wandb: Tracking run with wandb version 0.28.0
wandb: W&B syncing is set to `offline` in this directory. Run `wandb online` or set WANDB_MODE=online to enable cloud syncing.
wandb: Run data is saved locally in outputs/healed/opd_warm_fixed_keep50/wandb/offline-run-20260802_065345-ubll0rd3
wandb: View this run in the terminal with `wandb leet`

Loading checkpoint shards:   0%|          | 0/3 [00:00<?, ?it/s]
Loading checkpoint shards:  33%|███▎      | 1/3 [00:00<00:00,  8.84it/s]
Loading checkpoint shards:  67%|██████▋   | 2/3 [00:00<00:00,  8.77it/s]
Loading checkpoint shards: 100%|██████████| 3/3 [00:00<00:00,  9.60it/s]

Loading checkpoint shards:   0%|          | 0/2 [00:00<?, ?it/s]
Loading checkpoint shards:  50%|█████     | 1/2 [00:00<00:00,  6.22it/s]
Loading checkpoint shards: 100%|██████████| 2/2 [00:00<00:00,  8.26it/s]
reference anchor loaded from outputs/healed/keep50_warmup_fixed_s1224/step0150 on cuda:0 (beta=0.05)

Loading checkpoint shards:   0%|          | 0/2 [00:00<?, ?it/s]
Loading checkpoint shards:  50%|█████     | 1/2 [00:00<00:00,  6.12it/s]
Loading checkpoint shards: 100%|██████████| 2/2 [00:00<00:00,  8.28it/s]
starting vllm rollout server on GPU 2 (port 8377) ...
vllm server healthy in 36s
mixture distillation: 58360 gold trajectories from outputs/teacher_trajectories/dolci_combined_top128, lambda=0.5 decay=0.0

Filter:   0%|          | 0/169964 [00:00<?, ? examples/s]
Filter:   1%|          | 2000/169964 [00:00<00:09, 17826.37 examples/s]
Filter:   2%|▏         | 4000/169964 [00:00<00:09, 18427.83 examples/s]
Filter:   4%|▎         | 6000/169964 [00:00<00:08, 18411.37 examples/s]
Filter:   5%|▍         | 8000/169964 [00:00<00:08, 18281.65 examples/s]
Filter:   6%|▌         | 10000/169964 [00:00<00:08, 18093.92 examples/s]
Filter:   7%|▋         | 12000/169964 [00:00<00:08, 18120.46 examples/s]
Filter:   8%|▊         | 14000/169964 [00:00<00:08, 18160.07 examples/s]
Filter:   9%|▉         | 16000/169964 [00:00<00:08, 18157.59 examples/s]
Filter:  11%|█         | 18000/169964 [00:00<00:08, 18158.62 examples/s]
Filter:  12%|█▏        | 20000/169964 [00:01<00:08, 18365.15 examples/s]
Filter:  14%|█▍        | 24000/169964 [00:01<00:06, 22038.74 examples/s]
Filter:  16%|█▋        | 28000/169964 [00:01<00:05, 24198.29 examples/s]
Filter:  19%|█▉        | 32000/169964 [00:01<00:05, 25469.29 examples/s]
Filter:  21%|██        | 36000/169964 [00:01<00:05, 26281.40 examples/s]
Filter:  24%|██▎       | 40000/169964 [00:01<00:04, 26781.18 examples/s]
Filter:  26%|██▌       | 44000/169964 [00:01<00:04, 28044.75 examples/s]
Filter:  28%|██▊       | 48000/169964 [00:02<00:04, 29103.97 examples/s]
Filter:  31%|███       | 52000/169964 [00:02<00:03, 29897.17 examples/s]
Filter:  32%|███▏      | 55000/169964 [00:02<00:03, 28908.51 examples/s]
Filter:  34%|███▍      | 58000/169964 [00:02<00:04, 25929.09 examples/s]
Filter:  36%|███▌      | 61000/169964 [00:02<00:04, 23665.31 examples/s]
Filter:  38%|███▊      | 64000/169964 [00:02<00:04, 22401.88 examples/s]
Filter:  39%|███▉      | 67000/169964 [00:02<00:04, 21275.45 examples/s]
Filter:  41%|████      | 70000/169964 [00:03<00:04, 22095.84 examples/s]
Filter:  44%|████▎     | 74000/169964 [00:03<00:04, 23908.49 examples/s]
Filter:  46%|████▌     | 78000/169964 [00:03<00:03, 25028.53 examples/s]
Filter:  48%|████▊     | 82000/169964 [00:03<00:03, 25655.48 examples/s]
Filter:  50%|█████     | 85000/169964 [00:03<00:03, 23472.60 examples/s]
Filter:  52%|█████▏    | 88000/169964 [00:03<00:04, 18654.75 examples/s]
Filter:  54%|█████▎    | 91000/169964 [00:04<00:04, 16028.27 examples/s]
Filter:  55%|█████▍    | 93000/169964 [00:04<00:05, 14947.32 examples/s]
Filter:  56%|█████▌    | 95000/169964 [00:04<00:05, 14071.13 examples/s]
Filter:  57%|█████▋    | 97000/169964 [00:04<00:05, 13475.08 examples/s]
Filter:  58%|█████▊    | 99000/169964 [00:04<00:05, 12955.75 examples/s]
Filter:  59%|█████▉    | 101000/169964 [00:04<00:05, 12549.77 examples/s]
Filter:  61%|██████    | 103000/169964 [00:05<00:05, 12272.58 examples/s]
Filter:  62%|██████▏   | 105000/169964 [00:05<00:05, 12205.65 examples/s]
Filter:  63%|██████▎   | 107000/169964 [00:05<00:05, 12017.59 examples/s]
Filter:  64%|██████▍   | 109000/169964 [00:05<00:05, 11979.57 examples/s]
Filter:  65%|██████▌   | 111000/169964 [00:05<00:04, 11879.31 examples/s]
Filter:  66%|██████▋   | 113000/169964 [00:06<00:04, 11876.33 examples/s]
Filter:  68%|██████▊   | 115000/169964 [00:06<00:04, 11784.94 examples/s]
Filter:  69%|██████▉   | 117000/169964 [00:06<00:04, 11875.46 examples/s]
Filter:  70%|███████   | 119000/169964 [00:06<00:04, 11777.10 examples/s]
Filter:  71%|███████   | 121000/169964 [00:06<00:04, 11671.50 examples/s]
Filter:  72%|███████▏  | 123000/169964 [00:06<00:03, 12481.57 examples/s]
Filter:  74%|███████▎  | 125000/169964 [00:06<00:03, 13457.13 examples/s]
Filter:  75%|███████▍  | 127000/169964 [00:07<00:02, 14481.97 examples/s]
Filter:  76%|███████▌  | 129000/169964 [00:07<00:02, 15304.02 examples/s]
Filter:  77%|███████▋  | 131000/169964 [00:07<00:02, 15942.09 examples/s]
Filter:  78%|███████▊  | 133000/169964 [00:07<00:02, 16137.61 examples/s]
Filter:  79%|███████▉  | 135000/169964 [00:07<00:02, 16187.44 examples/s]
Filter:  81%|████████  | 137000/169964 [00:07<00:02, 16413.20 examples/s]
Filter:  82%|████████▏ | 139000/169964 [00:07<00:01, 16185.38 examples/s]
Filter:  83%|████████▎ | 141000/169964 [00:07<00:01, 16366.08 examples/s]
Filter:  84%|████████▍ | 143000/169964 [00:08<00:01, 16673.80 examples/s]
Filter:  85%|████████▌ | 145000/169964 [00:08<00:01, 16922.22 examples/s]
Filter:  86%|████████▋ | 147000/169964 [00:08<00:01, 16981.81 examples/s]
Filter:  88%|████████▊ | 149000/169964 [00:08<00:01, 16994.66 examples/s]
Filter:  89%|████████▉ | 151000/169964 [00:08<00:01, 17187.04 examples/s]
Filter:  90%|█████████ | 153000/169964 [00:08<00:00, 17006.39 examples/s]
Filter:  91%|█████████ | 155000/169964 [00:08<00:00, 16903.61 examples/s]
Filter:  92%|█████████▏| 157000/169964 [00:08<00:00, 16875.08 examples/s]
Filter:  94%|█████████▎| 159000/169964 [00:08<00:00, 17085.05 examples/s]
Filter:  95%|█████████▍| 161000/169964 [00:09<00:00, 16916.21 examples/s]
Filter:  96%|█████████▌| 163000/169964 [00:09<00:00, 16597.31 examples/s]
Filter:  97%|█████████▋| 165000/169964 [00:09<00:00, 16517.32 examples/s]
Filter:  98%|█████████▊| 167000/169964 [00:09<00:00, 16591.80 examples/s]
Filter:  99%|█████████▉| 169000/169964 [00:09<00:00, 16782.62 examples/s]
Filter: 100%|██████████| 169964/169964 [00:09<00:00, 17686.51 examples/s]
The attention mask is not set and cannot be inferred from input because pad token is same as eos token. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results.
168191 prompts | 2627 steps/epoch | 120 total steps | student params 3.70B | teacher overlap=True
{"step": 1, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.44963827088755237, "tokens": 194827, "cumulative_loss_tokens": 194827, "grad_norm": 2.0625, "lr": 2.0000000000000003e-06, "finish_rate": 0.957, "comp_len": 761.0, "dropped_truncated": 0, "gold_loss": 0.2115, "gold_lambda": 0.5, "rep_ratio": 2.401, "t_data_s": 0.0, "t_rollout_s": 56.4, "t_step_s": 182.8, "t_refresh_s": 0.3, "mem_gb": 17.24, "mem_gb_teacher": 20.26}
[eval step 1] sample: 'Misy fitsapana manokana ve mba hanombanana ny fahaiza-mamorona?'
{"step": 1, "gsm8k_n": 256, "gsm8k_quick_chat": 0.65234375, "t_eval_s": 23.8}
{"step": 2, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3940014742786295, "tokens": 187987, "cumulative_loss_tokens": 382814, "grad_norm": 2.0, "lr": 3e-06, "finish_rate": 0.992, "comp_len": 734.3, "dropped_truncated": 0, "gold_loss": 0.2068, "gold_lambda": 0.5, "rep_ratio": 2.369, "t_data_s": 0.0, "t_rollout_s": 56.2, "t_step_s": 157.9, "t_refresh_s": 0.3, "mem_gb": 17.22, "mem_gb_teacher": 20.24}
{"step": 3, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3842378264528005, "tokens": 175719, "cumulative_loss_tokens": 558533, "grad_norm": 1.9296875, "lr": 4.000000000000001e-06, "finish_rate": 0.969, "comp_len": 686.4, "dropped_truncated": 0, "gold_loss": 0.2386, "gold_lambda": 0.5, "rep_ratio": 2.451, "t_data_s": 0.0, "t_rollout_s": 52.3, "t_step_s": 150.2, "t_refresh_s": 0.3, "mem_gb": 17.33, "mem_gb_teacher": 20.27}
{"step": 4, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3169176116394187, "tokens": 183284, "cumulative_loss_tokens": 741817, "grad_norm": 1.703125, "lr": 5e-06, "finish_rate": 0.973, "comp_len": 716.0, "dropped_truncated": 0, "gold_loss": 0.2391, "gold_lambda": 0.5, "rep_ratio": 2.682, "t_data_s": 0.0, "t_rollout_s": 53.4, "t_step_s": 156.8, "t_refresh_s": 0.3, "mem_gb": 17.1, "mem_gb_teacher": 20.22}
{"step": 5, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.31469852179097824, "tokens": 169650, "cumulative_loss_tokens": 911467, "grad_norm": 2.625, "lr": 6e-06, "finish_rate": 0.988, "comp_len": 662.7, "dropped_truncated": 0, "gold_loss": 1.3211, "gold_lambda": 0.5, "rep_ratio": 2.526, "t_data_s": 0.0, "t_rollout_s": 50.1, "t_step_s": 144.7, "t_refresh_s": 0.3, "mem_gb": 17.17, "mem_gb_teacher": 20.24}
{"step": 6, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3012831195711605, "tokens": 182143, "cumulative_loss_tokens": 1093610, "grad_norm": 2.203125, "lr": 7e-06, "finish_rate": 0.949, "comp_len": 711.5, "dropped_truncated": 0, "gold_loss": 0.9775, "gold_lambda": 0.5, "rep_ratio": 2.403, "t_data_s": 0.0, "t_rollout_s": 56.4, "t_step_s": 154.9, "t_refresh_s": 0.3, "mem_gb": 17.26, "mem_gb_teacher": 20.26}
{"step": 7, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2884359588736265, "tokens": 182618, "cumulative_loss_tokens": 1276228, "grad_norm": 2.640625, "lr": 8.000000000000001e-06, "finish_rate": 0.977, "comp_len": 713.4, "dropped_truncated": 0, "gold_loss": 1.3168, "gold_lambda": 0.5, "rep_ratio": 2.266, "t_data_s": 0.0, "t_rollout_s": 53.8, "t_step_s": 154.1, "t_refresh_s": 0.3, "mem_gb": 17.47, "mem_gb_teacher": 20.28}
{"step": 8, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.25864222868998754, "tokens": 182352, "cumulative_loss_tokens": 1458580, "grad_norm": 2.9375, "lr": 9e-06, "finish_rate": 0.969, "comp_len": 712.3, "dropped_truncated": 0, "gold_loss": 1.3305, "gold_lambda": 0.5, "rep_ratio": 2.607, "t_data_s": 0.0, "t_rollout_s": 53.3, "t_step_s": 153.4, "t_refresh_s": 0.3, "mem_gb": 17.53, "mem_gb_teacher": 20.29}
{"step": 9, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.4240309239163299, "tokens": 161654, "cumulative_loss_tokens": 1620234, "grad_norm": 1.8046875, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 631.5, "dropped_truncated": 0, "gold_loss": 0.2139, "gold_lambda": 0.5, "rep_ratio": 2.185, "t_data_s": 0.0, "t_rollout_s": 49.1, "t_step_s": 143.8, "t_refresh_s": 0.3, "mem_gb": 17.52, "mem_gb_teacher": 20.28}
{"step": 10, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3422506856649129, "tokens": 182466, "cumulative_loss_tokens": 1802700, "grad_norm": 1.4453125, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 712.8, "dropped_truncated": 0, "gold_loss": 0.2024, "gold_lambda": 0.5, "rep_ratio": 2.523, "t_data_s": 0.0, "t_rollout_s": 53.5, "t_step_s": 153.2, "t_refresh_s": 0.3, "mem_gb": 17.34, "mem_gb_teacher": 20.25}
[eval step 10] sample: 'Misy fitsapana manokana ve mba hanombanana ny fahaiza-mamorona?'
{"step": 11, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3194563395594054, "tokens": 162668, "cumulative_loss_tokens": 1965368, "grad_norm": 2.875, "lr": 1e-05, "finish_rate": 0.996, "comp_len": 635.4, "dropped_truncated": 0, "gold_loss": 1.4128, "gold_lambda": 0.5, "rep_ratio": 2.353, "t_data_s": 0.0, "t_rollout_s": 49.0, "t_step_s": 144.6, "t_refresh_s": 0.3, "mem_gb": 17.31, "mem_gb_teacher": 20.25}
{"step": 12, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.30020437081658174, "tokens": 172670, "cumulative_loss_tokens": 2138038, "grad_norm": 2.5, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 674.5, "dropped_truncated": 0, "gold_loss": 1.3403, "gold_lambda": 0.5, "rep_ratio": 2.41, "t_data_s": 0.0, "t_rollout_s": 52.5, "t_step_s": 152.4, "t_refresh_s": 0.3, "mem_gb": 17.45, "mem_gb_teacher": 20.3}
{"step": 13, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.4090317845232767, "tokens": 160408, "cumulative_loss_tokens": 2298446, "grad_norm": 3.203125, "lr": 1e-05, "finish_rate": 0.992, "comp_len": 626.6, "dropped_truncated": 0, "gold_loss": 1.4213, "gold_lambda": 0.5, "rep_ratio": 2.23, "t_data_s": 0.0, "t_rollout_s": 47.7, "t_step_s": 141.4, "t_refresh_s": 0.3, "mem_gb": 17.07, "mem_gb_teacher": 20.25}
{"step": 14, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.34899632673731384, "tokens": 159410, "cumulative_loss_tokens": 2457856, "grad_norm": 2.6875, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 622.7, "dropped_truncated": 0, "gold_loss": 1.3901, "gold_lambda": 0.5, "rep_ratio": 2.422, "t_data_s": 0.0, "t_rollout_s": 47.6, "t_step_s": 141.7, "t_refresh_s": 0.3, "mem_gb": 17.12, "mem_gb_teacher": 20.25}
{"step": 15, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.34543543075337274, "tokens": 166862, "cumulative_loss_tokens": 2624718, "grad_norm": 1.15625, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 651.8, "dropped_truncated": 0, "gold_loss": 0.2286, "gold_lambda": 0.5, "rep_ratio": 2.459, "t_data_s": 0.0, "t_rollout_s": 48.7, "t_step_s": 145.8, "t_refresh_s": 0.3, "mem_gb": 17.25, "mem_gb_teacher": 20.24}
{"step": 16, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3638701808236742, "tokens": 154124, "cumulative_loss_tokens": 2778842, "grad_norm": 1.2109375, "lr": 1e-05, "finish_rate": 0.988, "comp_len": 602.0, "dropped_truncated": 0, "gold_loss": 0.2253, "gold_lambda": 0.5, "rep_ratio": 2.369, "t_data_s": 0.0, "t_rollout_s": 47.2, "t_step_s": 141.5, "t_refresh_s": 0.3, "mem_gb": 17.09, "mem_gb_teacher": 20.23}
{"step": 17, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3916856637959853, "tokens": 153093, "cumulative_loss_tokens": 2931935, "grad_norm": 2.0625, "lr": 1e-05, "finish_rate": 0.992, "comp_len": 598.0, "dropped_truncated": 0, "gold_loss": 1.0651, "gold_lambda": 0.5, "rep_ratio": 2.331, "t_data_s": 0.0, "t_rollout_s": 47.5, "t_step_s": 149.3, "t_refresh_s": 0.3, "mem_gb": 17.26, "mem_gb_teacher": 20.27}
{"step": 18, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3789892320845085, "tokens": 167800, "cumulative_loss_tokens": 3099735, "grad_norm": 1.96875, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 655.5, "dropped_truncated": 0, "gold_loss": 1.0761, "gold_lambda": 0.5, "rep_ratio": 2.334, "t_data_s": 0.0, "t_rollout_s": 52.0, "t_step_s": 153.1, "t_refresh_s": 0.3, "mem_gb": 17.18, "mem_gb_teacher": 20.24}
{"step": 19, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.35074403341776794, "tokens": 167173, "cumulative_loss_tokens": 3266908, "grad_norm": 2.390625, "lr": 1e-05, "finish_rate": 0.988, "comp_len": 653.0, "dropped_truncated": 0, "gold_loss": 1.3153, "gold_lambda": 0.5, "rep_ratio": 2.337, "t_data_s": 0.0, "t_rollout_s": 50.2, "t_step_s": 143.5, "t_refresh_s": 0.3, "mem_gb": 17.32, "mem_gb_teacher": 20.25}
{"step": 20, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2626116644125829, "tokens": 190018, "cumulative_loss_tokens": 3456926, "grad_norm": 2.375, "lr": 1e-05, "finish_rate": 0.969, "comp_len": 742.3, "dropped_truncated": 0, "gold_loss": 1.4092, "gold_lambda": 0.5, "rep_ratio": 2.615, "t_data_s": 0.0, "t_rollout_s": 55.8, "t_step_s": 158.7, "t_refresh_s": 0.3, "mem_gb": 17.61, "mem_gb_teacher": 20.29}
[eval step 20] sample: 'Misy fitapana manokana ny fahaiza-mamorona.'
{"step": 20, "gsm8k_n": 256, "gsm8k_quick_chat": 0.6171875, "t_eval_s": 21.4}
{"step": 21, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.36320468789413923, "tokens": 161040, "cumulative_loss_tokens": 3617966, "grad_norm": 1.4921875, "lr": 1e-05, "finish_rate": 0.977, "comp_len": 629.1, "dropped_truncated": 0, "gold_loss": 0.2034, "gold_lambda": 0.5, "rep_ratio": 2.491, "t_data_s": 0.0, "t_rollout_s": 47.5, "t_step_s": 141.8, "t_refresh_s": 0.3, "mem_gb": 17.31, "mem_gb_teacher": 20.25}
{"step": 22, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3560513313296494, "tokens": 186196, "cumulative_loss_tokens": 3804162, "grad_norm": 1.140625, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 727.3, "dropped_truncated": 0, "gold_loss": 0.2012, "gold_lambda": 0.5, "rep_ratio": 2.621, "t_data_s": 0.0, "t_rollout_s": 56.3, "t_step_s": 157.9, "t_refresh_s": 0.3, "mem_gb": 17.48, "mem_gb_teacher": 20.28}
{"step": 23, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.4002737975893586, "tokens": 172291, "cumulative_loss_tokens": 3976453, "grad_norm": 1.5546875, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 673.0, "dropped_truncated": 0, "gold_loss": 0.7825, "gold_lambda": 0.5, "rep_ratio": 2.324, "t_data_s": 0.0, "t_rollout_s": 50.2, "t_step_s": 146.6, "t_refresh_s": 0.3, "mem_gb": 17.11, "mem_gb_teacher": 20.25}
{"step": 24, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3151660050172998, "tokens": 164852, "cumulative_loss_tokens": 4141305, "grad_norm": 1.6875, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 644.0, "dropped_truncated": 0, "gold_loss": 0.801, "gold_lambda": 0.5, "rep_ratio": 2.331, "t_data_s": 0.0, "t_rollout_s": 49.6, "t_step_s": 143.3, "t_refresh_s": 0.3, "mem_gb": 17.59, "mem_gb_teacher": 20.31}
{"step": 25, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.313829974284698, "tokens": 196304, "cumulative_loss_tokens": 4337609, "grad_norm": 2.234375, "lr": 1e-05, "finish_rate": 0.965, "comp_len": 766.8, "dropped_truncated": 0, "gold_loss": 1.6377, "gold_lambda": 0.5, "rep_ratio": 2.688, "t_data_s": 0.0, "t_rollout_s": 57.5, "t_step_s": 156.7, "t_refresh_s": 0.3, "mem_gb": 17.76, "mem_gb_teacher": 20.32}
{"step": 26, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.27340404820149083, "tokens": 174461, "cumulative_loss_tokens": 4512070, "grad_norm": 1.9375, "lr": 1e-05, "finish_rate": 0.977, "comp_len": 681.5, "dropped_truncated": 0, "gold_loss": 1.3934, "gold_lambda": 0.5, "rep_ratio": 2.549, "t_data_s": 0.0, "t_rollout_s": 52.1, "t_step_s": 146.5, "t_refresh_s": 0.3, "mem_gb": 17.47, "mem_gb_teacher": 20.28}
{"step": 27, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3913407735489407, "tokens": 181479, "cumulative_loss_tokens": 4693549, "grad_norm": 1.8671875, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 708.9, "dropped_truncated": 0, "gold_loss": 1.0072, "gold_lambda": 0.5, "rep_ratio": 2.356, "t_data_s": 0.0, "t_rollout_s": 52.5, "t_step_s": 146.6, "t_refresh_s": 0.3, "mem_gb": 17.33, "mem_gb_teacher": 20.26}
{"step": 28, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3790620736299349, "tokens": 177593, "cumulative_loss_tokens": 4871142, "grad_norm": 1.6015625, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 693.7, "dropped_truncated": 0, "gold_loss": 1.04, "gold_lambda": 0.5, "rep_ratio": 2.369, "t_data_s": 0.0, "t_rollout_s": 51.6, "t_step_s": 147.2, "t_refresh_s": 0.3, "mem_gb": 17.29, "mem_gb_teacher": 20.25}
{"step": 29, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3367005885269224, "tokens": 169115, "cumulative_loss_tokens": 5040257, "grad_norm": 1.34375, "lr": 1e-05, "finish_rate": 0.957, "comp_len": 660.6, "dropped_truncated": 0, "gold_loss": 0.2224, "gold_lambda": 0.5, "rep_ratio": 2.308, "t_data_s": 0.0, "t_rollout_s": 52.3, "t_step_s": 150.0, "t_refresh_s": 0.3, "mem_gb": 17.5, "mem_gb_teacher": 20.28}
{"step": 30, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3335959742327791, "tokens": 192932, "cumulative_loss_tokens": 5233189, "grad_norm": 1.765625, "lr": 1e-05, "finish_rate": 0.961, "comp_len": 753.6, "dropped_truncated": 0, "gold_loss": 0.1972, "gold_lambda": 0.5, "rep_ratio": 2.476, "t_data_s": 0.0, "t_rollout_s": 57.4, "t_step_s": 161.6, "t_refresh_s": 0.3, "mem_gb": 17.65, "mem_gb_teacher": 20.3}
[eval step 30] sample: 'Misy fitapana manokana ny fahaiza-mamorona.'
{"step": 31, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3171257981787659, "tokens": 191095, "cumulative_loss_tokens": 5424284, "grad_norm": 1.8046875, "lr": 1e-05, "finish_rate": 0.969, "comp_len": 746.5, "dropped_truncated": 0, "gold_loss": 1.2779, "gold_lambda": 0.5, "rep_ratio": 2.497, "t_data_s": 0.0, "t_rollout_s": 56.9, "t_step_s": 156.9, "t_refresh_s": 0.3, "mem_gb": 17.18, "mem_gb_teacher": 20.26}
{"step": 32, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3311199295843264, "tokens": 187659, "cumulative_loss_tokens": 5611943, "grad_norm": 1.6171875, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 733.0, "dropped_truncated": 0, "gold_loss": 1.0749, "gold_lambda": 0.5, "rep_ratio": 2.448, "t_data_s": 0.0, "t_rollout_s": 56.4, "t_step_s": 161.5, "t_refresh_s": 0.3, "mem_gb": 17.37, "mem_gb_teacher": 20.29}
{"step": 33, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3653596817669372, "tokens": 162176, "cumulative_loss_tokens": 5774119, "grad_norm": 1.96875, "lr": 1e-05, "finish_rate": 0.977, "comp_len": 633.5, "dropped_truncated": 0, "gold_loss": 1.3587, "gold_lambda": 0.5, "rep_ratio": 2.488, "t_data_s": 0.0, "t_rollout_s": 50.0, "t_step_s": 147.9, "t_refresh_s": 0.3, "mem_gb": 17.77, "mem_gb_teacher": 20.31}
{"step": 34, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3303874825869901, "tokens": 175637, "cumulative_loss_tokens": 5949756, "grad_norm": 2.125, "lr": 1e-05, "finish_rate": 0.977, "comp_len": 686.1, "dropped_truncated": 0, "gold_loss": 1.6698, "gold_lambda": 0.5, "rep_ratio": 2.444, "t_data_s": 0.0, "t_rollout_s": 53.0, "t_step_s": 150.9, "t_refresh_s": 0.3, "mem_gb": 17.3, "mem_gb_teacher": 20.27}
{"step": 35, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.338662115701657, "tokens": 164097, "cumulative_loss_tokens": 6113853, "grad_norm": 1.2578125, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 641.0, "dropped_truncated": 0, "gold_loss": 0.1611, "gold_lambda": 0.5, "rep_ratio": 2.39, "t_data_s": 0.0, "t_rollout_s": 48.7, "t_step_s": 140.7, "t_refresh_s": 0.3, "mem_gb": 17.31, "mem_gb_teacher": 20.25}
{"step": 36, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3760627886982097, "tokens": 180353, "cumulative_loss_tokens": 6294206, "grad_norm": 1.171875, "lr": 1e-05, "finish_rate": 0.949, "comp_len": 704.5, "dropped_truncated": 0, "gold_loss": 0.1814, "gold_lambda": 0.5, "rep_ratio": 2.598, "t_data_s": 0.0, "t_rollout_s": 54.3, "t_step_s": 154.0, "t_refresh_s": 0.3, "mem_gb": 17.34, "mem_gb_teacher": 20.27}
{"step": 37, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3777555677864465, "tokens": 179099, "cumulative_loss_tokens": 6473305, "grad_norm": 1.734375, "lr": 1e-05, "finish_rate": 0.988, "comp_len": 699.6, "dropped_truncated": 0, "gold_loss": 1.1694, "gold_lambda": 0.5, "rep_ratio": 2.291, "t_data_s": 0.0, "t_rollout_s": 52.4, "t_step_s": 147.9, "t_refresh_s": 0.3, "mem_gb": 17.28, "mem_gb_teacher": 20.25}
{"step": 38, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2655141159261545, "tokens": 183837, "cumulative_loss_tokens": 6657142, "grad_norm": 1.7890625, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 718.1, "dropped_truncated": 0, "gold_loss": 1.0682, "gold_lambda": 0.5, "rep_ratio": 2.283, "t_data_s": 0.0, "t_rollout_s": 52.4, "t_step_s": 149.0, "t_refresh_s": 0.3, "mem_gb": 16.97, "mem_gb_teacher": 20.22}
{"step": 39, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.26126282006497425, "tokens": 181345, "cumulative_loss_tokens": 6838487, "grad_norm": 1.140625, "lr": 1e-05, "finish_rate": 0.969, "comp_len": 708.4, "dropped_truncated": 0, "gold_loss": 0.1805, "gold_lambda": 0.5, "rep_ratio": 2.463, "t_data_s": 0.0, "t_rollout_s": 53.9, "t_step_s": 152.2, "t_refresh_s": 0.3, "mem_gb": 17.13, "mem_gb_teacher": 20.23}
{"step": 40, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.4208335484820406, "tokens": 166853, "cumulative_loss_tokens": 7005340, "grad_norm": 1.3125, "lr": 1e-05, "finish_rate": 0.977, "comp_len": 651.8, "dropped_truncated": 0, "gold_loss": 0.1644, "gold_lambda": 0.5, "rep_ratio": 2.677, "t_data_s": 0.0, "t_rollout_s": 48.8, "t_step_s": 140.2, "t_refresh_s": 0.3, "mem_gb": 17.35, "mem_gb_teacher": 20.29}
[eval step 40] sample: 'Masy fitapana manokana ve mba hanombanana ny fahaiza-mamorona?\n\nMasy fitapana (Masy) is a traditional Indian dish made from a combination of spices, including flava, a Indian seasoning, and pananas, w'
{"step": 40, "gsm8k_n": 256, "gsm8k_quick_chat": 0.62890625, "t_eval_s": 21.6}
{"step": 41, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.33376682261101664, "tokens": 161263, "cumulative_loss_tokens": 7166603, "grad_norm": 1.765625, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 629.9, "dropped_truncated": 0, "gold_loss": 1.0706, "gold_lambda": 0.5, "rep_ratio": 2.454, "t_data_s": 0.0, "t_rollout_s": 47.2, "t_step_s": 136.0, "t_refresh_s": 0.3, "mem_gb": 17.22, "mem_gb_teacher": 20.26}
{"step": 42, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.34152158943635763, "tokens": 171806, "cumulative_loss_tokens": 7338409, "grad_norm": 1.8125, "lr": 1e-05, "finish_rate": 0.992, "comp_len": 671.1, "dropped_truncated": 0, "gold_loss": 1.0227, "gold_lambda": 0.5, "rep_ratio": 2.453, "t_data_s": 0.0, "t_rollout_s": 50.1, "t_step_s": 141.3, "t_refresh_s": 0.3, "mem_gb": 17.12, "mem_gb_teacher": 20.23}
{"step": 43, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.30709751060902213, "tokens": 173000, "cumulative_loss_tokens": 7511409, "grad_norm": 1.765625, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 675.8, "dropped_truncated": 0, "gold_loss": 1.2339, "gold_lambda": 0.5, "rep_ratio": 2.397, "t_data_s": 0.0, "t_rollout_s": 51.4, "t_step_s": 144.4, "t_refresh_s": 0.3, "mem_gb": 17.07, "mem_gb_teacher": 20.24}
{"step": 44, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.38481537799269605, "tokens": 160690, "cumulative_loss_tokens": 7672099, "grad_norm": 2.046875, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 627.7, "dropped_truncated": 0, "gold_loss": 1.3762, "gold_lambda": 0.5, "rep_ratio": 2.414, "t_data_s": 0.0, "t_rollout_s": 47.9, "t_step_s": 138.1, "t_refresh_s": 0.3, "mem_gb": 17.33, "mem_gb_teacher": 20.25}
{"step": 45, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3474364483801316, "tokens": 163851, "cumulative_loss_tokens": 7835950, "grad_norm": 1.609375, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 640.0, "dropped_truncated": 0, "gold_loss": 1.1411, "gold_lambda": 0.5, "rep_ratio": 2.204, "t_data_s": 0.0, "t_rollout_s": 49.1, "t_step_s": 143.4, "t_refresh_s": 0.3, "mem_gb": 17.42, "mem_gb_teacher": 20.27}
{"step": 46, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2480004573858833, "tokens": 174070, "cumulative_loss_tokens": 8010020, "grad_norm": 1.796875, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 680.0, "dropped_truncated": 0, "gold_loss": 1.1273, "gold_lambda": 0.5, "rep_ratio": 2.398, "t_data_s": 0.0, "t_rollout_s": 50.5, "t_step_s": 144.5, "t_refresh_s": 0.3, "mem_gb": 17.11, "mem_gb_teacher": 20.24}
{"step": 47, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3238008658347637, "tokens": 175419, "cumulative_loss_tokens": 8185439, "grad_norm": 1.15625, "lr": 1e-05, "finish_rate": 0.965, "comp_len": 685.2, "dropped_truncated": 0, "gold_loss": 0.2272, "gold_lambda": 0.5, "rep_ratio": 2.395, "t_data_s": 0.0, "t_rollout_s": 52.7, "t_step_s": 151.6, "t_refresh_s": 0.3, "mem_gb": 17.06, "mem_gb_teacher": 20.24}
{"step": 48, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.26486295279158406, "tokens": 174191, "cumulative_loss_tokens": 8359630, "grad_norm": 1.0078125, "lr": 1e-05, "finish_rate": 0.965, "comp_len": 680.4, "dropped_truncated": 0, "gold_loss": 0.2168, "gold_lambda": 0.5, "rep_ratio": 2.41, "t_data_s": 0.0, "t_rollout_s": 51.6, "t_step_s": 148.9, "t_refresh_s": 0.3, "mem_gb": 17.0, "mem_gb_teacher": 20.23}
{"step": 49, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3201779053440231, "tokens": 175865, "cumulative_loss_tokens": 8535495, "grad_norm": 1.84375, "lr": 1e-05, "finish_rate": 0.988, "comp_len": 687.0, "dropped_truncated": 0, "gold_loss": 1.2798, "gold_lambda": 0.5, "rep_ratio": 2.35, "t_data_s": 0.0, "t_rollout_s": 52.2, "t_step_s": 147.4, "t_refresh_s": 0.3, "mem_gb": 17.22, "mem_gb_teacher": 20.26}
{"step": 50, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3491487042176898, "tokens": 173904, "cumulative_loss_tokens": 8709399, "grad_norm": 1.609375, "lr": 1e-05, "finish_rate": 0.988, "comp_len": 679.3, "dropped_truncated": 0, "gold_loss": 1.1223, "gold_lambda": 0.5, "rep_ratio": 2.19, "t_data_s": 0.0, "t_rollout_s": 52.0, "t_step_s": 146.4, "t_refresh_s": 0.3, "mem_gb": 17.47, "mem_gb_teacher": 20.27}
[eval step 50] sample: 'Masy fitapana manokana ve mba hanombanana ny fahaiza-mamorona?\n\nMasy fitapana (Masy) is a traditional Indian dish made from a combination of spices, including flava, a Indian seasoning, and a Indian v'
{"step": 51, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.4068168763832476, "tokens": 149202, "cumulative_loss_tokens": 8858601, "grad_norm": 1.84375, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 582.8, "dropped_truncated": 0, "gold_loss": 0.9763, "gold_lambda": 0.5, "rep_ratio": 2.411, "t_data_s": 0.0, "t_rollout_s": 44.8, "t_step_s": 130.7, "t_refresh_s": 0.3, "mem_gb": 17.24, "mem_gb_teacher": 20.26}
{"step": 52, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3949036657737358, "tokens": 157801, "cumulative_loss_tokens": 9016402, "grad_norm": 1.5390625, "lr": 1e-05, "finish_rate": 0.996, "comp_len": 616.4, "dropped_truncated": 0, "gold_loss": 0.7635, "gold_lambda": 0.5, "rep_ratio": 2.429, "t_data_s": 0.0, "t_rollout_s": 48.4, "t_step_s": 137.8, "t_refresh_s": 0.3, "mem_gb": 16.94, "mem_gb_teacher": 20.22}
{"step": 53, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3542952813312953, "tokens": 173066, "cumulative_loss_tokens": 9189468, "grad_norm": 1.375, "lr": 1e-05, "finish_rate": 0.961, "comp_len": 676.0, "dropped_truncated": 0, "gold_loss": 0.888, "gold_lambda": 0.5, "rep_ratio": 2.359, "t_data_s": 0.0, "t_rollout_s": 52.3, "t_step_s": 149.8, "t_refresh_s": 0.3, "mem_gb": 17.61, "mem_gb_teacher": 20.31}
{"step": 54, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3200471835924996, "tokens": 183743, "cumulative_loss_tokens": 9373211, "grad_norm": 1.6796875, "lr": 1e-05, "finish_rate": 0.965, "comp_len": 717.7, "dropped_truncated": 0, "gold_loss": 1.0623, "gold_lambda": 0.5, "rep_ratio": 2.51, "t_data_s": 0.0, "t_rollout_s": 54.3, "t_step_s": 153.0, "t_refresh_s": 0.3, "mem_gb": 17.3, "mem_gb_teacher": 20.26}
{"step": 55, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3881143610330633, "tokens": 161627, "cumulative_loss_tokens": 9534838, "grad_norm": 2.515625, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 631.4, "dropped_truncated": 0, "gold_loss": 1.6456, "gold_lambda": 0.5, "rep_ratio": 2.152, "t_data_s": 0.0, "t_rollout_s": 50.7, "t_step_s": 143.3, "t_refresh_s": 0.3, "mem_gb": 17.43, "mem_gb_teacher": 20.29}
{"step": 56, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.37364923851747267, "tokens": 168945, "cumulative_loss_tokens": 9703783, "grad_norm": 1.6953125, "lr": 1e-05, "finish_rate": 0.957, "comp_len": 659.9, "dropped_truncated": 0, "gold_loss": 1.1163, "gold_lambda": 0.5, "rep_ratio": 2.215, "t_data_s": 0.0, "t_rollout_s": 51.2, "t_step_s": 148.9, "t_refresh_s": 0.3, "mem_gb": 17.32, "mem_gb_teacher": 20.25}
{"step": 57, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.4587748102189034, "tokens": 168237, "cumulative_loss_tokens": 9872020, "grad_norm": 1.5390625, "lr": 1e-05, "finish_rate": 0.977, "comp_len": 657.2, "dropped_truncated": 0, "gold_loss": 0.9107, "gold_lambda": 0.5, "rep_ratio": 2.531, "t_data_s": 0.0, "t_rollout_s": 49.2, "t_step_s": 144.3, "t_refresh_s": 0.3, "mem_gb": 17.16, "mem_gb_teacher": 20.23}
{"step": 58, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.26425971459508646, "tokens": 184409, "cumulative_loss_tokens": 10056429, "grad_norm": 1.5, "lr": 1e-05, "finish_rate": 0.957, "comp_len": 720.3, "dropped_truncated": 0, "gold_loss": 1.125, "gold_lambda": 0.5, "rep_ratio": 2.684, "t_data_s": 0.0, "t_rollout_s": 54.7, "t_step_s": 153.4, "t_refresh_s": 0.3, "mem_gb": 17.21, "mem_gb_teacher": 20.25}
{"step": 59, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2719602728308547, "tokens": 189369, "cumulative_loss_tokens": 10245798, "grad_norm": 1.671875, "lr": 1e-05, "finish_rate": 0.945, "comp_len": 739.7, "dropped_truncated": 0, "gold_loss": 0.9619, "gold_lambda": 0.5, "rep_ratio": 2.581, "t_data_s": 0.0, "t_rollout_s": 56.5, "t_step_s": 154.0, "t_refresh_s": 0.3, "mem_gb": 17.25, "mem_gb_teacher": 20.24}
{"step": 60, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.35498727428496857, "tokens": 166983, "cumulative_loss_tokens": 10412781, "grad_norm": 1.84375, "lr": 1e-05, "finish_rate": 0.953, "comp_len": 652.3, "dropped_truncated": 0, "gold_loss": 1.1915, "gold_lambda": 0.5, "rep_ratio": 2.391, "t_data_s": 0.0, "t_rollout_s": 49.4, "t_step_s": 142.7, "t_refresh_s": 0.3, "mem_gb": 17.2, "mem_gb_teacher": 20.25}
[eval step 60] sample: 'Misy fitapana manokana ny fahaiza-mamorona, mba hanombanana, is a term that describes a situation where a person is engaging in activities that are not conducive to their well-being or health, often d'
{"step": 60, "gsm8k_n": 256, "gsm8k_quick_chat": 0.6171875, "t_eval_s": 21.5}
{"step": 61, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.36016367042313613, "tokens": 195255, "cumulative_loss_tokens": 10608036, "grad_norm": 1.1640625, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 762.7, "dropped_truncated": 0, "gold_loss": 0.2184, "gold_lambda": 0.5, "rep_ratio": 2.441, "t_data_s": 0.0, "t_rollout_s": 56.4, "t_step_s": 157.9, "t_refresh_s": 0.3, "mem_gb": 17.14, "mem_gb_teacher": 20.25}
{"step": 62, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2672880692024946, "tokens": 183249, "cumulative_loss_tokens": 10791285, "grad_norm": 1.0625, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 715.8, "dropped_truncated": 0, "gold_loss": 0.2236, "gold_lambda": 0.5, "rep_ratio": 2.47, "t_data_s": 0.0, "t_rollout_s": 54.1, "t_step_s": 151.8, "t_refresh_s": 0.3, "mem_gb": 17.37, "mem_gb_teacher": 20.26}
{"step": 63, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.272046200843007, "tokens": 169354, "cumulative_loss_tokens": 10960639, "grad_norm": 1.078125, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 661.5, "dropped_truncated": 0, "gold_loss": 0.2677, "gold_lambda": 0.5, "rep_ratio": 2.399, "t_data_s": 0.0, "t_rollout_s": 50.5, "t_step_s": 146.3, "t_refresh_s": 0.3, "mem_gb": 16.99, "mem_gb_teacher": 20.22}
{"step": 64, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.36498770492525573, "tokens": 162379, "cumulative_loss_tokens": 11123018, "grad_norm": 1.3359375, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 634.3, "dropped_truncated": 0, "gold_loss": 0.2643, "gold_lambda": 0.5, "rep_ratio": 2.482, "t_data_s": 0.0, "t_rollout_s": 50.7, "t_step_s": 148.6, "t_refresh_s": 0.3, "mem_gb": 17.24, "mem_gb_teacher": 20.25}
{"step": 65, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3866863870528368, "tokens": 155402, "cumulative_loss_tokens": 11278420, "grad_norm": 1.5390625, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 607.0, "dropped_truncated": 0, "gold_loss": 0.8323, "gold_lambda": 0.5, "rep_ratio": 2.378, "t_data_s": 0.0, "t_rollout_s": 47.0, "t_step_s": 136.8, "t_refresh_s": 0.3, "mem_gb": 17.26, "mem_gb_teacher": 20.27}
{"step": 66, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2787221595843288, "tokens": 183694, "cumulative_loss_tokens": 11462114, "grad_norm": 1.875, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 717.6, "dropped_truncated": 0, "gold_loss": 1.1385, "gold_lambda": 0.5, "rep_ratio": 2.538, "t_data_s": 0.0, "t_rollout_s": 53.9, "t_step_s": 152.2, "t_refresh_s": 0.3, "mem_gb": 17.04, "mem_gb_teacher": 20.22}
{"step": 67, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.27035813788552, "tokens": 176040, "cumulative_loss_tokens": 11638154, "grad_norm": 1.71875, "lr": 1e-05, "finish_rate": 0.949, "comp_len": 687.7, "dropped_truncated": 0, "gold_loss": 0.9716, "gold_lambda": 0.5, "rep_ratio": 2.298, "t_data_s": 0.0, "t_rollout_s": 51.3, "t_step_s": 142.2, "t_refresh_s": 0.3, "mem_gb": 17.37, "mem_gb_teacher": 20.28}
{"step": 68, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.28064265166714214, "tokens": 163139, "cumulative_loss_tokens": 11801293, "grad_norm": 1.53125, "lr": 1e-05, "finish_rate": 0.996, "comp_len": 637.3, "dropped_truncated": 0, "gold_loss": 0.9747, "gold_lambda": 0.5, "rep_ratio": 2.236, "t_data_s": 0.0, "t_rollout_s": 48.4, "t_step_s": 137.7, "t_refresh_s": 0.3, "mem_gb": 17.12, "mem_gb_teacher": 20.22}
{"step": 69, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.26903990990510324, "tokens": 171609, "cumulative_loss_tokens": 11972902, "grad_norm": 2.0625, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 670.3, "dropped_truncated": 0, "gold_loss": 1.3383, "gold_lambda": 0.5, "rep_ratio": 2.405, "t_data_s": 0.0, "t_rollout_s": 51.2, "t_step_s": 145.0, "t_refresh_s": 0.3, "mem_gb": 17.11, "mem_gb_teacher": 20.23}
{"step": 70, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3467570015864296, "tokens": 158831, "cumulative_loss_tokens": 12131733, "grad_norm": 2.0625, "lr": 1e-05, "finish_rate": 0.992, "comp_len": 620.4, "dropped_truncated": 0, "gold_loss": 1.2858, "gold_lambda": 0.5, "rep_ratio": 2.309, "t_data_s": 0.0, "t_rollout_s": 48.3, "t_step_s": 136.2, "t_refresh_s": 0.3, "mem_gb": 17.13, "mem_gb_teacher": 20.26}
[eval step 70] sample: 'Misy fitapana manokana ny fahaiza-mamorona, mba hanombanana, is a term used to describe a situation where a person is engaging in a sexual relationship with more than one person at the same time, ofte'
{"step": 71, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3942991111679687, "tokens": 131738, "cumulative_loss_tokens": 12263471, "grad_norm": 1.671875, "lr": 1e-05, "finish_rate": 0.996, "comp_len": 514.6, "dropped_truncated": 0, "gold_loss": 0.9927, "gold_lambda": 0.5, "rep_ratio": 2.479, "t_data_s": 0.0, "t_rollout_s": 42.0, "t_step_s": 123.0, "t_refresh_s": 0.3, "mem_gb": 17.01, "mem_gb_teacher": 20.22}
{"step": 72, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2885391192831585, "tokens": 177469, "cumulative_loss_tokens": 12440940, "grad_norm": 1.71875, "lr": 1e-05, "finish_rate": 0.965, "comp_len": 693.2, "dropped_truncated": 0, "gold_loss": 1.2827, "gold_lambda": 0.5, "rep_ratio": 2.388, "t_data_s": 0.0, "t_rollout_s": 52.2, "t_step_s": 148.6, "t_refresh_s": 0.3, "mem_gb": 17.1, "mem_gb_teacher": 20.24}
{"step": 73, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3441034620278951, "tokens": 153979, "cumulative_loss_tokens": 12594919, "grad_norm": 1.1875, "lr": 1e-05, "finish_rate": 0.988, "comp_len": 601.5, "dropped_truncated": 0, "gold_loss": 0.2327, "gold_lambda": 0.5, "rep_ratio": 2.452, "t_data_s": 0.0, "t_rollout_s": 47.3, "t_step_s": 140.2, "t_refresh_s": 0.3, "mem_gb": 16.98, "mem_gb_teacher": 20.22}
{"step": 74, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2929858381382192, "tokens": 160418, "cumulative_loss_tokens": 12755337, "grad_norm": 1.0703125, "lr": 1e-05, "finish_rate": 0.969, "comp_len": 626.6, "dropped_truncated": 0, "gold_loss": 0.2401, "gold_lambda": 0.5, "rep_ratio": 2.408, "t_data_s": 0.0, "t_rollout_s": 48.7, "t_step_s": 144.0, "t_refresh_s": 0.3, "mem_gb": 17.21, "mem_gb_teacher": 20.24}
{"step": 75, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.35599221217074245, "tokens": 160043, "cumulative_loss_tokens": 12915380, "grad_norm": 1.5703125, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 625.2, "dropped_truncated": 0, "gold_loss": 1.0205, "gold_lambda": 0.5, "rep_ratio": 2.571, "t_data_s": 0.0, "t_rollout_s": 47.8, "t_step_s": 136.3, "t_refresh_s": 0.3, "mem_gb": 17.06, "mem_gb_teacher": 20.22}
{"step": 76, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.24107641524497922, "tokens": 196064, "cumulative_loss_tokens": 13111444, "grad_norm": 1.5078125, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 765.9, "dropped_truncated": 0, "gold_loss": 0.945, "gold_lambda": 0.5, "rep_ratio": 2.558, "t_data_s": 0.0, "t_rollout_s": 56.1, "t_step_s": 154.5, "t_refresh_s": 0.3, "mem_gb": 17.26, "mem_gb_teacher": 20.27}
{"step": 77, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2918642508978692, "tokens": 168701, "cumulative_loss_tokens": 13280145, "grad_norm": 1.484375, "lr": 1e-05, "finish_rate": 0.969, "comp_len": 659.0, "dropped_truncated": 0, "gold_loss": 0.9558, "gold_lambda": 0.5, "rep_ratio": 2.496, "t_data_s": 0.0, "t_rollout_s": 51.7, "t_step_s": 146.1, "t_refresh_s": 0.3, "mem_gb": 17.23, "mem_gb_teacher": 20.24}
{"step": 78, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3819451671292889, "tokens": 133293, "cumulative_loss_tokens": 13413438, "grad_norm": 1.5625, "lr": 1e-05, "finish_rate": 0.988, "comp_len": 520.7, "dropped_truncated": 0, "gold_loss": 1.0728, "gold_lambda": 0.5, "rep_ratio": 2.506, "t_data_s": 0.0, "t_rollout_s": 43.8, "t_step_s": 128.9, "t_refresh_s": 0.3, "mem_gb": 17.52, "mem_gb_teacher": 20.28}
{"step": 79, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2689037877672512, "tokens": 178569, "cumulative_loss_tokens": 13592007, "grad_norm": 1.6953125, "lr": 1e-05, "finish_rate": 0.934, "comp_len": 697.5, "dropped_truncated": 0, "gold_loss": 1.2626, "gold_lambda": 0.5, "rep_ratio": 2.574, "t_data_s": 0.0, "t_rollout_s": 53.3, "t_step_s": 150.3, "t_refresh_s": 0.3, "mem_gb": 17.2, "mem_gb_teacher": 20.24}
{"step": 80, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.31916272058224376, "tokens": 170669, "cumulative_loss_tokens": 13762676, "grad_norm": 1.7109375, "lr": 1e-05, "finish_rate": 0.977, "comp_len": 666.7, "dropped_truncated": 0, "gold_loss": 1.2327, "gold_lambda": 0.5, "rep_ratio": 2.53, "t_data_s": 0.0, "t_rollout_s": 51.5, "t_step_s": 145.4, "t_refresh_s": 0.3, "mem_gb": 17.48, "mem_gb_teacher": 20.28}
[eval step 80] sample: 'Misy fitapana manokana ny fahaiza-mamorona, mba hanombanana, is a term that describes a situation where a person is engaging in activities that are not conducive to their well-being or health, often d'
{"step": 80, "gsm8k_n": 256, "gsm8k_quick_chat": 0.59765625, "t_eval_s": 21.4}
{"step": 81, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2485308842504299, "tokens": 186465, "cumulative_loss_tokens": 13949141, "grad_norm": 1.953125, "lr": 1e-05, "finish_rate": 0.969, "comp_len": 728.4, "dropped_truncated": 0, "gold_loss": 1.1764, "gold_lambda": 0.5, "rep_ratio": 2.547, "t_data_s": 0.0, "t_rollout_s": 55.5, "t_step_s": 152.9, "t_refresh_s": 0.3, "mem_gb": 17.07, "mem_gb_teacher": 20.23}
{"step": 82, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.25967463855320316, "tokens": 190239, "cumulative_loss_tokens": 14139380, "grad_norm": 1.6484375, "lr": 1e-05, "finish_rate": 0.965, "comp_len": 743.1, "dropped_truncated": 0, "gold_loss": 1.3093, "gold_lambda": 0.5, "rep_ratio": 2.526, "t_data_s": 0.0, "t_rollout_s": 57.3, "t_step_s": 158.5, "t_refresh_s": 0.3, "mem_gb": 17.07, "mem_gb_teacher": 20.24}
{"step": 83, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.33700879973768977, "tokens": 182259, "cumulative_loss_tokens": 14321639, "grad_norm": 1.6875, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 711.9, "dropped_truncated": 0, "gold_loss": 1.0788, "gold_lambda": 0.5, "rep_ratio": 2.505, "t_data_s": 0.0, "t_rollout_s": 55.0, "t_step_s": 152.8, "t_refresh_s": 0.3, "mem_gb": 17.47, "mem_gb_teacher": 20.27}
{"step": 84, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2682194592400478, "tokens": 170072, "cumulative_loss_tokens": 14491711, "grad_norm": 1.3984375, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 664.3, "dropped_truncated": 0, "gold_loss": 0.7509, "gold_lambda": 0.5, "rep_ratio": 2.597, "t_data_s": 0.0, "t_rollout_s": 51.4, "t_step_s": 144.0, "t_refresh_s": 0.3, "mem_gb": 17.44, "mem_gb_teacher": 20.27}
{"step": 85, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.23743917622813884, "tokens": 199223, "cumulative_loss_tokens": 14690934, "grad_norm": 1.8359375, "lr": 1e-05, "finish_rate": 0.961, "comp_len": 778.2, "dropped_truncated": 0, "gold_loss": 0.8562, "gold_lambda": 0.5, "rep_ratio": 2.627, "t_data_s": 0.0, "t_rollout_s": 58.4, "t_step_s": 159.8, "t_refresh_s": 0.3, "mem_gb": 17.48, "mem_gb_teacher": 20.28}
{"step": 86, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.40081178832628034, "tokens": 179491, "cumulative_loss_tokens": 14870425, "grad_norm": 1.96875, "lr": 1e-05, "finish_rate": 0.969, "comp_len": 701.1, "dropped_truncated": 0, "gold_loss": 0.7619, "gold_lambda": 0.5, "rep_ratio": 2.334, "t_data_s": 0.0, "t_rollout_s": 55.1, "t_step_s": 154.6, "t_refresh_s": 0.3, "mem_gb": 17.68, "mem_gb_teacher": 20.32}
{"step": 87, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.31119211183407003, "tokens": 161086, "cumulative_loss_tokens": 15031511, "grad_norm": 1.609375, "lr": 1e-05, "finish_rate": 0.965, "comp_len": 629.2, "dropped_truncated": 0, "gold_loss": 1.1431, "gold_lambda": 0.5, "rep_ratio": 2.341, "t_data_s": 0.0, "t_rollout_s": 48.7, "t_step_s": 139.4, "t_refresh_s": 0.3, "mem_gb": 17.28, "mem_gb_teacher": 20.25}
{"step": 88, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.21312830959642717, "tokens": 172945, "cumulative_loss_tokens": 15204456, "grad_norm": 1.53125, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 675.6, "dropped_truncated": 0, "gold_loss": 1.0969, "gold_lambda": 0.5, "rep_ratio": 2.558, "t_data_s": 0.0, "t_rollout_s": 52.2, "t_step_s": 148.0, "t_refresh_s": 0.3, "mem_gb": 17.49, "mem_gb_teacher": 20.28}
{"step": 89, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.32083614396197874, "tokens": 197781, "cumulative_loss_tokens": 15402237, "grad_norm": 1.796875, "lr": 1e-05, "finish_rate": 0.957, "comp_len": 772.6, "dropped_truncated": 0, "gold_loss": 1.1581, "gold_lambda": 0.5, "rep_ratio": 2.371, "t_data_s": 0.0, "t_rollout_s": 59.1, "t_step_s": 158.0, "t_refresh_s": 0.3, "mem_gb": 17.13, "mem_gb_teacher": 20.24}
{"step": 90, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3539003647110563, "tokens": 177862, "cumulative_loss_tokens": 15580099, "grad_norm": 1.8046875, "lr": 1e-05, "finish_rate": 0.957, "comp_len": 694.8, "dropped_truncated": 0, "gold_loss": 1.1751, "gold_lambda": 0.5, "rep_ratio": 2.484, "t_data_s": 0.0, "t_rollout_s": 53.6, "t_step_s": 152.0, "t_refresh_s": 0.3, "mem_gb": 17.37, "mem_gb_teacher": 20.27}
[eval step 90] sample: 'Masy fitapana manokana ny fahaiza-mamorona?'
{"step": 91, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3003133533208709, "tokens": 190730, "cumulative_loss_tokens": 15770829, "grad_norm": 1.1328125, "lr": 1e-05, "finish_rate": 0.977, "comp_len": 745.0, "dropped_truncated": 0, "gold_loss": 0.2162, "gold_lambda": 0.5, "rep_ratio": 2.401, "t_data_s": 0.0, "t_rollout_s": 56.0, "t_step_s": 156.9, "t_refresh_s": 0.3, "mem_gb": 17.01, "mem_gb_teacher": 20.25}
{"step": 92, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3466024366871271, "tokens": 169647, "cumulative_loss_tokens": 15940476, "grad_norm": 1.34375, "lr": 1e-05, "finish_rate": 0.969, "comp_len": 662.7, "dropped_truncated": 0, "gold_loss": 0.2275, "gold_lambda": 0.5, "rep_ratio": 2.384, "t_data_s": 0.0, "t_rollout_s": 50.4, "t_step_s": 144.5, "t_refresh_s": 0.3, "mem_gb": 17.16, "mem_gb_teacher": 20.24}
{"step": 93, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.23514000853877456, "tokens": 167638, "cumulative_loss_tokens": 16108114, "grad_norm": 1.8359375, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 654.8, "dropped_truncated": 0, "gold_loss": 1.6188, "gold_lambda": 0.5, "rep_ratio": 2.248, "t_data_s": 0.0, "t_rollout_s": 50.4, "t_step_s": 145.5, "t_refresh_s": 0.3, "mem_gb": 17.14, "mem_gb_teacher": 20.25}
{"step": 94, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.31567410076907493, "tokens": 176303, "cumulative_loss_tokens": 16284417, "grad_norm": 1.6484375, "lr": 1e-05, "finish_rate": 0.992, "comp_len": 688.7, "dropped_truncated": 0, "gold_loss": 1.4047, "gold_lambda": 0.5, "rep_ratio": 2.276, "t_data_s": 0.0, "t_rollout_s": 52.5, "t_step_s": 150.1, "t_refresh_s": 0.3, "mem_gb": 17.19, "mem_gb_teacher": 20.23}
{"step": 95, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3306478749962856, "tokens": 173098, "cumulative_loss_tokens": 16457515, "grad_norm": 1.6796875, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 676.2, "dropped_truncated": 0, "gold_loss": 1.3459, "gold_lambda": 0.5, "rep_ratio": 2.403, "t_data_s": 0.0, "t_rollout_s": 51.7, "t_step_s": 145.0, "t_refresh_s": 0.3, "mem_gb": 17.25, "mem_gb_teacher": 20.24}
{"step": 96, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2879815189713935, "tokens": 176694, "cumulative_loss_tokens": 16634209, "grad_norm": 1.4140625, "lr": 1e-05, "finish_rate": 0.977, "comp_len": 690.2, "dropped_truncated": 0, "gold_loss": 1.2006, "gold_lambda": 0.5, "rep_ratio": 2.383, "t_data_s": 0.0, "t_rollout_s": 51.7, "t_step_s": 148.5, "t_refresh_s": 0.3, "mem_gb": 17.54, "mem_gb_teacher": 20.28}
{"step": 97, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.23541706224584372, "tokens": 193571, "cumulative_loss_tokens": 16827780, "grad_norm": 2.34375, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 756.1, "dropped_truncated": 0, "gold_loss": 1.2615, "gold_lambda": 0.5, "rep_ratio": 2.438, "t_data_s": 0.0, "t_rollout_s": 57.9, "t_step_s": 158.2, "t_refresh_s": 0.3, "mem_gb": 17.75, "mem_gb_teacher": 20.32}
{"step": 98, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.34905908556710274, "tokens": 149760, "cumulative_loss_tokens": 16977540, "grad_norm": 1.75, "lr": 1e-05, "finish_rate": 0.992, "comp_len": 585.0, "dropped_truncated": 0, "gold_loss": 1.1712, "gold_lambda": 0.5, "rep_ratio": 2.325, "t_data_s": 0.0, "t_rollout_s": 47.2, "t_step_s": 139.0, "t_refresh_s": 0.3, "mem_gb": 17.11, "mem_gb_teacher": 20.28}
{"step": 99, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2776076079423005, "tokens": 183255, "cumulative_loss_tokens": 17160795, "grad_norm": 0.9765625, "lr": 1e-05, "finish_rate": 0.988, "comp_len": 715.8, "dropped_truncated": 0, "gold_loss": 0.2535, "gold_lambda": 0.5, "rep_ratio": 2.47, "t_data_s": 0.0, "t_rollout_s": 53.3, "t_step_s": 153.1, "t_refresh_s": 0.3, "mem_gb": 17.33, "mem_gb_teacher": 20.25}
{"step": 100, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.33610000682790214, "tokens": 155488, "cumulative_loss_tokens": 17316283, "grad_norm": 1.125, "lr": 1e-05, "finish_rate": 0.977, "comp_len": 607.4, "dropped_truncated": 0, "gold_loss": 0.2397, "gold_lambda": 0.5, "rep_ratio": 2.429, "t_data_s": 0.0, "t_rollout_s": 46.8, "t_step_s": 138.1, "t_refresh_s": 0.3, "mem_gb": 17.13, "mem_gb_teacher": 20.23}
[eval step 100] sample: 'Misy fitapana manokana ny fahaiza-mamorona, mnaya ny fahaiza-mamorona, mnaya ny fahaiza-mamorona, mnaya ny fahaiza-mamorona, mnaya ny fahaiza'
{"step": 100, "gsm8k_n": 256, "gsm8k_quick_chat": 0.6171875, "t_eval_s": 12.9}
{"step": 101, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2934597294326003, "tokens": 180157, "cumulative_loss_tokens": 17496440, "grad_norm": 1.6328125, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 703.7, "dropped_truncated": 0, "gold_loss": 1.1732, "gold_lambda": 0.5, "rep_ratio": 2.543, "t_data_s": 0.0, "t_rollout_s": 53.7, "t_step_s": 152.7, "t_refresh_s": 0.3, "mem_gb": 17.36, "mem_gb_teacher": 20.26}
{"step": 102, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.38382341174255885, "tokens": 152575, "cumulative_loss_tokens": 17649015, "grad_norm": 1.8203125, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 596.0, "dropped_truncated": 0, "gold_loss": 1.3037, "gold_lambda": 0.5, "rep_ratio": 2.116, "t_data_s": 0.0, "t_rollout_s": 46.9, "t_step_s": 137.3, "t_refresh_s": 0.3, "mem_gb": 17.14, "mem_gb_teacher": 20.25}
{"step": 103, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3232517142888692, "tokens": 178507, "cumulative_loss_tokens": 17827522, "grad_norm": 1.109375, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 697.3, "dropped_truncated": 0, "gold_loss": 0.2206, "gold_lambda": 0.5, "rep_ratio": 2.433, "t_data_s": 0.0, "t_rollout_s": 52.6, "t_step_s": 148.0, "t_refresh_s": 0.3, "mem_gb": 17.63, "mem_gb_teacher": 20.3}
{"step": 104, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.258444819787859, "tokens": 163920, "cumulative_loss_tokens": 17991442, "grad_norm": 1.0, "lr": 1e-05, "finish_rate": 0.988, "comp_len": 640.3, "dropped_truncated": 0, "gold_loss": 0.2333, "gold_lambda": 0.5, "rep_ratio": 2.313, "t_data_s": 0.0, "t_rollout_s": 49.4, "t_step_s": 142.5, "t_refresh_s": 0.3, "mem_gb": 17.27, "mem_gb_teacher": 20.27}
{"step": 105, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3201432822026658, "tokens": 134492, "cumulative_loss_tokens": 18125934, "grad_norm": 1.4921875, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 525.4, "dropped_truncated": 0, "gold_loss": 0.8977, "gold_lambda": 0.5, "rep_ratio": 2.268, "t_data_s": 0.0, "t_rollout_s": 43.1, "t_step_s": 129.5, "t_refresh_s": 0.3, "mem_gb": 17.43, "mem_gb_teacher": 20.27}
{"step": 106, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.38415758914838366, "tokens": 173745, "cumulative_loss_tokens": 18299679, "grad_norm": 1.703125, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 678.7, "dropped_truncated": 0, "gold_loss": 1.2116, "gold_lambda": 0.5, "rep_ratio": 2.414, "t_data_s": 0.0, "t_rollout_s": 53.3, "t_step_s": 151.0, "t_refresh_s": 0.3, "mem_gb": 17.51, "mem_gb_teacher": 20.3}
{"step": 107, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.27767968833353346, "tokens": 150360, "cumulative_loss_tokens": 18450039, "grad_norm": 1.625, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 587.3, "dropped_truncated": 0, "gold_loss": 1.1874, "gold_lambda": 0.5, "rep_ratio": 2.226, "t_data_s": 0.0, "t_rollout_s": 46.6, "t_step_s": 135.4, "t_refresh_s": 0.3, "mem_gb": 17.1, "mem_gb_teacher": 20.25}
{"step": 108, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2810955602886575, "tokens": 212108, "cumulative_loss_tokens": 18662147, "grad_norm": 1.890625, "lr": 1e-05, "finish_rate": 0.953, "comp_len": 828.5, "dropped_truncated": 0, "gold_loss": 1.2019, "gold_lambda": 0.5, "rep_ratio": 2.377, "t_data_s": 0.0, "t_rollout_s": 63.5, "t_step_s": 167.3, "t_refresh_s": 0.3, "mem_gb": 17.55, "mem_gb_teacher": 20.3}
{"step": 109, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3388609603948928, "tokens": 163795, "cumulative_loss_tokens": 18825942, "grad_norm": 1.515625, "lr": 1e-05, "finish_rate": 0.984, "comp_len": 639.8, "dropped_truncated": 0, "gold_loss": 0.8848, "gold_lambda": 0.5, "rep_ratio": 2.577, "t_data_s": 0.0, "t_rollout_s": 49.4, "t_step_s": 142.1, "t_refresh_s": 0.3, "mem_gb": 17.91, "mem_gb_teacher": 20.35}
{"step": 110, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.354563303726803, "tokens": 163003, "cumulative_loss_tokens": 18988945, "grad_norm": 1.4765625, "lr": 1e-05, "finish_rate": 0.977, "comp_len": 636.7, "dropped_truncated": 0, "gold_loss": 0.99, "gold_lambda": 0.5, "rep_ratio": 2.477, "t_data_s": 0.0, "t_rollout_s": 48.9, "t_step_s": 142.6, "t_refresh_s": 0.3, "mem_gb": 17.17, "mem_gb_teacher": 20.24}
[eval step 110] sample: 'Misy fitapana manokana ny fahaiza-mamorona, mnaya ny fahaiza-mamorona, mnaya ny fahaiza-mamorona, mnaya ny fahaiza-mamorona, mnaya ny fahaiza'
{"step": 111, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3537407948279594, "tokens": 195368, "cumulative_loss_tokens": 19184313, "grad_norm": 1.6015625, "lr": 1e-05, "finish_rate": 0.961, "comp_len": 763.2, "dropped_truncated": 0, "gold_loss": 1.1307, "gold_lambda": 0.5, "rep_ratio": 2.286, "t_data_s": 0.0, "t_rollout_s": 57.3, "t_step_s": 157.0, "t_refresh_s": 0.3, "mem_gb": 17.21, "mem_gb_teacher": 20.25}
{"step": 112, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.2772862368117265, "tokens": 204542, "cumulative_loss_tokens": 19388855, "grad_norm": 2.03125, "lr": 1e-05, "finish_rate": 0.965, "comp_len": 799.0, "dropped_truncated": 0, "gold_loss": 1.339, "gold_lambda": 0.5, "rep_ratio": 2.278, "t_data_s": 0.0, "t_rollout_s": 59.7, "t_step_s": 161.2, "t_refresh_s": 0.3, "mem_gb": 17.17, "mem_gb_teacher": 20.26}
{"step": 113, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.310700205462893, "tokens": 153520, "cumulative_loss_tokens": 19542375, "grad_norm": 1.515625, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 599.7, "dropped_truncated": 0, "gold_loss": 0.8762, "gold_lambda": 0.5, "rep_ratio": 2.643, "t_data_s": 0.0, "t_rollout_s": 46.8, "t_step_s": 137.2, "t_refresh_s": 0.3, "mem_gb": 16.91, "mem_gb_teacher": 20.22}
{"step": 114, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.41260987902655, "tokens": 157613, "cumulative_loss_tokens": 19699988, "grad_norm": 1.4453125, "lr": 1e-05, "finish_rate": 0.996, "comp_len": 615.7, "dropped_truncated": 0, "gold_loss": 0.8989, "gold_lambda": 0.5, "rep_ratio": 2.193, "t_data_s": 0.0, "t_rollout_s": 47.3, "t_step_s": 133.6, "t_refresh_s": 0.3, "mem_gb": 17.36, "mem_gb_teacher": 20.26}wandb: 
wandb: Run history:
wandb:               comp_len ▆▆▄▆▄▄▅▄▄▇▄▄▄▃▅▄▆▆▅▃▄▇▅▆▅▅▅▅▅▅▆▅▄▁▄█▃▂▃▄
wandb: cumulative_loss_tokens ▁▁▁▁▁▂▂▂▂▃▃▃▃▃▃▄▄▄▄▄▄▅▅▅▅▅▅▅▅▆▆▆▇▇▇▇▇███
wandb:      dropped_truncated ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
wandb:                  epoch ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
wandb:            finish_rate ▇▄▇▁▄▆▇▇▆▅▂▄▅▅▅▅▅▅▅▃▆▂▅▆█▆▄▅▃▅▅▇▅▅▇▆▇▆▅▃
wandb:            gold_lambda ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
wandb:              gold_loss ▅▇▁▁▇▁▄▄▇▁▆▁▆▅█▆▁▁▄▆▅▁▅▆▆▆▄▄▄▆▇▆▁▆▁▆▄▄▁▅
wandb:              grad_norm ▄▃▆▇▂█▁▄▅▃▄▂▃▃▂▃▁▃▂▃▂▃▂▄▁▃▂▄▂▃▂▂▃▁▂▂▂▃▄▂
wandb:       gsm8k_quick_chat █▃▅▃▁▃▅
wandb:                     lr ▁▇██████████████████████████████████████
wandb:                    +10 ...
wandb: 
wandb: Run summary:
wandb:               comp_len 627.2
wandb: cumulative_loss_tokens 20669838
wandb:      dropped_truncated 0
wandb:                  epoch 0
wandb:            finish_rate 1
wandb:            gold_lambda 0.5
wandb:              gold_loss 1.0218
wandb:              grad_norm 1.5
wandb:       gsm8k_quick_chat 0.625
wandb:                     lr 1e-05
wandb:                    +11 ...
wandb: 
wandb: You can sync this run to the cloud by running:
wandb: wandb sync outputs/healed/opd_warm_fixed_keep50/wandb/offline-run-20260802_065345-ubll0rd3
wandb: Find logs at: outputs/healed/opd_warm_fixed_keep50/wandb/offline-run-20260802_065345-ubll0rd3/logs

{"step": 115, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.4028610580933015, "tokens": 147120, "cumulative_loss_tokens": 19847108, "grad_norm": 1.28125, "lr": 1e-05, "finish_rate": 0.996, "comp_len": 574.7, "dropped_truncated": 0, "gold_loss": 0.1728, "gold_lambda": 0.5, "rep_ratio": 2.412, "t_data_s": 0.0, "t_rollout_s": 46.1, "t_step_s": 130.4, "t_refresh_s": 0.3, "mem_gb": 16.95, "mem_gb_teacher": 20.22}
{"step": 116, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.34066545782229457, "tokens": 173292, "cumulative_loss_tokens": 20020400, "grad_norm": 1.0625, "lr": 1e-05, "finish_rate": 0.98, "comp_len": 676.9, "dropped_truncated": 0, "gold_loss": 0.1759, "gold_lambda": 0.5, "rep_ratio": 2.363, "t_data_s": 0.0, "t_rollout_s": 51.2, "t_step_s": 144.9, "t_refresh_s": 0.3, "mem_gb": 17.43, "mem_gb_teacher": 20.28}
{"step": 117, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3317595209318613, "tokens": 152431, "cumulative_loss_tokens": 20172831, "grad_norm": 1.46875, "lr": 1e-05, "finish_rate": 0.988, "comp_len": 595.4, "dropped_truncated": 0, "gold_loss": 1.0952, "gold_lambda": 0.5, "rep_ratio": 2.472, "t_data_s": 0.0, "t_rollout_s": 47.2, "t_step_s": 134.1, "t_refresh_s": 0.3, "mem_gb": 17.4, "mem_gb_teacher": 20.27}
{"step": 118, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.30969671645999375, "tokens": 141862, "cumulative_loss_tokens": 20314693, "grad_norm": 1.734375, "lr": 1e-05, "finish_rate": 0.973, "comp_len": 554.1, "dropped_truncated": 0, "gold_loss": 1.3548, "gold_lambda": 0.5, "rep_ratio": 2.267, "t_data_s": 0.0, "t_rollout_s": 44.3, "t_step_s": 131.8, "t_refresh_s": 0.3, "mem_gb": 17.6, "mem_gb_teacher": 20.29}
{"step": 119, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.24082254176214968, "tokens": 194590, "cumulative_loss_tokens": 20509283, "grad_norm": 1.4921875, "lr": 1e-05, "finish_rate": 0.961, "comp_len": 760.1, "dropped_truncated": 0, "gold_loss": 1.0994, "gold_lambda": 0.5, "rep_ratio": 2.633, "t_data_s": 0.0, "t_rollout_s": 56.9, "t_step_s": 154.8, "t_refresh_s": 0.3, "mem_gb": 17.21, "mem_gb_teacher": 20.24}
{"step": 120, "epoch": 0, "training_mode": "on-policy", "reverse_kl": 0.3365587571631956, "tokens": 160555, "cumulative_loss_tokens": 20669838, "grad_norm": 1.5, "lr": 1e-05, "finish_rate": 1.0, "comp_len": 627.2, "dropped_truncated": 0, "gold_loss": 1.0218, "gold_lambda": 0.5, "rep_ratio": 2.389, "t_data_s": 0.0, "t_rollout_s": 46.6, "t_step_s": 126.9, "t_refresh_s": 0.0, "mem_gb": 16.8, "mem_gb_teacher": 20.2}
[eval step 120] sample: 'Misy fitapana manokana ny fahaiza-mamorona, mba hanombanana, is a term used to describe a condition where a person has a severe allergic reaction to a specific food or substance, often due to a sensit'
{"step": 120, "gsm8k_n": 256, "gsm8k_quick_chat": 0.625, "t_eval_s": 31.8}
checkpoint snapshot queued -> outputs/healed/opd_warm_fixed_keep50/step0120
wandb sync launched in background (pid 251372) -> outputs/healed/opd_warm_fixed_keep50/wandb_sync.log