + agent_id: str, + regex: str | None = None, + _adapter_id=adapter_id, + ): + self.prepare_adapter_for_inference(adapter_id=_adapter_id) + response = await self.get_action(state, agent_id, regex) + return response + + policies[self.llm_id + "/" + adapter_id] = policy + + for adapter_id in self.past_agent_adapter_ids: + # define policy func + async def policy( + state: list[ChatTurn], + agent_id: str, + regex: str | None = None, + _adapter_id=adapter_id, + ): + self.prepare_adapter_for_inference(adapter_id=_adapter_id) + response = await self.get_action(state, agent_id, regex) + return response + + policies[self.llm_id + "/" + adapter_id] = policy + return policies + + def get_adapter_modules(self) -> dict[PolicyID, nn.Module]: + """ + Returns wrappers over the adapters which allows them be + interfaced like regular PyTorch models. + AdapterWrapper lives in adapter_wrapper.py; the huggingface modules already wrap + parameters here, so we surface them directly until an extra shim is required. + """ + trainable_objects = {an: self.hf_adapters[an] for an in self.adapter_ids} + return trainable_objects + + async def toggle_training_mode(self) -> None: + for adn in self.adapter_ids: + self.adapter_train_ids[adn] = self.short_id_generator() + await self.inference_backend.toggle_training_mode() + + async def toggle_eval_mode(self) -> None: + await self.inference_backend.toggle_eval_mode() + + def prepare_adapter_for_inference(self, adapter_id: AdapterID) -> None: + self.inference_backend.prepare_adapter( + adapter_id, + adapter_path=self.adapter_paths.get( + adapter_id, self.past_agent_adapter_paths.get(adapter_id, None) + ), + weights_got_updated=self.weights_got_updated[adapter_id], + ) + self.currently_loaded_adapter_id = adapter_id + self.weights_got_updated[adapter_id] = False + + # def _make_prompt_text(self, prompt: list[dict]) -> str: + # if self.enable_thinking is not None: + # prompt_text = self.tokenizer.apply_chat_template( + # prompt, + # tokenize=False, + # add_generation_prompt=True, + # enable_thinking=self.enable_thinking, + # ) + # else: + # prompt_text = self.tokenizer.apply_chat_template( + # prompt, + # tokenize=False, + # add_generation_prompt=True, + # ) + + # return prompt_text + + async def get_action( + self, state: list[ChatTurn], agent_id: str, regex: str | None = None + ) -> ChatTurn: + current_regex = regex if self.regex_max_attempts == -1 else None + pattern = re.compile(regex) if regex else None + nb_attempts = 0 + state = state[:] + while True: + context_token_ids = chat_turns_to_token_ids( + chats=state, + tokenizer=self.tokenizer, + enable_thinking=self.enable_thinking, + ) + policy_output = await self.inference_backend.generate( + input_token_ids=context_token_ids.tolist(), + extract_thinking=(self.max_thinking_characters > 0), + regex=current_regex, + ) + if ( + pattern is None + or (pattern.fullmatch(policy_output.content)) + or (nb_attempts >= self.regex_max_attempts) + ): + return ChatTurn( + agent_id=agent_id, + role="assistant", + content=policy_output.content, + reasoning_content=policy_output.reasoning_content, + out_token_ids=policy_output.out_token_ids, + log_probs=policy_output.log_probs, + is_state_end=False, + ) + else: + self.regex_retries_count += 1 + nb_attempts += 1 + logger.warning( + f"Response {policy_output.content} did not match regex: {regex}, retry {nb_attempts}/{self.regex_max_attempts}" + ) + if nb_attempts == self.regex_max_attempts: + current_regex = regex + # regex_prompt = ChatTurn( + # role="user", + # content=f"Invalid response format. Expected format (regex): {current_regex}\n Please try again and provide ONLY a response that matches this regex.", + # reasoning_content=None, + # log_probs=None, + # out_token_ids=None, + # is_state_end=False, + # ) + # state.append(regex_prompt) + + def export_adapters(self) -> None: + """ + Any peft wrapper, by default, saves all adapters, not just the one currently loaded. + """ + + # New version of the adapters available + for adapter_id in self.adapter_ids: + self.weights_got_updated[adapter_id] = True + for adapter_id in self.past_agent_adapter_ids: + self.weights_got_updated[adapter_id] = True + + adapter_id = self.adapter_ids[0] + self.hf_adapters[adapter_id].save_pretrained(self.save_path) + + def checkpoint_all_adapters(self, checkpoint_indicator: str) -> None: + """ + Checkpoints all adapters to the configured output directory. + """ + adapter_id = self.adapter_ids[0] + output_dir = os.path.join(self.output_directory, "checkpoints") + os.makedirs(output_dir, exist_ok=True) + date_str = datetime.now().strftime("%Y-%m-%d_%H-%M-%S") + agent_adapter_dir = f"{adapter_id}-{checkpoint_indicator}-{date_str}" + export_path = os.path.join(output_dir, agent_adapter_dir) + for adapter_id in self.adapter_ids: + if "agent" in adapter_id: + self.past_agent_adapter_paths[ + f"{agent_adapter_dir}_buffer" + ] = os.path.join(export_path, adapter_id) + self.past_agent_adapter_ids.append(f"{agent_adapter_dir}_buffer") + self.weights_got_updated[f"{agent_adapter_dir}_buffer"] = False + self.hf_adapters[adapter_id].save_pretrained(export_path) + + def short_id_generator(self) -> str: + """ + Generates a short unique ID for tracking adapter versions. + + Returns: + int: An 8-digit integer ID. + """ + return str(uuid.uuid4().int)[:8] diff --git a/src_code_for_reproducibility/models/scalar_critic.py b/src_code_for_reproducibility/models/scalar_critic.py new file mode 100644 index 0000000000000000000000000000000000000000..0b704dcc78fdfbed1c68b1ac469e9c7b51758211 --- /dev/null +++ b/src_code_for_reproducibility/models/scalar_critic.py @@ -0,0 +1,59 @@ +""" +File: mllm/models/scalar_critic.py +Summary: Defines a scalar critic network and helper utilities. +""" + +import torch +import torch.nn as nn +import torch.optim as optim +from peft import LoraConfig, get_peft_model +from transformers import AutoModelForCausalLM, AutoTokenizer + +from mllm.models.adapter_training_wrapper import AdapterWrapper + + +class ScalarCritic(nn.Module): + """ + A causal-LM critic_adapter + a scalar value head: + V_φ(s) = wᵀ h_last + b + Only LoRA adapters (inside critic_adapter) and the value head are trainable. + """ + + def __init__(self, critic_adapter: AdapterWrapper): + super().__init__() + self.critic_adapter = critic_adapter + hidden_size = self.critic_adapter.shared_llm.config.hidden_size + self.value_head = nn.Linear(hidden_size, 1).to( + dtype=critic_adapter.dtype, device=critic_adapter.device + ) + + def forward(self, input_ids, attention_mask=None, **kwargs): + # AdapterWrapper activates its own adapter internally + outputs = self.critic_adapter( + input_ids=input_ids, + attention_mask=attention_mask, + output_hidden_states=True, + **kwargs, + ) + h_last = outputs.hidden_states[-1] # (B, S, H) + values = self.value_head(h_last).squeeze(-1) # (B, S) + return values + + def parameters(self, recurse: bool = True): + """Iterator over *trainable* parameters for this critic.""" + # 1) LoRA params for *this* adapter + for p in self.critic_adapter.parameters(): + yield p + # 2) scalar head + yield from self.value_head.parameters() + + def gradient_checkpointing_enable(self, *args, **kwargs): + self.critic_adapter.gradient_checkpointing_enable(*args, **kwargs) + + @property + def dtype(self): + return self.critic_adapter.dtype + + @property + def device(self): + return self.critic_adapter.device diff --git a/src_code_for_reproducibility/training/__init__.py b/src_code_for_reproducibility/training/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..188fde562de5d8f658ef12708df9967f45cb2a7a --- /dev/null +++ b/src_code_for_reproducibility/training/__init__.py @@ -0,0 +1,4 @@ +""" +File: mllm/training/__init__.py +Summary: Exposes training submodules through the package namespace. +""" diff --git a/src_code_for_reproducibility/training/__pycache__/__init__.cpython-312.pyc b/src_code_for_reproducibility/training/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..eb3d8d4bee48323b7b756ccf91dc2a191e067e3b Binary files /dev/null and b/src_code_for_reproducibility/training/__pycache__/__init__.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/training/__pycache__/credit_methods.cpython-312.pyc b/src_code_for_reproducibility/training/__pycache__/credit_methods.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..46eae35640116412e52dc6fbbc7f7038c07dd49d Binary files /dev/null and b/src_code_for_reproducibility/training/__pycache__/credit_methods.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/training/__pycache__/tally_rollout.cpython-312.pyc b/src_code_for_reproducibility/training/__pycache__/tally_rollout.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..224f9e383d51bd946401cd4aa01919574b577bad Binary files /dev/null and b/src_code_for_reproducibility/training/__pycache__/tally_rollout.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/training/__pycache__/tally_tokenwise.cpython-312.pyc b/src_code_for_reproducibility/training/__pycache__/tally_tokenwise.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..31df01b89cad4eab566667f599501f06d824190e Binary files /dev/null and b/src_code_for_reproducibility/training/__pycache__/tally_tokenwise.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/training/__pycache__/trainer_common.cpython-312.pyc b/src_code_for_reproducibility/training/__pycache__/trainer_common.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..12a65f20e21690751f077533e91cb7f4fc300f14 Binary files /dev/null and b/src_code_for_reproducibility/training/__pycache__/trainer_common.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/training/__pycache__/trainer_independent.cpython-312.pyc b/src_code_for_reproducibility/training/__pycache__/trainer_independent.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..854620cf63cea8ec419bcb2a5f63788abcb01bf6 Binary files /dev/null and b/src_code_for_reproducibility/training/__pycache__/trainer_independent.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/training/__pycache__/training_data_utils.cpython-312.pyc b/src_code_for_reproducibility/training/__pycache__/training_data_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6b5a388e13e83425d7cf9aa370b5715d8b01701e Binary files /dev/null and b/src_code_for_reproducibility/training/__pycache__/training_data_utils.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/training/annealing_methods.py b/src_code_for_reproducibility/training/annealing_methods.py new file mode 100644 index 0000000000000000000000000000000000000000..591d91f7720880fc202b116b27b15b996c256dc4 --- /dev/null +++ b/src_code_for_reproducibility/training/annealing_methods.py @@ -0,0 +1,20 @@ +""" +File: mllm/training/annealing_methods.py +Summary: Implements annealing schedules used across training loops. +""" + +import numpy as np + + +def sigmoid_annealing(step: int, temperature: float) -> float: + """ + Smoothly ramp a scalar from 0 → 1 using a temperature-controlled sigmoid. + + Args: + step: Current training step or iteration. + temperature: Controls how sharp the transition is; larger values flatten the curve. + + Returns: + Float in [-1, 1] that can be rescaled for annealing schedules. + """ + return 2 / (1 + np.exp(-step / temperature)) - 1 diff --git a/src_code_for_reproducibility/training/credit_methods.py b/src_code_for_reproducibility/training/credit_methods.py new file mode 100644 index 0000000000000000000000000000000000000000..c29032630d06536e7efe6ceaae04092a616ce53a --- /dev/null +++ b/src_code_for_reproducibility/training/credit_methods.py @@ -0,0 +1,307 @@ +""" +File: mllm/training/credit_methods.py +Summary: Holds credit-assignment routines for reinforcement learning updates. +""" + +import torch + + +def whiten_advantages(advantages: torch.Tensor) -> torch.Tensor: + """ + Normalize a vector of advantages to zero mean / unit variance (global). + + Useful for variance reduction before computing gradients. + """ + whitened_advantages = (advantages - torch.mean(advantages)) / ( + torch.std(advantages) + 1e-9 + ) + return whitened_advantages + + +def whiten_advantages_time_step_wise( + advantages: torch.Tensor, # (B, T) +) -> torch.Tensor: + """ + Whiten advantages independently per timestep (column-wise mean/std). + + Helps when rollout lengths differ or certain positions have very different scales. + """ + assert advantages.dim() == 2, "Wrong dimensions." + whitened_advantages_time_step_wise = ( + advantages - advantages.mean(dim=0, keepdim=True) + ) / (advantages.std(dim=0, keepdim=True) + 1e-9) + return whitened_advantages_time_step_wise + + +def get_discounted_state_visitation_credits( + credits: torch.Tensor, discount_factor: float # (B, T) +) -> torch.Tensor: + """ + Apply geometric discounting to credits so earlier visits count less. + + Equivalent to per-timestep multiplication by ``gamma^t``. + """ + return credits * ( + discount_factor ** torch.arange(credits.shape[1], device=credits.device) + ) + + +def get_discounted_returns( + rewards: torch.Tensor, # (B, T) + discount_factor: float, +) -> torch.Tensor: + """ + Computes Monte Carlo discounted returns for a sequence of rewards. + + Args: + rewards (torch.Tensor): Array of rewards for each timestep. + + Returns: + torch.Tensor: Array of discounted returns. + """ + assert rewards.dim() == 2, "Wrong dimensions." + B, T = rewards.shape + discounted_returns = torch.zeros_like(rewards) + accumulator = torch.zeros(B, device=rewards.device, dtype=rewards.dtype) + for t in reversed(range(T)): + accumulator = rewards[:, t] + discount_factor * accumulator + discounted_returns[:, t] = accumulator + return discounted_returns + + +def get_rloo_credits(credits: torch.Tensor): # (B, S) + """Compute leave-one-out baselines for a batch of credits.""" + assert credits.dim() == 2, "Wrong dimensions." + rloo_baselines = torch.zeros_like(credits) + n = credits.shape[0] + if n == 1: + return credits, rloo_baselines + rloo_baselines = (torch.sum(credits, dim=0, keepdim=True) - credits) / (n - 1) + rloo_credits = credits - rloo_baselines + return rloo_credits, rloo_baselines + + +def get_generalized_advantage_estimates( + rewards: torch.Tensor, # (B, T) + value_estimates: torch.Tensor, # (B, T+1) + discount_factor: float, + lambda_coef: float, +) -> torch.Tensor: + """ + Compute Generalized Advantage Estimates (GAE). + + See https://arxiv.org/pdf/1506.02438 for derivation. + """ + assert rewards.dim() == value_estimates.dim() == 2, "Wrong dimensions." + + assert ( + rewards.shape[0] == value_estimates.shape[0] + ), f"Got shapes {rewards.shape} and {value_estimates.shape} of rewards and value estimates." + assert ( + rewards.shape[1] == value_estimates.shape[1] - 1 + ), f"Got shapes {rewards.shape} and {value_estimates.shape} of rewards and value estimates." + + T = rewards.shape[1] + tds = rewards + discount_factor * value_estimates[:, 1:] - value_estimates[:, :-1] + gaes = torch.zeros_like(tds) + acc = 0.0 + for t in reversed(range(T)): + acc = tds[:, t] + lambda_coef * discount_factor * acc + gaes[:, t] = acc + return gaes + + +def get_advantage_alignment_weights( + advantages: torch.Tensor, # (B, T) + exclude_k_equals_t: bool, + gamma: float, + discount_t: bool, +) -> torch.Tensor: + """ + The advantage alignment credit is calculated as + + \[ + A^*(s_t, a_t, b_t) = A^1(s_t, a_t, b_t) + \beta \cdot + \left( \sum_{k < t} \gamma^{t-k} A^1(s_k, a_k, b_k) \right) + A^2(s_t, a_t, b_t) + \] + + Here, the weights are defined as \( \beta \cdot + \left( \sum_{k < t} \gamma^{t-k} A^1(s_k, a_k, b_k) \) + """ + T = advantages.shape[1] + discounted_advantages = advantages * ( + gamma * torch.ones((1, T), device=advantages.device) + ) ** (-torch.arange(0, T, 1, device=advantages.device)) + if exclude_k_equals_t: + sub = torch.eye(T, device=advantages.device) + else: + sub = torch.zeros((T, T), device=advantages.device) + # Identity is for \( k < t \), remove for \( k \leq t \) + ad_align_weights = discounted_advantages @ ( + torch.triu(torch.ones((T, T), device=advantages.device)) - sub + ) + t_discounts = (gamma * torch.ones((1, T), device=advantages.device)) ** ( + torch.arange(0, T, 1, device=advantages.device) + ) + ad_align_weights = t_discounts * ad_align_weights + if discount_t: + time_discounted_advantages = advantages * ( + gamma * torch.ones((1, T), device=advantages.device) + ) ** (torch.arange(0, T, 1, device=advantages.device)) + ad_align_weights = ad_align_weights - advantages + time_discounted_advantages + return ad_align_weights + + +def get_advantage_alignment_credits( + a1: torch.Tensor, # (B, S) + a1_alternative: torch.Tensor, # (B, S, A) + a2: torch.Tensor, # (B, S) + exclude_k_equals_t: bool, + beta: float, + gamma: float = 1.0, + use_old_ad_align: bool = False, + use_sign: bool = False, + clipping: float | None = None, + use_time_regularization: bool = False, + force_coop_first_step: bool = False, + use_variance_regularization: bool = False, + rloo_branch: bool = False, + reuse_baseline: bool = False, + mean_normalize_ad_align: bool = False, + whiten_adalign_advantages: bool = False, + whiten_adalign_advantages_time_step_wise: bool = False, + discount_t: bool = False, +) -> torch.Tensor: + """ + Calculate the advantage alignment credits with vectorization, as described in https://arxiv.org/abs/2406.14662. + + Recall that the advantage opponent shaping term of the AdAlign policy gradient is: + \[ + \beta \mathbb{E}_{\substack{ + \tau \sim \text{Pr}_{\mu}^{\pi^1, \pi^2} \\ + a_t' \sim \pi^1(\cdot \mid s_t) + }} + \left[\sum_{t=0}^\infty \gamma^{t}\left( \sum_{k\leq t} A^1(s_k,a^{\prime}_k,b_k) \right) A^{2}(s_t,a_t, b_t)\nabla_{\theta^1}\text{log } \pi^1(a_t|s_t) \right] + \] + + This method computes the following: + \[ + Credit(s_t, a_t, b_t) = \gamma^t \left[ A^1(s_t, a_t, b_t) + \beta \left( \sum_{k\leq t} A^1(s_k,a^{\prime}_k,b_k) \right) A^{2}(s_t,a_t, b_t) \right] + \] + + Args: + a1: Advantages of the main trajectories for the current agent. + a1_alternative: Advantages of the alternative trajectories for the current agent. + a2: Advantages of the main trajectories for the other agent. + discount_factor: Discount factor for the advantage alignment. + beta: Beta parameter for the advantage alignment. + gamma: Gamma parameter for the advantage alignment. + use_sign_in_ad_align: Whether to use sign in the advantage alignment. + + Returns: + torch.Tensor: The advantage alignment credits. + """ + + assert a1.dim() == a2.dim() == 2, "Advantages must be of shape (B, S)" + if a1_alternative is not None: + assert ( + a1_alternative.dim() == 3 + ), "Alternative advantages must be of shape (B, S, A)" + B, T, A = a1_alternative.shape + else: + B, T = a1.shape + assert a1.shape == a2.shape, "Not the same shape" + + sub_tensors = {} + + if use_old_ad_align: + ad_align_weights = get_advantage_alignment_weights( + advantages=a1, + exclude_k_equals_t=exclude_k_equals_t, + gamma=gamma, + discount_t=discount_t, + ) + sub_tensors["ad_align_weights_prev"] = ad_align_weights + if exclude_k_equals_t: + ad_align_weights = gamma * ad_align_weights + else: + assert a1_alternative is not None, "Alternative advantages must be provided" + if rloo_branch: + a1_alternative = torch.cat([a1.unsqueeze(2), a1_alternative], dim=2) + a1_alternative = a1_alternative.mean(dim=2) + a1, baseline = get_rloo_credits(a1) + if reuse_baseline: + a1_alternative = a1_alternative - baseline + else: + a1_alternative, _ = get_rloo_credits(a1_alternative) + assert a1.shape == a1_alternative.shape, "Not the same shape" + ad_align_weights = get_advantage_alignment_weights( + advantages=a1_alternative, + exclude_k_equals_t=exclude_k_equals_t, + gamma=gamma, + ) + sub_tensors["ad_align_weights"] = ad_align_weights + + # Use sign + if use_sign: + assert beta == 1.0, "beta should be 1.0 when using sign" + positive_signs = ad_align_weights > 0 + negative_signs = ad_align_weights < 0 + ad_align_weights[positive_signs] = 1 + ad_align_weights[negative_signs] = -1 + sub_tensors["ad_align_weights_sign"] = ad_align_weights + # (rest are 0) + + ################### + # Process weights + ################### + + # Use clipping + if clipping not in [0.0, None]: + upper_mask = ad_align_weights > 1 + lower_mask = ad_align_weights < -1 + + ad_align_weights = torch.clip( + ad_align_weights, + -clipping, + clipping, + ) + clipping_ratio = ( + torch.sum(upper_mask) + torch.sum(lower_mask) + ) / upper_mask.size + sub_tensors["clipped_ad_align_weights"] = ad_align_weights + + # 1/1+t Regularization + if use_time_regularization: + t_values = torch.arange(1, T + 1).to(ad_align_weights.device) + ad_align_weights = ad_align_weights / t_values + sub_tensors["time_regularized_ad_align_weights"] = ad_align_weights + + # Use coop on t=0 + if force_coop_first_step: + ad_align_weights[:, 0] = 1 + sub_tensors["coop_first_step_ad_align_weights"] = ad_align_weights + + #################################### + # Compose elements together + #################################### + + opp_shaping_terms = beta * ad_align_weights * a2 + sub_tensors["ad_align_opp_shaping_terms"] = opp_shaping_terms + + credits = a1 + opp_shaping_terms + if mean_normalize_ad_align: + credits = credits - credits.mean(dim=0) + sub_tensors["mean_normalized_ad_align_credits"] = credits + if whiten_adalign_advantages: + credits = (credits - credits.mean()) / (credits.std() + 1e-9) + sub_tensors["whitened_ad_align_credits"] = credits + if whiten_adalign_advantages_time_step_wise: + credits = (credits - credits.mean(dim=0, keepdim=True)) / ( + credits.std(dim=0, keepdim=True) + 1e-9 + ) + sub_tensors["whitened_ad_align_credits_time_step_wise"] = credits + sub_tensors["final_ad_align_credits"] = credits + + return credits, sub_tensors diff --git a/src_code_for_reproducibility/training/tally_metrics.py b/src_code_for_reproducibility/training/tally_metrics.py new file mode 100644 index 0000000000000000000000000000000000000000..a0073ab3be86bbaf448b6a898341a4f5a3087b24 --- /dev/null +++ b/src_code_for_reproducibility/training/tally_metrics.py @@ -0,0 +1,64 @@ +""" +File: mllm/training/tally_metrics.py +Summary: Transforms tally files into aggregated metric summaries. +""" + +import os +from numbers import Number +from typing import Union + +import wandb + + +class Tally: + """ + Minimal scalar-first tally. + - Keys are strings. + - First add stores a scalar; subsequent adds upgrade to a list of scalars. + """ + + def __init__(self): + self.stats = {} + + def reset(self): + """Reset all recorded metrics back to an empty dictionary.""" + self.stats = {} + + def _coerce_scalar(self, value: Union[int, float]) -> Union[int, float]: + """Ensure ``value`` is a plain Python scalar (detach tensors, etc.).""" + if hasattr(value, "item") and callable(getattr(value, "item")): + try: + value = value.item() + except Exception: + pass + if isinstance(value, Number): + return value + raise AssertionError("Metric must be a scalar number") + + def add_metric(self, path: str, metric: Union[int, float]): + """Accumulate a metric under ``path`` (scalar on first add, list thereafter).""" + metric = float(metric) + assert isinstance(path, str), "Path must be a string." + assert isinstance(metric, float), "Metric must be a scalar number." + + scalar = self._coerce_scalar(metric) + existing = self.stats.get(path) + if existing is None: + self.stats[path] = scalar + elif isinstance(existing, list): + existing.append(scalar) + else: + self.stats[path] = [existing, scalar] + + def save(self, identifier: str, folder: str): + """Persist the tally as a pickle file under ``folder``.""" + os.makedirs(name=folder, exist_ok=True) + try: + import pickle + + pkl_path = os.path.join(folder, f"{identifier}.tally.pkl") + payload = self.stats + with open(pkl_path, "wb") as f: + pickle.dump(payload, f, protocol=pickle.HIGHEST_PROTOCOL) + except Exception: + pass diff --git a/src_code_for_reproducibility/training/tally_rollout.py b/src_code_for_reproducibility/training/tally_rollout.py new file mode 100644 index 0000000000000000000000000000000000000000..04bb4f36d7f1c6759c3fb0f0102f29b117ea57c1 --- /dev/null +++ b/src_code_for_reproducibility/training/tally_rollout.py @@ -0,0 +1,116 @@ +""" +File: mllm/training/tally_rollout.py +Summary: Serializes rollout data into tallies for downstream processing. +""" + +import json +import os +from copy import deepcopy +from typing import Union + +import numpy as np +import pandas as pd +import torch +from transformers import AutoTokenizer + + +class RolloutTallyItem: + def __init__( + self, + crn_ids: list[str], + rollout_ids: list[str], + agent_ids: list[str], + metric_matrix: torch.Tensor, + ): + """Lightweight data container that keeps rollout-aligned metric matrices.""" + if isinstance(crn_ids, torch.Tensor): + crn_ids = crn_ids.detach().cpu().numpy() + if isinstance(rollout_ids, torch.Tensor): + rollout_ids = rollout_ids.detach().cpu().numpy() + if isinstance(agent_ids, torch.Tensor): + agent_ids = agent_ids.detach().cpu().numpy() + self.crn_ids = crn_ids + self.rollout_ids = rollout_ids + self.agent_ids = agent_ids + metric_matrix = metric_matrix.detach().cpu() + assert ( + 0 < metric_matrix.ndim <= 2 + ), "Metric matrix must have less than or equal to 2 dimensions" + if metric_matrix.ndim == 1: + metric_matrix = metric_matrix.reshape(1, -1) + # Convert to float32 if tensor is in BFloat16 format (not supported by numpy) + if metric_matrix.dtype == torch.bfloat16: + metric_matrix = metric_matrix.float() + self.metric_matrix = metric_matrix.numpy() + + +class RolloutTally: + """ + Tally is a utility class for collecting and storing training metrics. + It supports adding metrics at specified paths and saving them to disk. + """ + + def __init__(self): + """ + Initializes the RolloutTally object. + + Args: + tokenizer (AutoTokenizer): Tokenizer for converting token IDs to strings. + max_context_length (int, optional): Maximum context length for contextualized metrics. Defaults to 30. + """ + # Array-preserving structure (leaf lists hold numpy arrays / scalars) + self.metrics = {} + # Global ordered list of sample identifiers (crn_id, rollout_id) added in the order samples are processed + + def reset(self): + """Reset the tally to an empty dict.""" + self.metrics = {} + + def get_from_nested_dict(self, dictio: dict, path: str): + """Retrieve a nested entry, creating intermediate dicts as needed.""" + assert isinstance(path, list), "Path must be list." + for sp in path[:-1]: + dictio = dictio.setdefault(sp, {}) + return dictio.get(path[-1], None) + + def set_at_path(self, dictio: dict, path: str, value): + """Store ``value`` at ``path``; helper used by ``add_metric``.""" + for sp in path[:-1]: + dictio = dictio.setdefault(sp, {}) + dictio[path[-1]] = value + + def add_metric(self, path: list[str], rollout_tally_item: RolloutTallyItem): + """ + Adds a metric to the base tally at the specified path. + + Args: + path (list): List of keys representing the path in the base tally. + rollout_tally_item (RolloutTallyItem): The rollout tally item to add. + """ + rollout_tally_item = deepcopy(rollout_tally_item) + + # Update array-preserving tally + array_list = self.get_from_nested_dict(dictio=self.metrics, path=path) + if array_list is None: + self.set_at_path(dictio=self.metrics, path=path, value=[rollout_tally_item]) + else: + array_list.append(rollout_tally_item) + + def save(self, identifier: str, folder: str): + """Persist the tally as a pickle (metrics only) under ``folder``.""" + os.makedirs(name=folder, exist_ok=True) + + from datetime import datetime + + now = datetime.now() + + # Pickle only (fastest, exact structure with numpy/scalars at leaves) + try: + import pickle + + pkl_path = os.path.join(folder, f"{identifier}.rt_tally.pkl") + payload = {"metrics": self.metrics} + with open(pkl_path, "wb") as f: + pickle.dump(payload, f, protocol=pickle.HIGHEST_PROTOCOL) + except Exception: + pass diff --git a/src_code_for_reproducibility/training/tally_tokenwise.py b/src_code_for_reproducibility/training/tally_tokenwise.py new file mode 100644 index 0000000000000000000000000000000000000000..b7770e0cb79d5ed4e56a3f66b6982582c72e0bb7 --- /dev/null +++ b/src_code_for_reproducibility/training/tally_tokenwise.py @@ -0,0 +1,278 @@ +""" +File: mllm/training/tally_tokenwise.py +Summary: Converts token-level tallies into per-token statistics. +""" + +import json +import os +from typing import Any, Dict, List, Tuple, Union + +import numpy as np +import pandas as pd +import torch +from transformers import AutoTokenizer + + +class ContextualizedTokenwiseTally: + """ + Collect, store, and save token-level metrics per rollout. + + - One DataFrame per rollout_id in `paths` + - Index = timestep (int) + - Columns are added incrementally via `add_contexts()` and `add_data()` + - Cells may contain scalars, strings, or lists (dtype=object) + """ + + def __init__( + self, + tokenizer: AutoTokenizer, + paths: List[str], + max_context_length: int = 30, + ): + """ + Args: + tokenizer: HuggingFace tokenizer used to convert tids -> tokens + paths: rollout identifiers (parallel to batch dimension) + max_context_length: truncate context token lists to this length + """ + self.tokenizer = tokenizer + self.paths = paths + self.max_context_length = max_context_length + self.tally: Dict[str, pd.DataFrame] = {path: pd.DataFrame() for path in paths} + + # set later by setters + self.contexts: torch.Tensor | None = None + self.action_mask: torch.Tensor | None = None + self.range: Tuple[int, int] | None = None + + # --------- Utilities --------- + + def tids_to_str(self, tids: List[int]) -> List[str]: + """Convert a list of token IDs to a list of token strings.""" + return self.tokenizer.convert_ids_to_tokens(tids) + + def _ensure_ready(self): + """Validate that action mask and range are configured prior to writes.""" + assert self.action_mask is not None, "call set_action_mask(mask) first" + assert self.range is not None, "call set_range((start, end)) first" + + @staticmethod + def _sanitize_filename(name: Any) -> str: + """Make a safe filename from any rollout_id.""" + s = str(name) + bad = {os.sep, " ", ":", "|", "<", ">", '"', "'"} + if os.altsep is not None: + bad.add(os.altsep) + for ch in bad: + s = s.replace(ch, "_") + return s + + @staticmethod + def _pad_left(seq: List[Any], length: int, pad_val: Any = "") -> List[Any]: + """Left-pad a sequence to `length` with `pad_val`.""" + if len(seq) >= length: + return seq[-length:] + return [pad_val] * (length - len(seq)) + list(seq) + + # --------- Setters --------- + + def set_action_mask(self, action_mask: torch.Tensor): + """Register the (B, S) mask indicating which tokens correspond to actions.""" + self.action_mask = action_mask + + def set_range(self, range: Tuple[int, int]): + """Record which subset of ``paths`` the current mini-batch corresponds to.""" + self.range = range + + # --------- Column builders --------- + + def add_contexts(self, contexts: torch.Tensor): + """ + Add a single 'context' column (list[str]) for valid steps. + + Expects `contexts` with shape (B, S): token id at each timestep. + For each valid timestep t, we use the last N tokens up to and including t: + window = contexts[i, max(0, t - N + 1) : t + 1] + The list is left-padded with "" to always be length N. + """ + self._ensure_ready() + + current_paths = self.paths[self.range[0] : self.range[1]] + B, S = contexts.shape + N = self.max_context_length + + # to CPU ints once + contexts_cpu = contexts.detach().to("cpu") + + for i in range(B): + rollout_id = current_paths[i] + df = self.tally.get(rollout_id, pd.DataFrame()) + + valid_idx = torch.nonzero( + self.action_mask[i].bool(), as_tuple=False + ).squeeze(-1) + if valid_idx.numel() == 0: + self.tally[rollout_id] = df + continue + + idx_list = valid_idx.tolist() + + # ensure index contains valid steps + if df.empty: + df = pd.DataFrame(index=idx_list) + else: + new_index = sorted(set(df.index.tolist()) | set(idx_list)) + if list(df.index) != new_index: + df = df.reindex(new_index) + + # build context windows + ctx_token_lists = [] + for t in idx_list: + start = max(0, t - N + 1) + window_ids = contexts_cpu[i, start : t + 1].tolist() + window_toks = self.tids_to_str([int(x) for x in window_ids]) + if len(window_toks) < N: + window_toks = [""] * (N - len(window_toks)) + window_toks + else: + window_toks = window_toks[-N:] + ctx_token_lists.append(window_toks) + + # single 'context' column + if "context" not in df.columns: + df["context"] = pd.Series(index=df.index, dtype=object) + df.loc[idx_list, "context"] = pd.Series( + ctx_token_lists, index=idx_list, dtype=object + ) + + self.tally[rollout_id] = df + + def add_data( + self, + metric_id: str, + metrics: torch.Tensor, + to_tids: bool = False, + ): + """ + Add a metric column for valid steps. + + Args: + metric_id: column name + metrics: shape (B, S) for scalars/ids or (B, S, K) for top-k vectors + to_tids: if True, treat ints/lists of ints as tids and convert to tokens + """ + self._ensure_ready() + current_paths = self.paths[self.range[0] : self.range[1]] + + if metrics.dim() == 2: + B, S = metrics.shape + elif metrics.dim() == 3: + B, S, _ = metrics.shape + else: + raise ValueError("metrics must be (B, S) or (B, S, K)") + + for i in range(B): + rollout_id = current_paths[i] + df = self.tally.get(rollout_id, pd.DataFrame()) + + valid_idx = torch.nonzero( + self.action_mask[i].bool(), as_tuple=False + ).squeeze(-1) + if valid_idx.numel() == 0: + self.tally[rollout_id] = df + continue + + idx_list = valid_idx.detach().cpu().tolist() + + # Ensure index contains valid steps + if df.empty: + df = pd.DataFrame(index=idx_list) + else: + new_index = sorted(set(df.index.tolist()) | set(idx_list)) + if list(df.index) != new_index: + df = df.reindex(new_index) + + # Slice metrics at valid steps + m_valid = metrics[i][valid_idx] + + # -> pure python lists (1D list or list-of-lists) + values = m_valid.detach().cpu().tolist() + + # optional tids -> tokens + if to_tids: + + def _to_tokish(x): + if isinstance(x, list): + return self.tids_to_str([int(v) for v in x]) + else: + return self.tids_to_str([int(x)])[0] + + values = [_to_tokish(v) for v in values] + + # Ensure column exists with object dtype, then assign via aligned Series + if metric_id not in df.columns: + df[metric_id] = pd.Series(index=df.index, dtype=object) + + if isinstance(values, np.ndarray): + values = values.tolist() + + if len(values) != len(idx_list): + raise ValueError( + f"Length mismatch for '{metric_id}': values={len(values)} vs idx_list={len(idx_list)}" + ) + + df.loc[idx_list, metric_id] = pd.Series( + values, index=idx_list, dtype=object + ) + self.tally[rollout_id] = df + + # --------- Saving --------- + + def save(self, path: str): + """ + Write a manifest JSON and one CSV per rollout. + + - Manifest includes metadata only (safe to JSON). + - Each rollout CSV is written with index label 'timestep'. + - Only a single 'context' column (list[str]). + """ + if not self.tally or all(df.empty for df in self.tally.values()): + return + + os.makedirs(path, exist_ok=True) + from datetime import datetime + + now = datetime.now() + + manifest = { + "created_at": f"{now:%Y-%m-%d %H:%M:%S}", + "max_context_length": self.max_context_length, + "num_rollouts": len(self.tally), + "rollouts": [], + } + + for rid, df in self.tally.items(): + rid_str = str(rid) + safe_name = self._sanitize_filename(rid_str) + csv_path = os.path.join(path, f"{safe_name}_tokenwise.csv") + + # Put 'context' first, then the rest + cols = ["context"] + [c for c in df.columns if c != "context"] + try: + df[cols].to_csv(csv_path, index=True, index_label="timestep") + except Exception as e: + continue + + manifest["rollouts"].append( + { + "rollout_id": rid_str, + "csv": csv_path, + "num_rows": int(df.shape[0]), + "columns": cols, + } + ) + + manifest_path = os.path.join( + path, f"tokenwise_manifest_{now:%Y-%m-%d___%H-%M-%S}.json" + ) + with open(manifest_path, "w") as fp: + json.dump(manifest, fp, indent=2) diff --git a/src_code_for_reproducibility/training/tokenize_chats.py b/src_code_for_reproducibility/training/tokenize_chats.py new file mode 100644 index 0000000000000000000000000000000000000000..94da0030ec2afe19d5e5cd8a9a9e39b595d19975 --- /dev/null +++ b/src_code_for_reproducibility/training/tokenize_chats.py @@ -0,0 +1,128 @@ +""" +File: mllm/training/tokenize_chats.py +Summary: Tokenizes chat datasets and prepares tensors for training. +""" + +import logging +import sys + +import regex +import torch +from transformers import AutoTokenizer + +from mllm.training.training_data_utils import TrainingChatTurn, TrajectoryBatch + +logger = logging.getLogger(__name__) +logger.addHandler(logging.StreamHandler(sys.stdout)) + + +def process_training_chat( + tokenizer: AutoTokenizer, + chat_history: list[TrainingChatTurn], + entropy_mask_regex: str | None = None, + exploration_prompts_to_remove: list[str] = [], + use_engine_out_token_ids: bool = False, +) -> tuple[torch.IntTensor, torch.BoolTensor, torch.IntTensor, torch.BoolTensor]: + """Tokenize a single training chat and build aligned per-token masks. + + Given an ordered list of `TrainingChatTurn`, this function tokenizes each + turn independently using the tokenizer's chat template, then concatenates + all resulting token sequences. It also constructs three parallel 1D masks + that align with the concatenated tokens: + + - input_ids: token ids for the entire chat, turn by turn + - action_mask: True for tokens that belong to assistant turns (i.e., model + actions), False for tokens from other roles + - timesteps: per-token time step copied from the originating turn's + `time_step` + - state_ends_mask: True for the last token of any turn where + `is_state_end` is True, otherwise False + + Important details: + - Each turn is passed as a single-message list to + `tokenizer.apply_chat_template` and flattened; the per-turn outputs are + then concatenated in the original order. + - Turn boundaries are not explicitly encoded beyond what the chat template + inserts; masks provide alignment for learning signals and state endings. + - No truncation or padding is performed here; downstream code should handle + batching/padding as needed. + - Note on dtypes: `input_ids` will be a LongTensor (int64). `action_mask` + and `state_ends_mask` are BoolTensors. `timesteps` is currently created + as a float tensor; adjust the implementation if integer dtype is + required downstream. + + Args: + tokenizer: A Hugging Face tokenizer supporting `apply_chat_template`. + chat_history: Ordered list of `TrainingChatTurn` forming one dialogue. + + Returns: + A tuple of four 1D tensors, all of equal length N (the total number of + tokens across all turns), in the following order: + - input_ids (LongTensor) + - action_mask (BoolTensor) + - timesteps (FloatTensor as implemented; see note above) + - state_ends_mask (BoolTensor) + """ + state_ends_mask = [] + input_ids = [] + action_mask = [] + timesteps = [] + entropy_mask = [] + engine_log_probs = [] + for train_chat_turn in chat_history: + is_state_end = train_chat_turn.is_state_end + time_step = train_chat_turn.time_step + is_action = train_chat_turn.role == "assistant" + + # Remove exploration prompts from training data + for exploration_prompt in exploration_prompts_to_remove: + if exploration_prompt in train_chat_turn.content: + train_chat_turn.content = train_chat_turn.content.replace( + exploration_prompt, "" + ) + + chat_turn = { + "role": train_chat_turn.role, + "content": train_chat_turn.content, + } + if entropy_mask_regex is not None: + is_entropy_mask_true = ( + regex.search(entropy_mask_regex, train_chat_turn.content) is not None + ) + else: + is_entropy_mask_true = True + if is_action: + chat_turn_ids = train_chat_turn.out_token_ids + nb_chat_turns_ids = chat_turn_ids.numel() + action_mask.append(torch.ones(nb_chat_turns_ids, dtype=torch.bool)) + engine_log_probs.append(train_chat_turn.log_probs) + else: + chat_turn_ids = train_chat_turn.chat_template_token_ids + nb_chat_turns_ids = chat_turn_ids.numel() + action_mask.append(torch.zeros(nb_chat_turns_ids, dtype=torch.bool)) + engine_log_probs.append(torch.zeros(nb_chat_turns_ids, dtype=torch.float)) + nb_chat_turns_ids = chat_turn_ids.numel() + state_ends_mask.append(torch.zeros(nb_chat_turns_ids, dtype=torch.bool)) + if is_state_end: + state_ends_mask[-1][-1] = True # last token is state end + input_ids.append(chat_turn_ids) + entropy_mask.append(torch.ones(nb_chat_turns_ids, dtype=torch.bool)) + if not is_entropy_mask_true: + entropy_mask[-1] = entropy_mask[-1] * False + timesteps.append(torch.ones(nb_chat_turns_ids) * time_step) + input_ids = torch.cat(input_ids) + action_mask = torch.cat(action_mask) + entropy_mask = torch.cat(entropy_mask) + timesteps = torch.cat(timesteps) + timesteps = timesteps.to(torch.long) + state_ends_mask = torch.cat(state_ends_mask) + engine_log_probs = torch.cat(engine_log_probs) + + return ( + input_ids, + action_mask, + entropy_mask, + timesteps, + state_ends_mask, + engine_log_probs, + ) diff --git a/src_code_for_reproducibility/training/trainer_ad_align.py b/src_code_for_reproducibility/training/trainer_ad_align.py new file mode 100644 index 0000000000000000000000000000000000000000..14e18e51480e594355b3416555011223ff0e8f36 --- /dev/null +++ b/src_code_for_reproducibility/training/trainer_ad_align.py @@ -0,0 +1,505 @@ +""" +File: mllm/training/trainer_ad_align.py +Summary: Trainer specialized for the advantage-alignment objective. +""" + +import copy +import logging +import sys +from dataclasses import dataclass +from typing import Tuple + +import torch +from torch.nn.utils.rnn import pad_sequence + +from mllm.markov_games.rollout_tree import ( + ChatTurn, + RolloutTreeBranchNode, + RolloutTreeRootNode, +) +from mllm.training.credit_methods import ( + get_advantage_alignment_credits, + get_discounted_state_visitation_credits, +) +from mllm.training.tally_metrics import Tally +from mllm.training.tally_rollout import RolloutTally, RolloutTallyItem +from mllm.training.tally_tokenwise import ContextualizedTokenwiseTally +from mllm.training.tokenize_chats import process_training_chat +from mllm.training.trainer_common import BaseTrainer +from mllm.training.training_data_utils import ( + AdvantagePacket, + TrainingBatch, + TrainingChatTurn, + TrajectoryBatch, + get_main_chat_list_and_rewards, + get_tokenwise_credits, +) +from mllm.utils.resource_context import resource_logger_context + +logger = logging.getLogger(__name__) +logger.addHandler(logging.StreamHandler(sys.stdout)) + +RolloutId = int +AgentId = str + + +@dataclass +class AdAlignTrainingData: + """Holds tensorized rollouts plus precomputed advantages for one agent.""" + + agent_id: str + main_data: TrajectoryBatch + # list-of-tensors: per rollout advantages with length jT + main_advantages: list[torch.FloatTensor] | None = None + # list-of-tensors: per rollout matrix (jT, A) + alternative_advantages: list[torch.FloatTensor] | None = None + advantage_alignment_credits: list[torch.FloatTensor] | None = None + + +def get_alternative_chat_histories( + agent_id: str, root: RolloutTreeRootNode +) -> list[list[TrainingChatTurn], list[torch.FloatTensor]]: + """ + Traverse every unilateral branch under ``root`` and collect chat/reward histories. + + Returns + ------- + alternative_chats: + Flattened list of chat turns for each branch (ordered by branch depth). + alternative_rewards: + Matching list of reward tensors aligned with the chat history. + """ + current_node = root.child + branches = current_node.branches + pre_branch_chat = [] + pre_branch_rewards = [] + alternative_rewards = [] + alternative_chats = [] + while current_node is not None: + assert isinstance( + current_node, RolloutTreeBranchNode + ), "Current node should be a branch node." + main_node = current_node.main_child + branches = current_node.branches + current_node = main_node.child + + # Get the `A` alternative trajectories + alternative_nodes = branches[agent_id] + for alt_node in alternative_nodes: + post_branch_chat, post_branch_rewards = get_main_chat_list_and_rewards( + agent_id=agent_id, root=alt_node + ) + branch_chat = pre_branch_chat + post_branch_chat + alternative_chats.append(branch_chat) + alternative_rewards.append( + torch.cat([torch.tensor(pre_branch_rewards), post_branch_rewards]) + ) + + chat_turns: list[ChatTurn] = main_node.step_log.action_logs[agent_id].chat_turns + chat_turns: list[TrainingChatTurn] = [ + TrainingChatTurn(time_step=main_node.time_step, **turn.model_dump()) + for turn in chat_turns + ] + + pre_branch_chat.extend(chat_turns) + pre_branch_rewards.append( + main_node.step_log.simulation_step_log.rewards[agent_id] + ) + + return alternative_chats, alternative_rewards + + +class TrainerAdAlign(BaseTrainer): + """ + Extends the reinforce trainer to support Advantage Alignment. + """ + + def __init__( + self, + ad_align_beta: float, + ad_align_gamma: float, + ad_align_exclude_k_equals_t: bool, + ad_align_use_sign: bool, + ad_align_clipping: float, + ad_align_force_coop_first_step: bool, + use_old_ad_align: bool, + use_time_regularization: bool, + rloo_branch: bool, + reuse_baseline: bool, + ad_align_beta_anneal_step: int = -1, + ad_align_beta_anneal_rate: float = 0.5, + min_ad_align_beta: float = 0.1, + mean_normalize_ad_align: bool = False, + whiten_adalign_advantages: bool = False, + whiten_adalign_advantages_time_step_wise: bool = False, + ad_align_discount_t: bool = False, + *args, + **kwargs, + ): + """ + Initialize the advantage alignment trainer. + Args: + ad_align_beta: Beta parameter for the advantage alignment. + ad_align_gamma: Gamma parameter for the advantage alignment. + ad_align_exclude_k_equals_t: Whether to include k = t in the advantage alignment. + ad_align_use_sign: Whether to use sign in the advantage alignment. + ad_align_clipping: Clipping value for the advantage alignment. + ad_align_force_coop_first_step: Whether to force coop on the first step of the advantage alignment. + """ + super().__init__(*args, **kwargs) + self.ad_align_beta = ad_align_beta + self.ad_align_gamma = ad_align_gamma + self.ad_align_exclude_k_equals_t = ad_align_exclude_k_equals_t + self.ad_align_use_sign = ad_align_use_sign + self.ad_align_clipping = ad_align_clipping + self.ad_align_force_coop_first_step = ad_align_force_coop_first_step + self.use_old_ad_align = use_old_ad_align + self.use_time_regularization = use_time_regularization + self.rloo_branch = rloo_branch + self.reuse_baseline = reuse_baseline + self.ad_align_beta_anneal_step = ad_align_beta_anneal_step + self.ad_align_beta_anneal_rate = ad_align_beta_anneal_rate + self.min_ad_align_beta = min_ad_align_beta + self.past_ad_align_step = -1 + self.mean_normalize_ad_align = mean_normalize_ad_align + self.whiten_adalign_advantages = whiten_adalign_advantages + self.whiten_adalign_advantages_time_step_wise = ( + whiten_adalign_advantages_time_step_wise + ) + self.ad_align_discount_t = ad_align_discount_t + self.training_data: dict[AgentId, AdAlignTrainingData] = {} + self.debug_path_list: list[str] = [] + + def set_agent_trajectory_data( + self, agent_id: str, roots: list[RolloutTreeRootNode] + ): + """ + Materialize main and alternative trajectory tensors used by the advantage-alignment trainer. + """ + + B = len(roots) # Number of rollouts + + # For main rollouts + batch_rollout_ids = [] + batch_crn_ids = [] + batch_input_ids = [] + batch_action_mask = [] + batch_entropy_mask = [] + batch_timesteps = [] + batch_state_ends_mask = [] + batch_engine_log_probs = [] + batch_rewards = [] + + # For alternative actions rollouts + batch_branching_time_steps = [] + alternative_batch_input_ids = [] + alternative_batch_action_mask = [] + alternative_batch_entropy_mask = [] + alternative_batch_timesteps = [] + alternative_batch_state_ends_mask = [] + alternative_batch_engine_log_probs = [] + alternative_batch_rewards = [] + jT_list = [] + + try: + A = len(roots[0].child.branches[agent_id]) # Number of alternative actions + except: + A = 0 + + for root in roots: + rollout_id = root.id + self.debug_path_list.append( + "mgid:" + str(rollout_id) + "_agent_id:" + agent_id + ) + # Get main trajectory + batch_rollout_ids.append(rollout_id) + batch_crn_ids.append(root.crn_id) + main_chat, main_rewards = get_main_chat_list_and_rewards( + agent_id=agent_id, root=root + ) + ( + input_ids, + action_mask, + entropy_mask, + timesteps, + state_ends_mask, + engine_log_probs, + ) = process_training_chat( + tokenizer=self.tokenizer, + chat_history=main_chat, + entropy_mask_regex=self.entropy_mask_regex, + exploration_prompts_to_remove=self.exploration_prompts_to_remove, + ) + batch_input_ids.append(input_ids) + batch_action_mask.append(action_mask) + batch_entropy_mask.append(entropy_mask) + batch_timesteps.append(timesteps) + batch_state_ends_mask.append(state_ends_mask) + batch_engine_log_probs.append(engine_log_probs) + batch_rewards.append(main_rewards) + jT = ( + main_rewards.numel() + ) # Number of timesteps inferred from reward tensor length. + jT_list.append(jT) + if A > 0: + # We get the branching time steps for each of the `jT` time steps in the main trajectory. + branching_time_steps = [bt for item in range(jT) for bt in A * [item]] + batch_branching_time_steps.extend(branching_time_steps) + + # Get all of the (jT*A) alternative trajectories in the tree + # (jT is the number of time steps in the main trajectory, A is the number of alternative actions) + alternative_chats, alternative_rewards = get_alternative_chat_histories( + agent_id=agent_id, root=root + ) + assert ( + len(alternative_chats) == A * jT + ), "Incorrect number of alternative trajectories." + + for chat, rewards in zip(alternative_chats, alternative_rewards): + ( + input_ids, + action_mask, + entropy_mask, + timesteps, + state_ends_mask, + engine_log_probs, + ) = process_training_chat( + tokenizer=self.tokenizer, + chat_history=chat, + entropy_mask_regex=self.entropy_mask_regex, + exploration_prompts_to_remove=self.exploration_prompts_to_remove, + ) + alternative_batch_input_ids.append(input_ids) + alternative_batch_action_mask.append(action_mask) + alternative_batch_entropy_mask.append(entropy_mask) + alternative_batch_timesteps.append(timesteps) + alternative_batch_state_ends_mask.append(state_ends_mask) + alternative_batch_engine_log_probs.append(engine_log_probs) + alternative_batch_rewards.append(rewards) + + jT_list = torch.Tensor(jT_list) + + # Assert that number of alternative actions is constant + # assert len(set(nb_alternative_actions)) == 1, "Number of alternative actions must be constant" + # A = nb_alternative_actions[0] + + trajectory_batch = TrajectoryBatch( + rollout_ids=torch.tensor(batch_rollout_ids, dtype=torch.int32), # (B,) + crn_ids=torch.tensor(batch_crn_ids, dtype=torch.int32), + agent_ids=[agent_id] * len(batch_rollout_ids), + batch_input_ids=batch_input_ids, + batch_action_mask=batch_action_mask, + batch_entropy_mask=batch_entropy_mask, + batch_timesteps=batch_timesteps, + batch_state_ends_mask=batch_state_ends_mask, + batch_engine_log_probs=batch_engine_log_probs, + batch_rewards=batch_rewards, + ) + # Get Advantages & Train Critic + with resource_logger_context( + logger, "Get advantages with critic gradient accumulation" + ): + self.batch_advantages: torch.FloatTensor = ( + self.get_advantages_with_critic_gradient_accumulation(trajectory_batch) + ) # (B, jT) + + if A > 0: + # Here, `A` is the number of alternative actions / trajectories taken at each time step. + # For each of the `B` rollout perspectives, at each of its jT (`j` is for jagged, since each main rollout may be of a different length) steps, we take A alternate trajectories (from different actions). + # Therefore, we have ∑jT * A trajectories to process. If each of the main trajectories have T steps, we will have `B*T*A` to process. + with resource_logger_context(logger, "Create alternative trajectory batch"): + sum_jT = int(torch.sum(jT_list).item()) + jT_list = ( + jT_list.int().tolist() + ) # (jT,) # (we only want the advantages where we branched out) + alternative_trajectory_batch = TrajectoryBatch( + rollout_ids=torch.zeros(A * sum_jT, dtype=torch.int32), + crn_ids=torch.zeros(A * sum_jT, dtype=torch.int32), + agent_ids=[agent_id] * (A * sum_jT), + batch_input_ids=alternative_batch_input_ids, + batch_action_mask=alternative_batch_action_mask, + batch_entropy_mask=alternative_batch_entropy_mask, + batch_timesteps=alternative_batch_timesteps, + batch_state_ends_mask=alternative_batch_state_ends_mask, + batch_engine_log_probs=alternative_batch_engine_log_probs, + batch_rewards=alternative_batch_rewards, + ) + + # Get alternative advantages + # BAAs stands for batch alternative advantages + # (torch nested tensors have very little api support, so we have to do some odd manual work here) + with resource_logger_context( + logger, "Compute alternative advantage estimates" + ): + BAAs_list = self.get_advantages_with_critic_gradient_accumulation( + alternative_trajectory_batch + ) # list length (∑jT * A), each (jT',) + # Pad alternative advantages to (∑jT*A, P) + + BAAs_padded = pad_sequence( + BAAs_list, batch_first=True, padding_value=0.0 + ) + branch_idx = torch.tensor( + batch_branching_time_steps, + device=BAAs_padded.device, + dtype=torch.long, + ) + gathered = BAAs_padded.gather( + dim=1, index=branch_idx.unsqueeze(1) + ).squeeze(1) + # Reshape and split per rollout, then transpose to (jT_i, A) + gathered = gathered.view(A, sum_jT) # (A, ∑jT) + blocks = list( + torch.split(gathered, jT_list, dim=1) + ) # len B, shapes (A, jT_i) + BAAs = [ + blk.transpose(0, 1).contiguous() for blk in blocks + ] # list of (jT_i, A) + if self.ad_align_beta_anneal_step > 0: + max_rollout_id = torch.max(trajectory_batch.rollout_ids) + 1 + if ( + max_rollout_id % self.ad_align_beta_anneal_step == 0 + and self.past_ad_align_step != max_rollout_id + ): + self.ad_align_beta = max( + self.ad_align_beta * self.ad_align_beta_anneal_rate, + self.min_ad_align_beta, + ) + logger.info(f"Annealing ad_align_beta to {self.ad_align_beta}") + self.past_ad_align_step = max_rollout_id + self.training_data[agent_id] = AdAlignTrainingData( + agent_id=agent_id, + main_data=trajectory_batch, + main_advantages=self.batch_advantages, + alternative_advantages=BAAs if A > 0 else None, + ) + + def share_advantage_data(self) -> list[AdvantagePacket]: + """ + Share the advantage alignment data with other agents. + Returns: + AdvantagePacket: The advantage packet containing the agent's advantages. + """ + logger.info(f"Sharing advantage alignment data.") + advantage_packets = [] + for _, agent_data in self.training_data.items(): + advantage_packets.append( + AdvantagePacket( + agent_id=agent_data.agent_id, + rollout_ids=agent_data.main_data.rollout_ids, + main_advantages=agent_data.main_advantages, + ) + ) + return advantage_packets + + def receive_advantage_data(self, advantage_packets: list[AdvantagePacket]): + """ + Receive advantage packets from other players. + These contain the advantages of the other players' rollouts estimated by them. + """ + logger.info(f"Receiving advantage packets.") + + assert ( + len(advantage_packets) > 0 + ), "At least one advantage packet must be provided." + + for agent_id, agent_data in self.training_data.items(): + coagent_advantage_packets = [ + packet for packet in advantage_packets if packet.agent_id != agent_id + ] + agent_rollout_ids = agent_data.main_data.rollout_ids + agent_advantages = agent_data.main_advantages + co_agent_advantages = [] + for rollout_id in agent_rollout_ids: + for co_agent_packet in coagent_advantage_packets: + if rollout_id in co_agent_packet.rollout_ids: + index = torch.where(rollout_id == co_agent_packet.rollout_ids)[ + 0 + ].item() + co_agent_advantages.append( + co_agent_packet.main_advantages[index] + ) + # assumes that its two player game, with one co-agent + break + assert len(co_agent_advantages) == len(agent_advantages) + B = len(agent_advantages) + assert all( + a.shape[0] == b.shape[0] + for a, b in zip(co_agent_advantages, agent_advantages) + ), "Number of advantages must match for advantage alignment." + + # Get padded tensors (advantage alignment is invariant to padding) + lengths = torch.tensor( + [len(t) for t in agent_advantages], + device=self.device, + dtype=torch.long, + ) + padded_main_advantages = pad_sequence( + agent_advantages, batch_first=True, padding_value=0.0 + ) + if agent_data.alternative_advantages: + padded_alternative_advantages = pad_sequence( + agent_data.alternative_advantages, + batch_first=True, + padding_value=0.0, + ) # (B, P, A) + else: + padded_alternative_advantages = None + padded_co_agent_advantages = pad_sequence( + co_agent_advantages, batch_first=True, padding_value=0.0 + ) + + # Create training batch data + credits, sub_tensors = get_advantage_alignment_credits( + a1=padded_main_advantages, + a1_alternative=padded_alternative_advantages, + a2=padded_co_agent_advantages, + beta=self.ad_align_beta, + gamma=self.ad_align_gamma, + exclude_k_equals_t=self.ad_align_exclude_k_equals_t, + use_sign=self.ad_align_use_sign, + clipping=self.ad_align_clipping, + force_coop_first_step=self.ad_align_force_coop_first_step, + use_old_ad_align=self.use_old_ad_align, + use_time_regularization=self.use_time_regularization, + rloo_branch=self.rloo_branch, + reuse_baseline=self.reuse_baseline, + mean_normalize_ad_align=self.mean_normalize_ad_align, + whiten_adalign_advantages=self.whiten_adalign_advantages, + whiten_adalign_advantages_time_step_wise=self.whiten_adalign_advantages_time_step_wise, + discount_t=self.ad_align_discount_t, + ) + for key, value in sub_tensors.items(): + self.rollout_tally.add_metric( + path=[key], + rollout_tally_item=RolloutTallyItem( + crn_ids=agent_data.main_data.crn_ids, + rollout_ids=agent_data.main_data.rollout_ids, + agent_ids=agent_data.main_data.agent_ids, + metric_matrix=value, + ), + ) + + if not self.skip_discounted_state_visitation: + credits = get_discounted_state_visitation_credits( + credits, + self.discount_factor, + ) + self.rollout_tally.add_metric( + path=["discounted_state_visitation_credits"], + rollout_tally_item=RolloutTallyItem( + crn_ids=agent_data.main_data.crn_ids, + rollout_ids=agent_data.main_data.rollout_ids, + agent_ids=agent_data.main_data.agent_ids, + metric_matrix=sub_tensors[ + "discounted_state_visitation_credits" + ], + ), + ) + + # Slice back to jagged + advantage_alignment_credits = [credits[i, : lengths[i]] for i in range(B)] + # Replace stored training data for this agent by the concrete trajectory batch + # and attach the computed credits for policy gradient. + self.training_data[agent_id] = agent_data.main_data + self.training_data[agent_id].batch_credits = advantage_alignment_credits diff --git a/src_code_for_reproducibility/training/trainer_common.py b/src_code_for_reproducibility/training/trainer_common.py new file mode 100644 index 0000000000000000000000000000000000000000..0a0fb6f64707df0314ca4ed53611c2c592cc4377 --- /dev/null +++ b/src_code_for_reproducibility/training/trainer_common.py @@ -0,0 +1,1032 @@ +""" +File: mllm/training/trainer_common.py +Summary: Shared trainer utilities, base classes, and gradient helpers. +""" + +import logging +import os +import pickle +import sys +from abc import ABC, abstractmethod +from typing import Callable, Literal, Union + +import numpy as np +import torch +import torch.nn.functional as F +from accelerate import Accelerator +from pandas._libs.tslibs.offsets import CBMonthBegin +from peft import LoraConfig +from torch.nn.utils.rnn import pad_sequence +from transformers import AutoModelForCausalLM, AutoTokenizer + +from mllm.markov_games.rollout_tree import * +from mllm.markov_games.rollout_tree import RolloutTreeRootNode +from mllm.training.annealing_methods import sigmoid_annealing +from mllm.training.credit_methods import ( + get_discounted_returns, + get_generalized_advantage_estimates, + get_rloo_credits, + whiten_advantages, + whiten_advantages_time_step_wise, +) +from mllm.training.tally_metrics import Tally +from mllm.training.tally_rollout import RolloutTally, RolloutTallyItem +from mllm.training.tally_tokenwise import ContextualizedTokenwiseTally +from mllm.training.tokenize_chats import * +from mllm.training.tokenize_chats import process_training_chat +from mllm.training.training_data_utils import * +from mllm.training.training_data_utils import ( + TrainingBatch, + TrajectoryBatch, + get_tokenwise_credits, +) +from mllm.utils.resource_context import resource_logger_context + +logger = logging.getLogger(__name__) +logger.addHandler(logging.StreamHandler(sys.stdout)) + + +@dataclass +class TrainerAnnealingState: + annealing_step_counter: int = 0 + + +class BaseTrainer(ABC): + """ + Shared scaffolding for policy-gradient trainers (optimizer wiring, logging, etc.). + + Subclasses implement `set_agent_trajectory_data` / `share_advantage_data` + to plug in algorithm-specific behavior. + """ + + def __init__( + self, + policy: AutoModelForCausalLM, + policy_optimizer: torch.optim.Optimizer, + critic: Union[AutoModelForCausalLM, None], + critic_optimizer: Union[torch.optim.Optimizer, None], + tokenizer: AutoTokenizer, + lr_scheduler: torch.optim.lr_scheduler.LRScheduler, + critic_lr_scheduler: Union[torch.optim.lr_scheduler.LRScheduler, None], + ###################################################################### + entropy_coeff: float, + entropy_topk: int, + entropy_mask_regex: Union[str, None], + kl_coeff: float, + gradient_clipping: Union[float, None], + restrict_tokens: Union[list[str], None], + mini_batch_size: int, + use_gradient_checkpointing: bool, + temperature: float, + device: str, + whiten_advantages: bool, + whiten_advantages_time_step_wise: bool, + use_gae: bool, + use_gae_lambda_annealing: bool, + gae_lambda_annealing_limit: float, + gae_lambda_annealing_method: Literal["sigmoid_annealing"], + gae_lambda_annealing_method_params: dict, + pg_loss_normalization: Literal["batch", "nb_tokens"], + use_rloo: bool, + skip_discounted_state_visitation: bool, + discount_factor: float, + enable_tokenwise_logging: bool, + save_path: str, + reward_normalizing_constant: float = 1.0, + critic_loss_type: Literal["mse", "huber"] = "huber", + exploration_prompts_to_remove: list[str] = [], + filter_higher_refprob_tokens_kl: bool = False, + truncated_importance_sampling_ratio_cap: float = 0.0, + importance_sampling_strategy: Literal[ + "per_token", "per_sequence" + ] = "per_token", + no_rloo_grouping: bool = False, + ): + """ + Initialize the REINFORCE trainer with reward shaping for multi-agent or single-agent training. + + Args: + model (AutoModelForCausalLM): The main policy model. + tokenizer (AutoTokenizer): Tokenizer for the model. + optimizer (torch.optim.Optimizer): Optimizer for the policy model. + lr_scheduler (torch.optim.lr_scheduler.LRScheduler): Learning rate scheduler for the policy model. + critic (AutoModelForCausalLM or None): Critic model for value estimation (optional). + critic_optimizer (torch.optim.Optimizer or None): Optimizer for the critic model (optional). + critic_lr_scheduler (torch.optim.lr_scheduler.LRScheduler or None): LR scheduler for the critic (optional). + config (RtConfig): Configuration object for training. + """ + self.tokenizer = tokenizer + # self.tokenizer.padding_side = "left" # needed for flash attention + if self.tokenizer.pad_token_id is None: + self.tokenizer.pad_token_id = self.tokenizer.eos_token_id + self.lr_scheduler = lr_scheduler + self.accelerator = Accelerator() + ( + self.policy, + self.policy_optimizer, + self.critic, + self.critic_optimizer, + ) = self.accelerator.prepare(policy, policy_optimizer, critic, critic_optimizer) + + self.critic_lr_scheduler = critic_lr_scheduler + self.tally = Tally() + + if use_gradient_checkpointing == True: + self.policy.gradient_checkpointing_enable(dict(use_reentrant=False)) + if critic is not None: + self.critic.gradient_checkpointing_enable(dict(use_reentrant=False)) + + self.save_path = save_path + + # Load trainer state if it exists + self.trainer_annealing_state_path = os.path.join( + self.save_path, "trainer_annealing_state.pkl" + ) + if os.path.exists(self.trainer_annealing_state_path): + logger.info( + f"Loading trainer state from {self.trainer_annealing_state_path}" + ) + self.trainer_annealing_state = pickle.load( + open(self.trainer_annealing_state_path, "rb") + ) + else: + self.trainer_annealing_state = TrainerAnnealingState() + + # Load policy optimizer state if it exists + self.policy_optimizer_path = os.path.join( + self.save_path, "policy_optimizer_state.pt" + ) + if os.path.exists(self.policy_optimizer_path): + logger.info( + f"Loading policy optimizer state from {self.policy_optimizer_path}" + ) + self.policy_optimizer.load_state_dict( + torch.load(self.policy_optimizer_path) + ) + + # Load critic optimizer state if it exists + self.critic_optimizer_path = os.path.join( + self.save_path, "critic_optimizer_state.pt" + ) + if ( + os.path.exists(self.critic_optimizer_path) + and self.critic_optimizer is not None + ): + logger.info( + f"Loading critic optimizer state from {self.critic_optimizer_path}" + ) + self.critic_optimizer.load_state_dict( + torch.load(self.critic_optimizer_path) + ) + self.device = self.accelerator.device + self.entropy_coeff = entropy_coeff + self.entropy_topk = entropy_topk + self.entropy_mask_regex = entropy_mask_regex + self.kl_coeff = kl_coeff + self.gradient_clipping = gradient_clipping + self.restrict_tokens = restrict_tokens + self.mini_batch_size = mini_batch_size + self.use_gradient_checkpointing = use_gradient_checkpointing + self.temperature = temperature + self.use_gae = use_gae + self.whiten_advantages = whiten_advantages + self.whiten_advantages_time_step_wise = whiten_advantages_time_step_wise + self.use_rloo = use_rloo + self.skip_discounted_state_visitation = skip_discounted_state_visitation + self.use_gae_lambda_annealing = use_gae_lambda_annealing + self.gae_lambda_annealing_limit = gae_lambda_annealing_limit + if use_gae_lambda_annealing: + self.gae_lambda_annealing_method: Callable[ + [int], float + ] = lambda step: eval(gae_lambda_annealing_method)( + step=step, **gae_lambda_annealing_method_params + ) + self.discount_factor = discount_factor + self.enable_tokenwise_logging = enable_tokenwise_logging + self.reward_normalizing_constant = reward_normalizing_constant + self.pg_loss_normalization = pg_loss_normalization + self.critic_loss_type = critic_loss_type + self.exploration_prompts_to_remove = exploration_prompts_to_remove + # Common containers used by all trainers + self.training_data: dict = {} + self.debug_path_list: list[str] = [] + self.policy_gradient_data = None + self.tally = Tally() + self.rollout_tally = RolloutTally() + self.tokenwise_tally: Union[ContextualizedTokenwiseTally, None] = None + self.filter_higher_refprob_tokens_kl = filter_higher_refprob_tokens_kl + self.truncated_importance_sampling_ratio_cap = ( + truncated_importance_sampling_ratio_cap + ) + self.importance_sampling_strategy = importance_sampling_strategy + self.no_rloo_grouping = no_rloo_grouping + + def mask_non_restricted_token_logits(self, logits: torch.Tensor) -> torch.Tensor: + """ + Masks logits so that only allowed tokens (as specified in config.restrict_tokens) + and the EOS token are active. + All other logits are set to -inf, effectively removing them from the softmax. + + Args: + logits (torch.Tensor): The logits tensor of shape (B, S, V). + + Returns: + torch.Tensor: The masked logits tensor. + """ + # Gradients flow only through the kept logits; masking is recomputed per batch for clarity. + + if self.restrict_tokens is not None: + allowed_token_ids = [] + for token in self.restrict_tokens: + token_ids = self.tokenizer(token, add_special_tokens=False)["input_ids"] + allowed_token_ids.append(token_ids[0]) + allowed_token_ids.append( + self.tokenizer.eos_token_id + ) # This token should always be active + allowed_token_ids = torch.tensor(allowed_token_ids, device=logits.device) + # Mask log_probs and probs to only allowed tokens + mask = torch.zeros_like(logits).bool() # (B, S, V) + mask[..., allowed_token_ids] = True + logits = torch.where( + mask, + logits, + torch.tensor(-float("inf"), device=logits.device), + ) + + return logits + + def apply_reinforce_step( + self, + training_batch: TrainingBatch, + ) -> None: + """ + Applies a single REINFORCE policy gradient step using the provided batch of rollouts. + Handles batching, loss computation (including entropy and KL regularization), gradient accumulation, and optimizer step. + Optionally logs various metrics and statistics. + + Args: + paths (list[str]): List of game complete file paths for each rollout. + contexts (list[torch.Tensor]): List of context tensors for each rollout. + credits (list[torch.Tensor]): List of credit tensors (rewards/advantages) for each rollout. + action_masks (list[torch.Tensor]): List of action mask tensors for each rollout. + """ + with resource_logger_context(logger, "Apply reinforce step"): + self.policy.train() + mb_size = self.mini_batch_size + nb_rollouts = len(training_batch) + + # Initialize running mean logs + running_mean_logs = { + "rl_objective": 0.0, + "policy_gradient_loss": 0.0, + "policy_gradient_norm": 0.0, + "log_probs": 0.0, + "credits": 0.0, + "entropy": 0.0, + "engine_log_probs_diff_clampfrac": 0.0, + "tis_imp_ratio": 0.0, + "ref_log_probs_diff_clampfrac": 0.0, + "higher_refprob_frac": 0.0, + "tis_imp_ratio_clampfrac": 0.0, + } + if self.entropy_coeff != 0.0: + running_mean_logs["entropy"] = 0.0 + if self.kl_coeff != 0.0: + running_mean_logs["kl_divergence"] = 0.0 + + # Get total number of tokens generated + total_tokens_generated = 0 + for att_mask in training_batch.batch_action_mask: + total_tokens_generated += att_mask.sum() + + # Obtain loss normalization + if self.pg_loss_normalization == "nb_tokens": + normalization_factor = total_tokens_generated + elif self.pg_loss_normalization == "batch": + normalization_factor = np.ceil(nb_rollouts / mb_size).astype(int) + else: + raise ValueError( + f"Invalid pg_loss_normalization: {self.pg_loss_normalization}" + ) + + # Gradient accumulation for each mini-batch + for mb in range(0, nb_rollouts, mb_size): + logger.info(f"Processing mini-batch {mb} of {nb_rollouts}") + loss = 0.0 + training_mb = training_batch[mb : mb + mb_size] + training_mb = training_mb.get_padded_tensors() + training_mb.to(self.device) + ( + tokens_mb, + action_mask_mb, + entropy_mask_mb, + credits_mb, + engine_log_probs_mb, + timesteps_mb, + ) = ( + training_mb.batch_input_ids, + training_mb.batch_action_mask, + training_mb.batch_entropy_mask, + training_mb.batch_credits, + training_mb.batch_engine_log_probs, + training_mb.batch_timesteps, + ) + + # Next token prediction + contexts_mb = tokens_mb[:, :-1] + shifted_contexts_mb = tokens_mb[:, 1:] + action_mask_mb = action_mask_mb[:, 1:] + entropy_mask_mb = entropy_mask_mb[:, 1:] + credits_mb = credits_mb[:, 1:] + engine_log_probs_mb = engine_log_probs_mb[:, 1:] + timesteps_mb = timesteps_mb[:, 1:] + + if self.enable_tokenwise_logging: + self.tokenwise_tally.set_action_mask(action_mask=action_mask_mb) + self.tokenwise_tally.set_range(range=(mb, mb + mb_size)) + self.tokenwise_tally.add_contexts(contexts=contexts_mb) + self.tokenwise_tally.add_data( + metric_id="next_token", + metrics=shifted_contexts_mb, + to_tids=True, + ) + self.tokenwise_tally.add_data( + metric_id="entropy_mask", + metrics=entropy_mask_mb, + ) + + if self.enable_tokenwise_logging: + self.tokenwise_tally.add_data( + metric_id="next_token_credit", metrics=credits_mb + ) + + # Forward pass + cast to FP-32 for higher prec. Causal LM attention masks are implicit; + # wire up a custom mask here only if the policy deviates from standard autoregressive behavior. + logits = self.policy(input_ids=contexts_mb)[0] # (B, S, V) + + # Mask non-restricted tokens + if self.restrict_tokens is not None: + logits = self.mask_non_restricted_token_logits(logits) + + logits /= self.temperature # (B, S, V) + + # Compute new log probabilities + log_probs = F.log_softmax(logits, dim=-1) # (B, S, V) + + # Get log probabilities of actions taken during rollouts + action_log_probs = log_probs.gather( + dim=-1, index=shifted_contexts_mb.unsqueeze(-1) + ).squeeze( + -1 + ) # (B, S) + if self.pg_loss_normalization == "batch": + den_running_mean = action_mask_mb.sum() * normalization_factor + else: + den_running_mean = normalization_factor + running_mean_logs["log_probs"] += ( + action_log_probs * action_mask_mb + ).sum().item() / den_running_mean + running_mean_logs["credits"] += ( + credits_mb * action_mask_mb + ).sum().item() / den_running_mean + + if self.enable_tokenwise_logging: + self.tokenwise_tally.add_data( + metric_id="next_token_log_prob", + metrics=action_log_probs, + ) + self.tokenwise_tally.add_data( + metric_id="engine_next_token_log_prob", + metrics=engine_log_probs_mb, + ) + self.tokenwise_tally.add_data( + metric_id="next_token_prob", + metrics=torch.exp(action_log_probs), + ) + top_k_indices = torch.topk(logits, k=5, dim=-1).indices + self.tokenwise_tally.add_data( + metric_id=f"top_{5}_tids", + metrics=top_k_indices, + to_tids=True, + ) + self.tokenwise_tally.add_data( + metric_id=f"top_{5}_probs", + metrics=torch.exp(log_probs).gather( + dim=-1, index=top_k_indices + ), + ) + + rewarded_action_log_probs = ( + action_mask_mb * credits_mb * action_log_probs + ) + # (B, S) + INVALID_LOGPROB = 1.0 + CLAMP_VALUE = 40.0 + masked_action_log_probs = torch.masked_fill( + action_log_probs, ~action_mask_mb, INVALID_LOGPROB + ) + masked_engine_log_probs = torch.masked_fill( + engine_log_probs_mb, ~action_mask_mb, INVALID_LOGPROB + ) + with torch.no_grad(): + action_engine_log_probs_diff = ( + masked_action_log_probs - masked_engine_log_probs + ).clamp(-CLAMP_VALUE, CLAMP_VALUE) + running_mean_logs["engine_log_probs_diff_clampfrac"] += ( + action_engine_log_probs_diff.abs() + .eq(CLAMP_VALUE) + .float() + .sum() + .item() + / den_running_mean + ) + if self.importance_sampling_strategy == "per_sequence": + tis_imp_ratio = torch.zeros_like(action_engine_log_probs_diff) + for mb_idx in range(action_engine_log_probs_diff.shape[0]): + valid_token_mask = action_mask_mb[mb_idx] + timestep_ids = timesteps_mb[mb_idx][valid_token_mask] + timestep_logprob_diffs = action_engine_log_probs_diff[mb_idx][ + valid_token_mask + ] + max_timestep = int(timestep_ids.max().item()) + 1 + timestep_sums = torch.zeros( + max_timestep, + device=action_engine_log_probs_diff.device, + dtype=action_engine_log_probs_diff.dtype, + ) + timestep_sums.scatter_add_( + 0, timestep_ids, timestep_logprob_diffs + ) + timestep_ratios = torch.exp(timestep_sums) + tis_imp_ratio[ + mb_idx, valid_token_mask + ] = timestep_ratios.gather(0, timestep_ids) + else: + tis_imp_ratio = torch.exp(action_engine_log_probs_diff) + running_mean_logs["tis_imp_ratio"] += ( + tis_imp_ratio * action_mask_mb + ).sum().item() / den_running_mean + if self.truncated_importance_sampling_ratio_cap > 0.0: + tis_imp_ratio = torch.clamp( + tis_imp_ratio, max=self.truncated_importance_sampling_ratio_cap + ) + running_mean_logs["tis_imp_ratio_clampfrac"] += ( + tis_imp_ratio.eq(self.truncated_importance_sampling_ratio_cap) + .float() + .sum() + .item() + ) / den_running_mean + rewarded_action_log_probs = ( + rewarded_action_log_probs * tis_imp_ratio + ) + + if self.enable_tokenwise_logging: + self.tokenwise_tally.add_data( + metric_id="next_token_clogπ", + metrics=rewarded_action_log_probs, + ) + + # Add value term to loss + if self.pg_loss_normalization == "batch": + nb_act_tokens = action_mask_mb.sum() + mb_value = -rewarded_action_log_probs.sum() / nb_act_tokens + else: + mb_value = -rewarded_action_log_probs.sum() + + loss += mb_value + running_mean_logs["rl_objective"] += mb_value.item() / den_running_mean + + # ------------------------------------------------- + # Entropy Regularization + # ------------------------------------------------- + # Only apply entropy on distribution defined over most probable tokens + if self.entropy_topk is not None: + top_k_indices = torch.topk( + logits, k=self.entropy_topk, dim=-1 + ).indices + entropy_logits = logits.gather(dim=-1, index=top_k_indices) + else: + entropy_logits = logits + + token_entropy_terms = -F.softmax( + entropy_logits, dim=-1 + ) * F.log_softmax( + entropy_logits, dim=-1 + ) # (B, S, T) + token_entropy_terms *= ( + action_mask_mb[:, :, None] * entropy_mask_mb[:, :, None] + ) # only get loss on specific action tokens + + mb_entropy = token_entropy_terms.sum(dim=-1) + + if self.enable_tokenwise_logging: + self.tokenwise_tally.add_data( + metric_id="entropy", + metrics=mb_entropy, + ) + if self.pg_loss_normalization == "batch": + nb_act_tokens = action_mask_mb.sum() + mb_entropy = -mb_entropy.sum() / nb_act_tokens + else: + mb_entropy = -mb_entropy.sum() + running_mean_logs["entropy"] += -mb_entropy.item() / den_running_mean + if self.entropy_coeff != 0.0: + mb_entropy *= self.entropy_coeff + loss += mb_entropy + + # ------------------------------------------------- + # KL-DIVERGENCE + # ------------------------------------------------- + if self.kl_coeff != 0.0: + ref_model_logits = self.policy.get_base_model_logits(contexts_mb) + ref_model_logits = ref_model_logits / self.temperature + # (B, S, V) + ref_model_logits = self.mask_non_restricted_token_logits( + logits=ref_model_logits + ) + # (B, S, V) + ref_model_log_probs = F.log_softmax(ref_model_logits, dim=-1) + # (B, S, V) + ref_model_action_log_probs = ref_model_log_probs.gather( + dim=-1, index=shifted_contexts_mb.unsqueeze(-1) + ).squeeze( + -1 + ) # (B,S) + # Approximating KL Divergence (see refs in docstring) + # Ref 1: http://joschu.net/blog/kl-approx.html + # Ref 2: https://github.dev/huggingface/trl/blob/main/trl/trainer/grpo_trainer.py#L1332 + masked_ref_model_action_log_probs = torch.masked_fill( + ref_model_action_log_probs, ~action_mask_mb, INVALID_LOGPROB + ) + action_log_probs_diff = ( + masked_ref_model_action_log_probs - masked_action_log_probs + ).clamp(-CLAMP_VALUE, CLAMP_VALUE) + running_mean_logs["ref_log_probs_diff_clampfrac"] += ( + action_log_probs_diff.abs().eq(CLAMP_VALUE).float().sum().item() + / den_running_mean + ) + if self.filter_higher_refprob_tokens_kl: + higher_refprob_tokens_mask = action_log_probs_diff > 0.0 + running_mean_logs["higher_refprob_frac"] += ( + higher_refprob_tokens_mask.sum().item() / den_running_mean + ) + action_log_probs_diff = action_log_probs_diff * ( + ~higher_refprob_tokens_mask + ) + kl_div = torch.expm1(action_log_probs_diff) - action_log_probs_diff + kl_div *= action_mask_mb # We only care about KLD of action tokens + if self.truncated_importance_sampling_ratio_cap > 0.0: + kl_div = kl_div * tis_imp_ratio + kl_div *= self.kl_coeff + if self.enable_tokenwise_logging: + self.tokenwise_tally.add_data( + metric_id="ref_model_next_token_log_prob", + metrics=ref_model_action_log_probs, + ) + self.tokenwise_tally.add_data( + metric_id="kl_divergence", + metrics=kl_div, + ) + + if self.pg_loss_normalization == "batch": + nb_act_tokens = action_mask_mb.sum() + mb_kl = kl_div.sum() / nb_act_tokens + else: + mb_kl = kl_div.sum() + running_mean_logs["kl_divergence"] += ( + mb_kl.item() / den_running_mean + ) + loss += mb_kl + + # Accumulate gradient + running_mean_logs["policy_gradient_loss"] += ( + loss.item() / den_running_mean + ) + loss /= normalization_factor + self.accelerator.backward(loss) + + # ensure gpu memory is freed + del training_mb + del log_probs + del logits + del loss + del action_log_probs + del rewarded_action_log_probs + + logger.info( + f"Accumulated the policy gradient loss for {total_tokens_generated} tokens." + ) + + # Clip gradients and take step + if self.gradient_clipping is not None: + grad_norm = self.accelerator.clip_grad_norm_( + self.policy.parameters(), self.gradient_clipping + ) + running_mean_logs["policy_gradient_norm"] += grad_norm.item() + + # Take step + self.policy_optimizer.step() + self.policy_optimizer.zero_grad() + + # Store logs + for key, value in running_mean_logs.items(): + self.tally.add_metric(path=key, metric=value) + + # Clear accelerator state so we do not accumulate references between optimizer steps. + self.accelerator.clear(self.policy, self.policy_optimizer) + import gc + + gc.collect() + torch.cuda.empty_cache() + return running_mean_logs + + def get_advantages_with_critic_gradient_accumulation( + self, trajectories: TrajectoryBatch, critic_loss_scaling_factor: float = 2.0 + ) -> torch.FloatTensor: + """ + Compute (and optionally whiten) advantages while training the critic in mini-batches. + Uses GAE if enabled, otherwise uses Monte Carlo returns. + Optionally trains the critic if GAE is used. + Returns: + advantages: NestedFloatTensors + """ + + mb_size = self.mini_batch_size + batch_size = trajectories.rollout_ids.shape[0] + agent_id = trajectories.agent_ids[0] + batch_rewards = trajectories.batch_rewards + + ###################################### + # use critic for advantage estimation + ###################################### + if self.use_gae: + if "buffer" in agent_id: + self.critic.eval() + training = False + else: + self.critic.train() + training = True + advantages = [] + # critic_loss_scaling_factor comes learning single critic for two agents + normalization_factor = ( + np.ceil(batch_size / mb_size).astype(int) * critic_loss_scaling_factor + ) + # For each minibatch + for mb in range(0, batch_size, mb_size): + trajectory_mb = trajectories[mb : mb + mb_size] + trajectory_mb.to(self.device) + rewards_mb = trajectory_mb.batch_rewards + ( + tokens_mb, + state_ends_mask_mb, + timestep_counts, + ) = trajectory_mb.get_padded_tensors_for_critic() + # critic causal attention up to end flags + if training: + vals_estimate_full = self.critic(tokens_mb) + else: + with torch.no_grad(): + vals_estimate_full = self.critic(tokens_mb) + + # if vals_estimate_full.dim() == 3: + # vals_estimate_full = vals_estimate_full.squeeze(-1) + + # Select only positions where states end, per sample → list of (jT,) + B = tokens_mb.shape[0] + vals_list = [ + vals_estimate_full[b][state_ends_mask_mb[b]] for b in range(B) + ] + + # Pad to (B, max_jT) = (B, S) + vals_estimate_mb = pad_sequence( + vals_list, batch_first=True, padding_value=0.0 + ) + dtype = vals_estimate_mb.dtype + rewards_mb = pad_sequence( + rewards_mb, batch_first=True, padding_value=0.0 + ).to( + dtype=dtype + ) # (B, S) + self.rollout_tally.add_metric( + path=["batch_rewards"], + rollout_tally_item=RolloutTallyItem( + crn_ids=trajectory_mb.crn_ids, + rollout_ids=trajectory_mb.rollout_ids, + agent_ids=trajectory_mb.agent_ids, + metric_matrix=rewards_mb, + ), + ) + if self.reward_normalizing_constant != 1.0: + rewards_mb /= self.reward_normalizing_constant + + det_vals_estimate_mb = vals_estimate_mb.detach() # (B, max_jT) + self.rollout_tally.add_metric( + path=["mb_value_estimates_critic"], + rollout_tally_item=RolloutTallyItem( + crn_ids=trajectory_mb.crn_ids, + rollout_ids=trajectory_mb.rollout_ids, + agent_ids=trajectory_mb.agent_ids, + metric_matrix=det_vals_estimate_mb, + ), + ) + + # Append a 0 value to the end of the value estimates + if det_vals_estimate_mb.shape[1] == rewards_mb.shape[1]: + Bsize = det_vals_estimate_mb.shape[0] + device = det_vals_estimate_mb.device + dtype = det_vals_estimate_mb.dtype + det_vals_estimate_mb = torch.cat( + [ + det_vals_estimate_mb, + torch.zeros((Bsize, 1), device=device, dtype=dtype), + ], + dim=1, + ) # (B, max_jT+1) + else: + raise ValueError( + "Incompatible shapes for value estimates and rewards." + ) + + # Get annealed lambda + if self.use_gae_lambda_annealing: + annealing_constant = self.gae_lambda_annealing_method( + step=self.trainer_annealing_state.annealing_step_counter + ) + annealed_lambda = ( + self.gae_lambda_annealing_limit * annealing_constant + ) + self.tally.add_metric( + path="annealed_lambda", metric=annealed_lambda + ) + else: + annealed_lambda = self.gae_lambda_annealing_limit + + # Get GAE advantages + gae_advantages = get_generalized_advantage_estimates( + rewards=rewards_mb, + value_estimates=det_vals_estimate_mb, + discount_factor=self.discount_factor, + lambda_coef=annealed_lambda, + ) # (B, max_jT) + self.rollout_tally.add_metric( + path=["mb_gae_advantages"], + rollout_tally_item=RolloutTallyItem( + crn_ids=trajectory_mb.crn_ids, + rollout_ids=trajectory_mb.rollout_ids, + agent_ids=trajectory_mb.agent_ids, + metric_matrix=gae_advantages, + ), + ) + if training: + targets = ( + gae_advantages.to(dtype=dtype) + det_vals_estimate_mb[:, :-1] + ) # (B, max_jT) # A(s, a, b) + V(s) = Q(s, a, b) + self.rollout_tally.add_metric( + path=["mb_targets_critic"], + rollout_tally_item=RolloutTallyItem( + crn_ids=trajectory_mb.crn_ids, + rollout_ids=trajectory_mb.rollout_ids, + agent_ids=trajectory_mb.agent_ids, + metric_matrix=targets, + ), + ) + if self.critic_loss_type == "mse": + loss = F.mse_loss( + input=vals_estimate_mb, + target=targets, + ) + elif self.critic_loss_type == "huber": + loss = F.huber_loss( + input=vals_estimate_mb, + target=targets, + ) + self.tally.add_metric(path=["mb_critic_loss"], metric=loss.item()) + # Accumulate gradient + loss /= normalization_factor + self.accelerator.backward(loss) + del loss + del targets + del vals_estimate_mb + del trajectory_mb + del vals_estimate_full + + # Get jagged back using timestep_counts + advantages.extend( + [gae_advantages[i, : timestep_counts[i]] for i in range(B)] + ) + + ###################################### + # use exclusively Monte Carlo returns & rloo for advantage estimation + ###################################### + else: + lengths = [len(c) for c in batch_rewards] + padded_rewards = pad_sequence( + batch_rewards, batch_first=True, padding_value=0.0 + ) + self.rollout_tally.add_metric( + path=["mb_rewards"], + rollout_tally_item=RolloutTallyItem( + crn_ids=trajectories.crn_ids, + rollout_ids=trajectories.rollout_ids, + agent_ids=trajectories.agent_ids, + metric_matrix=padded_rewards, + ), + ) + if self.reward_normalizing_constant != 1.0: + padded_rewards /= self.reward_normalizing_constant + padded_advantages = get_discounted_returns( + rewards=padded_rewards, + discount_factor=self.discount_factor, + ) # no baseline for now + if self.use_rloo: + is_grouped_by_rng = ( + trajectories.crn_ids.unique().shape[0] + != trajectories.crn_ids.shape[0] + ) + if is_grouped_by_rng and not self.no_rloo_grouping: + for crn_id in trajectories.crn_ids.unique(): + rng_mask = trajectories.crn_ids == crn_id + rng_advantages = padded_advantages[rng_mask] + rng_advantages, _ = get_rloo_credits(credits=rng_advantages) + padded_advantages[rng_mask] = rng_advantages + else: + padded_advantages, _ = get_rloo_credits(credits=padded_advantages) + self.rollout_tally.add_metric( + path=["mb_rloo_advantages"], + rollout_tally_item=RolloutTallyItem( + crn_ids=trajectories.crn_ids, + rollout_ids=trajectories.rollout_ids, + agent_ids=trajectories.agent_ids, + metric_matrix=padded_advantages, + ), + ) + advantages = [ + padded_advantages[i, : lengths[i]] + for i in range(padded_advantages.shape[0]) + ] + + if self.whiten_advantages_time_step_wise or self.whiten_advantages: + lengths = [len(c) for c in advantages] + padded_advantages = pad_sequence( + advantages, batch_first=True, padding_value=0.0 + ) + if self.whiten_advantages_time_step_wise: + whitened_padded_advantages = whiten_advantages_time_step_wise( + padded_advantages + ) + path = ["mb_whitened_advantages_time_step_wise"] + elif self.whiten_advantages: + whitened_padded_advantages = whiten_advantages(padded_advantages) + path = ["mb_whitened_advantages"] + self.rollout_tally.add_metric( + path=path, + rollout_tally_item=RolloutTallyItem( + crn_ids=trajectories.crn_ids, + rollout_ids=trajectories.rollout_ids, + agent_ids=trajectories.agent_ids, + metric_matrix=whitened_padded_advantages, + ), + ) + advantages = [ + whitened_padded_advantages[i, : lengths[i]] + for i in range(whitened_padded_advantages.shape[0]) + ] + + self.trainer_annealing_state.annealing_step_counter += 1 + + return advantages + + @abstractmethod + def set_agent_trajectory_data( + self, agent_id: str, roots: list[RolloutTreeRootNode] + ) -> None: + """ + Populate self.training_data for a single agent using the provided rollout trees. + """ + pass + + def set_trajectory_data( + self, roots: list[RolloutTreeRootNode], agent_ids: list[str] + ) -> None: + """ + Convenience wrapper to ingest trajectory data for every training agent. + """ + for agent_id in agent_ids: + self.set_agent_trajectory_data(agent_id, roots) + + @abstractmethod + def share_advantage_data(self) -> list[AdvantagePacket]: + pass + + @abstractmethod + def receive_advantage_data(self, advantage_packets: list[AdvantagePacket]) -> None: + pass + + def set_policy_gradient_data(self, agent_ids: list[str]) -> None: + """ + Reset and rebuild the policy-gradient minibatches before iterating through agents. + """ + self.policy_gradient_data = None + for agent_id in agent_ids: + assert "buffer" not in agent_id, "Buffer agents do not train policy" + trajectory_batch = self.training_data[agent_id] + tokenwise_batch_credits = get_tokenwise_credits( + batch_timesteps=trajectory_batch.batch_timesteps, + batch_credits=trajectory_batch.batch_credits, + ) + policy_gradient_data = TrainingBatch( + rollout_ids=trajectory_batch.rollout_ids, + batch_input_ids=trajectory_batch.batch_input_ids, + batch_action_mask=trajectory_batch.batch_action_mask, + batch_entropy_mask=trajectory_batch.batch_entropy_mask, + batch_credits=tokenwise_batch_credits, + batch_engine_log_probs=trajectory_batch.batch_engine_log_probs, + batch_timesteps=trajectory_batch.batch_timesteps, + ) + if self.policy_gradient_data is None: + self.policy_gradient_data = policy_gradient_data + else: + self.policy_gradient_data.append(policy_gradient_data) + + self.training_data = {} + self.tokenwise_tally = ContextualizedTokenwiseTally( + tokenizer=self.tokenizer, + paths=self.debug_path_list, + ) + + def train(self) -> None: + """ + Entry point for policy updates: prepare batches, compute gradients, and update parameters. + """ + assert self.policy_gradient_data is not None, "Policy gradient data is not set" + if self.critic_optimizer is not None: + if self.gradient_clipping is not None: + grad_norm = self.accelerator.clip_grad_norm_( + self.critic.parameters(), self.gradient_clipping + ) + self.tally.add_metric( + path="gradient_norm_critic", metric=grad_norm.item() + ) + # Take step + self.critic_optimizer.step() + self.critic_optimizer.zero_grad() + self.accelerator.clear(self.critic, self.critic_optimizer) + import gc + + gc.collect() + torch.cuda.empty_cache() + running_mean_logs = self.apply_reinforce_step( + training_batch=self.policy_gradient_data + ) + return running_mean_logs + + def export_training_tally(self, identifier: str, folder: str) -> None: + """ + Saves and resets the collected training metrics using the tally object. + """ + os.makedirs(folder, exist_ok=True) + self.tally.save(identifier=identifier, folder=folder) + self.tokenwise_tally.save( + path=os.path.join(folder, f"{identifier}_tokenwise.csv") + ) + self.rollout_tally.save(identifier=identifier, folder=folder) + self.tally.reset() + self.tokenwise_tally = None + self.rollout_tally.reset() + self.debug_path_list = [] + + def export_optimizer_states(self) -> None: + """ + Saves the optimizer states for both the main model and critic (if it exists). + """ + try: + os.makedirs(self.save_path, exist_ok=True) + + torch.save(self.policy_optimizer.state_dict(), self.policy_optimizer_path) + logger.info(f"Saved main optimizer state to {self.policy_optimizer_path}") + + if self.critic_optimizer is not None: + torch.save( + self.critic_optimizer.state_dict(), self.critic_optimizer_path + ) + logger.info( + f"Saved critic optimizer state to {self.critic_optimizer_path}" + ) + except Exception as e: + logger.error(f"Error saving optimizer states: {str(e)}") + raise + + def export_trainer_annealing_state(self) -> None: + """ + Saves the trainer state. + """ + with open(self.trainer_annealing_state_path, "wb") as f: + pickle.dump(self.trainer_annealing_state, f) + logger.info(f"Saved trainer state to {self.trainer_annealing_state_path}") + + def export_trainer_states(self) -> None: + """ + Saves the trainer states. + """ + self.export_optimizer_states() + self.export_trainer_annealing_state() diff --git a/src_code_for_reproducibility/training/trainer_independent.py b/src_code_for_reproducibility/training/trainer_independent.py new file mode 100644 index 0000000000000000000000000000000000000000..807c69b3e903024028d8255ae3e79eb6537e609b --- /dev/null +++ b/src_code_for_reproducibility/training/trainer_independent.py @@ -0,0 +1,159 @@ +""" +File: mllm/training/trainer_independent.py +Summary: Trainer for independently optimizing each agent. +""" + +import logging +import os +import sys +from typing import Union + +import torch +import torch.nn.functional as F +from accelerate import Accelerator +from pandas._libs.tslibs.offsets import CBMonthBegin +from peft import LoraConfig +from torch.nn.utils.rnn import pad_sequence +from transformers import AutoModelForCausalLM, AutoTokenizer + +from mllm.markov_games.rollout_tree import * +from mllm.markov_games.rollout_tree import RolloutTreeRootNode +from mllm.training.credit_methods import ( + get_discounted_returns, + get_discounted_state_visitation_credits, + get_generalized_advantage_estimates, + get_rloo_credits, +) +from mllm.training.tally_metrics import Tally +from mllm.training.tally_tokenwise import ContextualizedTokenwiseTally +from mllm.training.tokenize_chats import * +from mllm.training.tokenize_chats import process_training_chat +from mllm.training.trainer_common import BaseTrainer +from mllm.training.training_data_utils import * +from mllm.training.training_data_utils import ( + TrainingBatch, + TrajectoryBatch, + get_tokenwise_credits, +) +from mllm.utils.resource_context import resource_logger_context + +logger = logging.getLogger(__name__) +logger.addHandler(logging.StreamHandler(sys.stdout)) + + +@dataclass +class TrainingData: + """Caches per-agent trajectory tensors plus their computed advantages.""" + + agent_id: str + main_data: TrajectoryBatch + # list-of-tensors: per rollout advantages with length jT + main_advantages: list[torch.FloatTensor] | None = None + + +class TrainerNaive(BaseTrainer): + def set_agent_trajectory_data( + self, agent_id: str, roots: list[RolloutTreeRootNode] + ) -> None: + """ + Tokenize rollouts for a given agent and cache the tensors used for training. + """ + # Reset per-agent buffers; extend this logic if joint training batches are needed. + self.policy_gradient_data = None + + # Tensorize Chats + rollout_ids = [] + crn_ids = [] # common random number id + batch_input_ids = [] + batch_action_mask = [] + batch_entropy_mask = [] + batch_timesteps = [] + batch_state_ends_mask = [] + batch_engine_log_probs = [] + batch_rewards = [] + for root in roots: + rollout_id = root.id + self.debug_path_list.append( + "mgid:" + str(rollout_id) + "_agent_id:" + agent_id + ) + rollout_ids.append(rollout_id) + crn_ids.append(root.crn_id) + chat, rewards = get_main_chat_list_and_rewards(agent_id=agent_id, root=root) + ( + input_ids, + action_mask, + entropy_mask, + timesteps, + state_ends_mask, + engine_log_probs, + ) = process_training_chat( + tokenizer=self.tokenizer, + chat_history=chat, + entropy_mask_regex=self.entropy_mask_regex, + exploration_prompts_to_remove=self.exploration_prompts_to_remove, + ) + batch_input_ids.append(input_ids) + batch_action_mask.append(action_mask) + batch_entropy_mask.append(entropy_mask) + batch_timesteps.append(timesteps) + batch_state_ends_mask.append(state_ends_mask) + batch_engine_log_probs.append(engine_log_probs) + batch_rewards.append(rewards) + + trajectory_batch = TrajectoryBatch( + rollout_ids=torch.tensor(rollout_ids, dtype=torch.int32), + crn_ids=torch.tensor(crn_ids, dtype=torch.int32), + agent_ids=[agent_id] * len(rollout_ids), + batch_input_ids=batch_input_ids, + batch_action_mask=batch_action_mask, + batch_entropy_mask=batch_entropy_mask, + batch_timesteps=batch_timesteps, + batch_state_ends_mask=batch_state_ends_mask, + batch_rewards=batch_rewards, + batch_engine_log_probs=batch_engine_log_probs, + ) + + # Get Advantages + batch_advantages: torch.FloatTensor = ( + self.get_advantages_with_critic_gradient_accumulation(trajectory_batch) + ) + + # Discount state visitation (the mathematically correct way) + if not self.skip_discounted_state_visitation: + for i in range(len(batch_advantages)): + batch_advantages[i] = get_discounted_state_visitation_credits( + batch_advantages[i].unsqueeze(0), + self.discount_factor, + ).squeeze(0) + + self.training_data[agent_id] = TrainingData( + agent_id=agent_id, + main_data=trajectory_batch, + main_advantages=batch_advantages, + ) + + def receive_advantage_data(self, advantage_packets: list[AdvantagePacket]): + """ + This trainer ignores the advantages of the other trainers. + """ + for agent_id, agent_data in self.training_data.items(): + self.training_data[agent_id] = agent_data.main_data + self.training_data[agent_id].batch_credits = agent_data.main_advantages + + def share_advantage_data(self) -> list[AdvantagePacket]: + """ + Share the advantage data with other agents. + Returns: + AdvantagePacket: The advantage packet containing the agent's advantages. + """ + logger.info(f"Sharing advantage data.") + advantage_packets = [] + for agent_id, agent_data in self.training_data.items(): + advantage_packets.append( + AdvantagePacket( + agent_id=agent_id, + rollout_ids=agent_data.main_data.rollout_ids, + main_advantages=agent_data.main_advantages, + ) + ) + return advantage_packets diff --git a/src_code_for_reproducibility/training/trainer_sum_rewards.py b/src_code_for_reproducibility/training/trainer_sum_rewards.py new file mode 100644 index 0000000000000000000000000000000000000000..336a542bbf13691a9041bcf15da063f3183db4fe --- /dev/null +++ b/src_code_for_reproducibility/training/trainer_sum_rewards.py @@ -0,0 +1,127 @@ +""" +File: mllm/training/trainer_sum_rewards.py +Summary: Trainer that optimizes the sum-of-rewards objective. +""" + +import logging +import os +import sys +from typing import Union + +import torch +import torch.nn.functional as F +from accelerate import Accelerator +from pandas._libs.tslibs.offsets import CBMonthBegin +from peft import LoraConfig +from torch.nn.utils.rnn import pad_sequence +from transformers import AutoModelForCausalLM, AutoTokenizer + +from mllm.markov_games.rollout_tree import * +from mllm.markov_games.rollout_tree import RolloutTreeRootNode +from mllm.training.credit_methods import ( + get_discounted_returns, + get_discounted_state_visitation_credits, + get_generalized_advantage_estimates, + get_rloo_credits, +) +from mllm.training.tally_metrics import Tally +from mllm.training.tally_rollout import RolloutTally, RolloutTallyItem +from mllm.training.tally_tokenwise import ContextualizedTokenwiseTally +from mllm.training.tokenize_chats import * +from mllm.training.tokenize_chats import process_training_chat +from mllm.training.trainer_common import BaseTrainer +from mllm.training.trainer_independent import TrainerNaive, TrainingData +from mllm.training.training_data_utils import * +from mllm.training.training_data_utils import ( + AdvantagePacket, + TrainingBatch, + TrajectoryBatch, + get_tokenwise_credits, +) +from mllm.utils.resource_context import resource_logger_context + +logger = logging.getLogger(__name__) +logger.addHandler(logging.StreamHandler(sys.stdout)) + + +class TrainerSumRewards(TrainerNaive): + def receive_advantage_data(self, advantage_packets: list[AdvantagePacket]): + """Sum peer advantages onto this agent's advantages to optimize joint reward.""" + logger.info(f"Receiving advantage packets.") + + assert ( + len(advantage_packets) > 0 + ), "At least one advantage packet must be provided." + + for agent_id, agent_data in self.training_data.items(): + coagent_advantage_packets = [ + packet for packet in advantage_packets if packet.agent_id != agent_id + ] + agent_rollout_ids = agent_data.main_data.rollout_ids + agent_advantages = agent_data.main_advantages + co_agent_advantages = [] + for rollout_id in agent_rollout_ids: + for co_agent_packet in coagent_advantage_packets: + if rollout_id in co_agent_packet.rollout_ids: + index = torch.where(rollout_id == co_agent_packet.rollout_ids)[ + 0 + ].item() + co_agent_advantages.append( + co_agent_packet.main_advantages[index] + ) + # assumes that its two player game, with one co-agent + break + assert len(co_agent_advantages) == len(agent_advantages) + B = len(agent_advantages) + assert all( + a.shape[0] == b.shape[0] + for a, b in zip(co_agent_advantages, agent_advantages) + ), "Number of advantages must match in order to sum them up." + + # Get padded tensors (advantage alignment is invariant to padding) + lengths = torch.tensor( + [len(t) for t in agent_advantages], + device=self.device, + dtype=torch.long, + ) + padded_main_advantages = pad_sequence( + agent_advantages, batch_first=True, padding_value=0.0 + ) + + padded_co_agent_advantages = pad_sequence( + co_agent_advantages, batch_first=True, padding_value=0.0 + ) + + # Create training batch data + sum_of_ad_credits = padded_main_advantages + padded_co_agent_advantages + self.rollout_tally.add_metric( + path=["sum_of_ad_credits"], + rollout_tally_item=RolloutTallyItem( + crn_ids=agent_data.main_data.crn_ids, + rollout_ids=agent_data.main_data.rollout_ids, + agent_ids=agent_data.main_data.agent_ids, + metric_matrix=sum_of_ad_credits, + ), + ) + + if not self.skip_discounted_state_visitation: + sum_of_ad_credits = get_discounted_state_visitation_credits( + sum_of_ad_credits, + self.discount_factor, + ) + self.rollout_tally.add_metric( + path=["discounted_state_visitation_credits"], + rollout_tally_item=RolloutTallyItem( + crn_ids=agent_data.main_data.crn_ids, + rollout_ids=agent_data.main_data.rollout_ids, + agent_ids=agent_data.main_data.agent_ids, + metric_matrix=sub_tensors[ + "discounted_state_visitation_credits" + ], + ), + ) + + # Slice back to jagged and convert to tokenwise credits + sum_of_ad_credits = [sum_of_ad_credits[i, : lengths[i]] for i in range(B)] + self.training_data[agent_id] = agent_data.main_data + self.training_data[agent_id].batch_credits = sum_of_ad_credits diff --git a/src_code_for_reproducibility/training/training_data_utils.py b/src_code_for_reproducibility/training/training_data_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..2685ee93b27a1208c9ee36a3605bd3321f4fe9fc --- /dev/null +++ b/src_code_for_reproducibility/training/training_data_utils.py @@ -0,0 +1,395 @@ +""" +File: mllm/training/training_data_utils.py +Summary: Utilities for loading, filtering, and batching training data. +""" + +from dataclasses import dataclass +from typing import Literal, Optional, Tuple + +import torch +from torch.nn.utils.rnn import pad_sequence + +from mllm.markov_games.rollout_tree import ( + ChatTurn, + RolloutTreeBranchNode, + RolloutTreeNode, + RolloutTreeRootNode, +) + + +@dataclass +class AdvantagePacket: + """Message used by trainers to share per-rollout advantages.""" + + agent_id: str + rollout_ids: torch.IntTensor # (B,) + # list-of-tensors + main_advantages: list[torch.FloatTensor] + + +class TrainingChatTurn: + """ + Lightweight ChatTurn variant that records the timestep index alongside role/content. + """ + + def __init__( + self, + time_step: int, + role: str, + agent_id: str, + content: str, + chat_template_token_ids: list[int], + reasoning_content: str, + is_state_end: bool, + out_token_ids: Optional[list[int]] = None, + log_probs: Optional[list[float]] = None, + ) -> None: + self.time_step = time_step + self.role = role + self.agent_id = agent_id + self.content = content + self.chat_template_token_ids = chat_template_token_ids + self.reasoning_content = reasoning_content + self.is_state_end = is_state_end + self.out_token_ids = out_token_ids + self.log_probs = log_probs + + def dict(self): + return { + "time_step": self.time_step, + "role": self.role, + "agent_id": self.agent_id, + "content": self.content, + "chat_template_token_ids": self.chat_template_token_ids, + "reasoning_content": self.reasoning_content, + "is_state_end": self.is_state_end, + "out_token_ids": self.out_token_ids, + "log_probs": self.log_probs, + } + + +def get_main_chat_list_and_rewards( + agent_id: str, root: RolloutTreeRootNode | RolloutTreeNode +) -> Tuple[list[TrainingChatTurn], torch.FloatTensor]: + """ + This method traverses a rollout tree and returns a the list of ChatTurn + for an agent. If it encounters a branch node, it follows the main path. + """ + # Currently follows only the main branch; extend if side branches must be included. + if isinstance(root, RolloutTreeRootNode): + current_node = root.child + else: + current_node = root + + chat = [] + rewards = [] + while current_node is not None: + if isinstance(current_node, RolloutTreeBranchNode): + current_node = current_node.main_child + reward: float = current_node.step_log.simulation_step_log.rewards[agent_id] + rewards.append(reward) + chat_turns: list[TrainingChatTurn] = current_node.step_log.action_logs[ + agent_id + ].chat_turns + chat_turns = [ + TrainingChatTurn(time_step=current_node.time_step, **turn.model_dump()) + for turn in chat_turns + ] + chat.extend(chat_turns) + current_node = current_node.child + return chat, torch.FloatTensor(rewards) + + +def get_tokenwise_credits( + # B := batch size, S := number of tokens / seq. length, T := number of states. `j` stands for jagged (see pytorch nested tensors.) + batch_timesteps: torch.IntTensor | torch.Tensor, # (B, jS), + batch_credits: torch.FloatTensor | torch.Tensor, # (B, jT) +) -> torch.FloatTensor | torch.Tensor: # (B, jS) + """ + Expand per-state credits so every token at that timestep receives the same value. + """ + # The explicit loops keep jagged tensor semantics simple; optimize later if profiling warrants it. + batch_token_credits = [] + for credits, timesteps in zip(batch_credits, batch_timesteps): + token_credits = torch.zeros_like( + timesteps, + dtype=credits.dtype, + device=timesteps.device, + ) + for idx, credit in enumerate(credits): + token_credits[timesteps == idx] = credit + batch_token_credits.append(token_credits) + return batch_token_credits + + +@dataclass +class TrajectoryBatch: + """ + Tensorized batch of trajectories using list-of-tensors for jagged dimensions. + """ + + # B := batch size, S := number of tokens / seq. length, T := number of states. + rollout_ids: torch.IntTensor # (B,) + crn_ids: torch.IntTensor # (B,) + agent_ids: list[str] # (B,) + batch_input_ids: list[torch.LongTensor] # List[(jS,)] + batch_action_mask: list[torch.BoolTensor] # List[(jS,)] + batch_entropy_mask: list[torch.BoolTensor] # List[(jS,)] + batch_timesteps: list[torch.IntTensor] # List[(jS,)] + batch_state_ends_mask: list[torch.BoolTensor] # List[(jS,)] + batch_engine_log_probs: Optional[list[torch.FloatTensor]] # List[(jS,)] + batch_rewards: list[torch.FloatTensor] # List[(jT,)] + batch_credits: Optional[list[torch.FloatTensor]] = None # List[(jS,)] + + def __post_init__(self): + """ + Validate per-sample consistency. + """ + B = self.rollout_ids.shape[0] + assert ( + self.crn_ids.shape[0] == B + ), "RNG IDs must have length equal to batch size." + assert ( + len(self.agent_ids) == B + ), "agent_ids must have length equal to batch size." + assert ( + len(self.batch_input_ids) + == len(self.batch_action_mask) + == len(self.batch_entropy_mask) + == len(self.batch_timesteps) + == len(self.batch_state_ends_mask) + == len(self.batch_engine_log_probs) + == len(self.batch_rewards) + == B + ), "Jagged lists must all have length equal to batch size." + + for b in range(B): + nb_rewards = int(self.batch_rewards[b].shape[0]) + nb_timesteps = int(torch.max(self.batch_timesteps[b]).item()) + 1 + assert ( + nb_rewards == nb_timesteps + ), "Number of rewards and timesteps mismatch." + assert ( + self.batch_input_ids[b].shape[0] + == self.batch_action_mask[b].shape[0] + == self.batch_entropy_mask[b].shape[0] + == self.batch_engine_log_probs[b].shape[0] + == self.batch_timesteps[b].shape[0] + ), "Tensors must have the same shape along the jagged dimension." + assert ( + int(self.batch_state_ends_mask[b].sum()) + == self.batch_rewards[b].shape[0] + ), "Number of rewards must match number of state ends." + + """ + Entries: + Here, we ignore the batch dimension. + input_ids: + All of the tokens of both the user and the assistant, flattened. + action_mask: + Set to true on the tokens of the assistant (tokens generated by the model). + timesteps: + Therefore, max(timesteps) = Ns - 1. + state_ends_idx: + Indices of the tokens at which state descriptions end. + rewards: + rewards[t] := R_t(s_t, a_t) + Example: + position: "0 1 2 3 4 5 6 7 8 9 10 11 12 13 14" + input_ids: "U U U a a a U a U a a a U U U" (U := User, a := Assistant) + action_mask: "x x x ✓ ✓ ✓ x ✓ x ✓ ✓ ✓ x x x" + timestep: "0 0 0 0 0 0 1 1 1 1 1 1 2 2 2" + state_ends_dx: [2, 6, 14] + rewards: [r0, r1, r2] + """ + + def __getitem__(self, key) -> "TrajectoryBatch": + if isinstance(key, slice): + return TrajectoryBatch( + rollout_ids=self.rollout_ids.__getitem__(key), + crn_ids=self.crn_ids.__getitem__(key), + agent_ids=self.agent_ids[key], + batch_input_ids=self.batch_input_ids[key], + batch_action_mask=self.batch_action_mask[key], + batch_entropy_mask=self.batch_entropy_mask[key], + batch_timesteps=self.batch_timesteps[key], + batch_state_ends_mask=self.batch_state_ends_mask[key], + batch_engine_log_probs=self.batch_engine_log_probs[key], + batch_rewards=self.batch_rewards[key], + batch_credits=self.batch_credits[key] if self.batch_credits else None, + ) + + def __len__(self): + return len(self.batch_input_ids) + + def to(self, device): + self.rollout_ids = self.rollout_ids.to(device) + self.crn_ids = self.crn_ids.to(device) + self.batch_input_ids = [t.to(device) for t in self.batch_input_ids] + self.batch_action_mask = [t.to(device) for t in self.batch_action_mask] + self.batch_entropy_mask = [t.to(device) for t in self.batch_entropy_mask] + self.batch_timesteps = [t.to(device) for t in self.batch_timesteps] + self.batch_state_ends_mask = [t.to(device) for t in self.batch_state_ends_mask] + self.batch_engine_log_probs = [ + t.to(device) for t in self.batch_engine_log_probs + ] + self.batch_rewards = [t.to(device) for t in self.batch_rewards] + self.batch_credits = ( + [t.to(device) for t in self.batch_credits] if self.batch_credits else None + ) + + def get_padded_tensors_for_critic(self): + """ + Returns: + padded_batch_input_ids: (B, P) + padded_batch_state_ends_mask: (B, P) + timestep_counts: (B,) tensor of ints indicating number of states per sample + """ + padded_batch_input_ids = pad_sequence( + self.batch_input_ids, batch_first=True, padding_value=0 + ) + padded_batch_state_ends_mask = pad_sequence( + self.batch_state_ends_mask, batch_first=True, padding_value=0 + ).bool() + # number of states equals number of True in state_ends_mask + timestep_counts = torch.tensor( + [int(mask.sum().item()) for mask in self.batch_state_ends_mask], + device=padded_batch_input_ids.device, + dtype=torch.long, + ) + return padded_batch_input_ids, padded_batch_state_ends_mask, timestep_counts + + +timestep = int + + +@dataclass +class PaddedTensorTrainingBatch: + """Helper struct returned by ``TrainingBatch.get_padded_tensors``.""" + + batch_input_ids: torch.LongTensor | torch.Tensor + batch_action_mask: torch.BoolTensor | torch.Tensor + batch_entropy_mask: Optional[torch.BoolTensor | torch.Tensor] + batch_credits: torch.FloatTensor | torch.Tensor + batch_engine_log_probs: torch.FloatTensor | torch.Tensor + batch_timesteps: torch.IntTensor | torch.Tensor + + def __len__(self): + return self.batch_input_ids.shape[0] + + def to(self, device): + self.batch_input_ids = self.batch_input_ids.to(device) + self.batch_action_mask = self.batch_action_mask.to(device) + self.batch_entropy_mask = self.batch_entropy_mask.to(device) + self.batch_credits = self.batch_credits.to(device) + self.batch_engine_log_probs = self.batch_engine_log_probs.to(device) + self.batch_timesteps = self.batch_timesteps.to(device) + + +@dataclass +class TrainingBatch: + rollout_ids: torch.IntTensor | torch.Tensor # (B,) + batch_input_ids: list[torch.LongTensor] # List[(jS,)] + batch_action_mask: list[torch.BoolTensor] # List[(jS,)] + batch_entropy_mask: Optional[list[torch.BoolTensor]] # List[(jS,)] + batch_credits: list[torch.FloatTensor] # List[(jS,)] + batch_engine_log_probs: list[torch.FloatTensor] # List[(jS,)] + batch_timesteps: list[torch.IntTensor] # List[(jS,)] + + def __post_init__(self): + # Ensure batch dimension is present + assert ( + len(self.batch_input_ids) + == len(self.batch_action_mask) + == len(self.batch_entropy_mask) + == len(self.batch_credits) + == len(self.batch_engine_log_probs) + == len(self.batch_timesteps) + == self.rollout_ids.shape[0] + ), "Jagged lists must all have length equal to batch size." + for inp, mask, cred, engine_log_prob, timestep in zip( + self.batch_input_ids, + self.batch_action_mask, + self.batch_credits, + self.batch_engine_log_probs, + self.batch_timesteps, + ): + assert ( + inp.shape[0] + == mask.shape[0] + == cred.shape[0] + == engine_log_prob.shape[0] + == timestep.shape[0] + ), "Tensors must have the same shapes along the jagged dimension." + + def __getitem__(self, key) -> "TrainingBatch": + if isinstance(key, slice): + return TrainingBatch( + rollout_ids=self.rollout_ids.__getitem__(key), + batch_input_ids=self.batch_input_ids[key], + batch_action_mask=self.batch_action_mask[key], + batch_entropy_mask=self.batch_entropy_mask[key], + batch_credits=self.batch_credits[key], + batch_engine_log_probs=self.batch_engine_log_probs[key], + batch_timesteps=self.batch_timesteps[key], + ) + + def __len__(self): + return len(self.batch_input_ids) + + def to(self, device): + self.rollout_ids = self.rollout_ids.to(device) + self.batch_input_ids = [t.to(device) for t in self.batch_input_ids] + self.batch_action_mask = [t.to(device) for t in self.batch_action_mask] + self.batch_entropy_mask = [t.to(device) for t in self.batch_entropy_mask] + self.batch_credits = [t.to(device) for t in self.batch_credits] + self.batch_engine_log_probs = [ + t.to(device) for t in self.batch_engine_log_probs + ] + self.batch_timesteps = [t.to(device) for t in self.batch_timesteps] + + def get_padded_tensors(self, padding: float = 0.0): + """ + Materialize right-padded tensors so PyTorch ops can run on uniform shapes. + """ + padded_batch_input_ids = pad_sequence( + self.batch_input_ids, batch_first=True, padding_value=int(padding) + ) + padded_batch_action_mask = pad_sequence( + [m.to(dtype=torch.bool) for m in self.batch_action_mask], + batch_first=True, + padding_value=False, + ) + padded_batch_entropy_mask = pad_sequence( + self.batch_entropy_mask, batch_first=True, padding_value=False + ) + padded_batch_credits = pad_sequence( + self.batch_credits, batch_first=True, padding_value=float(padding) + ) + padded_batch_engine_log_probs = pad_sequence( + self.batch_engine_log_probs, batch_first=True, padding_value=float(padding) + ) + padded_batch_timesteps = pad_sequence( + self.batch_timesteps, batch_first=True, padding_value=0 + ) + + return PaddedTensorTrainingBatch( + padded_batch_input_ids, + padded_batch_action_mask, + padded_batch_entropy_mask, + padded_batch_credits, + padded_batch_engine_log_probs, + padded_batch_timesteps, + ) + + def append(self, other: "TrainingBatch"): + self.rollout_ids = torch.cat([self.rollout_ids, other.rollout_ids]) + self.batch_input_ids.extend(other.batch_input_ids) + self.batch_action_mask.extend(other.batch_action_mask) + self.batch_entropy_mask.extend(other.batch_entropy_mask) + self.batch_credits.extend(other.batch_credits) + self.batch_engine_log_probs.extend(other.batch_engine_log_probs) + self.batch_timesteps.extend(other.batch_timesteps) + + +timestep = int diff --git a/src_code_for_reproducibility/utils/__init__.py b/src_code_for_reproducibility/utils/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..f8f86a6250af7b404a43dc38a7d58aef50dbeb6d --- /dev/null +++ b/src_code_for_reproducibility/utils/__init__.py @@ -0,0 +1,4 @@ +""" +File: mllm/utils/__init__.py +Summary: Utility package exposing helper modules. +""" diff --git a/src_code_for_reproducibility/utils/__pycache__/__init__.cpython-312.pyc b/src_code_for_reproducibility/utils/__pycache__/__init__.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..e63b428c00db55d1fd97a390ba95c8e0e2b8f6a7 Binary files /dev/null and b/src_code_for_reproducibility/utils/__pycache__/__init__.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/utils/__pycache__/dict_get_path.cpython-312.pyc b/src_code_for_reproducibility/utils/__pycache__/dict_get_path.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..bf519a8c2dd0e9288fbeec449932cb32f838bea2 Binary files /dev/null and b/src_code_for_reproducibility/utils/__pycache__/dict_get_path.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/utils/__pycache__/get_coagent_id.cpython-312.pyc b/src_code_for_reproducibility/utils/__pycache__/get_coagent_id.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..a954fa00312e4c613fe3feb2fe2248988dc6f00d Binary files /dev/null and b/src_code_for_reproducibility/utils/__pycache__/get_coagent_id.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/utils/__pycache__/resource_context.cpython-312.pyc b/src_code_for_reproducibility/utils/__pycache__/resource_context.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..084c8810b7f0ced9d7ecc96db6933f4bd0a8931d Binary files /dev/null and b/src_code_for_reproducibility/utils/__pycache__/resource_context.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/utils/__pycache__/rollout_tree_gather_utils.cpython-312.pyc b/src_code_for_reproducibility/utils/__pycache__/rollout_tree_gather_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..348e542fbd081961b48ba11f57ac98c99d11e41c Binary files /dev/null and b/src_code_for_reproducibility/utils/__pycache__/rollout_tree_gather_utils.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/utils/__pycache__/rollout_tree_stats.cpython-312.pyc b/src_code_for_reproducibility/utils/__pycache__/rollout_tree_stats.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..34a48d6500c7c7e62e47a7def5f05e8824eaac0c Binary files /dev/null and b/src_code_for_reproducibility/utils/__pycache__/rollout_tree_stats.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/utils/__pycache__/short_id_gen.cpython-312.pyc b/src_code_for_reproducibility/utils/__pycache__/short_id_gen.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6f20495b7f4d59902c0ff611c84ba21a2311e77b Binary files /dev/null and b/src_code_for_reproducibility/utils/__pycache__/short_id_gen.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/utils/__pycache__/stat_pack.cpython-312.pyc b/src_code_for_reproducibility/utils/__pycache__/stat_pack.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..9a454b258bff6fdfbd14a21b1b02cf2e957747f8 Binary files /dev/null and b/src_code_for_reproducibility/utils/__pycache__/stat_pack.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/utils/__pycache__/update_start_epoch.cpython-312.pyc b/src_code_for_reproducibility/utils/__pycache__/update_start_epoch.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6f9d3c6bca8b73cbb23dff3263b3a5f12f598599 Binary files /dev/null and b/src_code_for_reproducibility/utils/__pycache__/update_start_epoch.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/utils/__pycache__/wandb_utils.cpython-312.pyc b/src_code_for_reproducibility/utils/__pycache__/wandb_utils.cpython-312.pyc new file mode 100644 index 0000000000000000000000000000000000000000..e4da2d0a7c94f30a7b0dcb724c0e4eee16bebd6d Binary files /dev/null and b/src_code_for_reproducibility/utils/__pycache__/wandb_utils.cpython-312.pyc differ diff --git a/src_code_for_reproducibility/utils/get_coagent_id.py b/src_code_for_reproducibility/utils/get_coagent_id.py new file mode 100644 index 0000000000000000000000000000000000000000..f51674757ebb4ba1b0c18a36dd4ea9257564f890 --- /dev/null +++ b/src_code_for_reproducibility/utils/get_coagent_id.py @@ -0,0 +1,10 @@ +""" +File: mllm/utils/get_coagent_id.py +Summary: Helper for deriving co-agent identifiers from rollout metadata. +""" + + +def get_coagent_id(ids: list[str], agent_id: str) -> str | None: + for id in ids: + if id != agent_id: + return id diff --git a/src_code_for_reproducibility/utils/rollout_tree_stats.py b/src_code_for_reproducibility/utils/rollout_tree_stats.py new file mode 100644 index 0000000000000000000000000000000000000000..4725160156230d7efb89588c765fb5b63a7bbbe1 --- /dev/null +++ b/src_code_for_reproducibility/utils/rollout_tree_stats.py @@ -0,0 +1,55 @@ +""" +File: mllm/utils/rollout_tree_stats.py +Summary: Computes descriptive statistics from rollout tree collections. +""" + +from typing import Any, Callable, List, Tuple + +from mllm.markov_games.rollout_tree import RolloutTreeRootNode +from mllm.markov_games.simulation import SimulationStepLog +from mllm.utils.rollout_tree_gather_utils import ( + gather_simulation_step_logs, + get_rollout_tree_paths, +) +from mllm.utils.stat_pack import StatPack + + +def get_rollout_tree_stat_tally( + rollout_tree: RolloutTreeRootNode, + metrics: List[Callable[[SimulationStepLog], List[Tuple[str, float]]]], +) -> StatPack: + stat_tally = StatPack() + # get simulation step logs + node_list = get_rollout_tree_paths(rollout_tree)[0] + simulation_step_logs = gather_simulation_step_logs(node_list) + for simulation_step_log in simulation_step_logs: + for metric in metrics: + metric_result = metric(simulation_step_log) + if metric_result is not None: + for key, value in metric_result: + stat_tally.add_stat(key, value) + return stat_tally + + +def get_rollout_tree_mean_stats( + rollout_tree: RolloutTreeRootNode, metrics: List[Callable[[SimulationStepLog], Any]] +) -> StatPack: + """Get the mean stats for a rollout tree.""" + stat_tally = get_rollout_tree_stat_tally(rollout_tree, metrics) + return stat_tally.mean() + + +def get_mean_rollout_tree_stats( + rollout_trees: List[RolloutTreeRootNode], + metrics: List[Callable[[SimulationStepLog], Any]], +) -> StatPack: + """Get the mean stats for a list of rollout trees.""" + # Compute per-rollout means first, then aggregate them across the entire batch. + stat_tallies = [ + get_rollout_tree_mean_stats(rollout_tree, metrics) + for rollout_tree in rollout_trees + ] + mean_stat_tally = StatPack() + for stat_tally in stat_tallies: + mean_stat_tally.add_stats(stat_tally) + return mean_stat_tally.mean() diff --git a/src_code_for_reproducibility/utils/short_id_gen.py b/src_code_for_reproducibility/utils/short_id_gen.py new file mode 100644 index 0000000000000000000000000000000000000000..6c08ffdc3362c767ea8916496ea5b0e1c01dbd7e --- /dev/null +++ b/src_code_for_reproducibility/utils/short_id_gen.py @@ -0,0 +1,16 @@ +""" +File: mllm/utils/short_id_gen.py +Summary: Generates short unique identifiers for experiment assets. +""" + +import uuid + + +def generate_short_id() -> int: + """ + Generates a short unique ID for tracking adapter versions. + + Returns: + int: An 8-digit integer ID. + """ + return int(str(uuid.uuid4().int)[:8]) diff --git a/src_code_for_reproducibility/utils/update_start_epoch.py b/src_code_for_reproducibility/utils/update_start_epoch.py new file mode 100644 index 0000000000000000000000000000000000000000..24a2ab0ae28dc2f5bccc61a98b920e165bc9d813 --- /dev/null +++ b/src_code_for_reproducibility/utils/update_start_epoch.py @@ -0,0 +1,17 @@ +""" +File: mllm/utils/update_start_epoch.py +Summary: Updates persisted start-epoch metadata when resuming runs. +""" + +import os + + +# During run, set hydra.run.dir=./outputs/{folder} +def update_start_epoch(cfg, output_directory): + if cfg["experiment"]["resume_experiment"]: + folders = [ + f for f in os.listdir(output_directory) if f.startswith("iteration_") + ] + iterations = [int(f.split("_")[1]) for f in folders] if folders else [0] + cfg["experiment"]["start_epoch"] = max(iterations) + return None