| exp_name,global_step,model_name,json_relpath,prompt_level_strict_acc,prompt_level_strict_acc_stderr,inst_level_strict_acc,inst_level_strict_acc_stderr,prompt_level_loose_acc,prompt_level_loose_acc_stderr,inst_level_loose_acc,inst_level_loose_acc_stderr,gpqa_pass@1:1_samples,gpqa_pass@1:1_samples_stderr | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3,0,_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3_global_step_0_actor_huggingface,verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3/eval_results_ood/global_step_0/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3_global_step_0_actor_huggingface/results_2025-08-19T03-03-31.558565.json,0.17744916820702403,0.016440744379699782,0.30935251798561153,0.00047499830510106683,0.19408502772643252,0.017019380550749415,0.33093525179856115,0.0004844989872326643,0.2777777777777778,0.03191178226713546 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3,10,_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3_global_step_10_actor_huggingface,verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3/eval_results_ood/global_step_10/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3_global_step_10_actor_huggingface/results_2025-08-19T02-59-44.015414.json,0.19963031423290203,0.01720131626488992,0.33093525179856115,0.0005111477801350833,0.21811460258780038,0.01777121177665385,0.3513189448441247,0.0005226896624887547,0.25757575757575757,0.031156269519646836 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3,20,_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3_global_step_20_actor_huggingface,verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3/eval_results_ood/global_step_20/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3_global_step_20_actor_huggingface/results_2025-08-19T02-45-28.477295.json,0.21811460258780038,0.017771211776653847,0.3489208633093525,0.000507384709085127,0.2587800369685767,0.018846992560712535,0.39568345323741005,0.000510820842534251,0.24242424242424243,0.030532892233932046 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3,30,_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3_global_step_30_actor_huggingface,verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3/eval_results_ood/global_step_30/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3_global_step_30_actor_huggingface/results_2025-08-19T02-31-43.587146.json,0.24214417744916822,0.018434587800223085,0.3513189448441247,0.0005128574097267274,0.27726432532347506,0.019263706963479364,0.3908872901678657,0.0005208737343045235,0.2878787878787879,0.03225883512300992 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3,40,_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3_global_step_40_actor_huggingface,verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3/eval_results_ood/global_step_40/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3_global_step_40_actor_huggingface/results_2025-08-19T02-32-20.502116.json,0.2809611829944547,0.019342047683777005,0.40167865707434053,0.0005330003632167457,0.3364140480591497,0.020332406004701285,0.45443645083932854,0.0005524936482910329,0.2878787878787879,0.03225883512300993 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3,50,_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3_global_step_50_actor_huggingface,verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3/eval_results_ood/global_step_50/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3_global_step_50_actor_huggingface/results_2025-08-19T02-27-37.276111.json,0.2476894639556377,0.01857613928518519,0.38489208633093525,0.00047565725932562395,0.2846580406654344,0.019418769106486006,0.42685851318944845,0.0004762003483399427,0.3181818181818182,0.033184773338453315 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3,60,_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3_global_step_60_actor_huggingface,verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3/eval_results_ood/global_step_60/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3_global_step_60_actor_huggingface/results_2025-08-19T02-24-48.573923.json,0.25508317929759705,0.018758491950414142,0.3872901678657074,0.0004929038935043971,0.3197781885397412,0.020070251556580192,0.44244604316546765,0.0005299981399131118,0.2777777777777778,0.031911782267135466 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3,70,_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3_global_step_70_actor_huggingface,verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3/eval_results_ood/global_step_70/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3_global_step_70_actor_huggingface/results_2025-08-19T02-23-41.225302.json,0.25693160813308685,0.018802962575636843,0.40167865707434053,0.0005063029486749525,0.2957486136783734,0.019639405361071123,0.44844124700239807,0.0005175166114037128,0.2828282828282828,0.03208779558786753 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3,80,_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3_global_step_80_actor_huggingface,verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3/eval_results_ood/global_step_80/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3_global_step_80_actor_huggingface/results_2025-08-19T03-22-27.759640.json,0.2698706099815157,0.01910208752649431,0.4064748201438849,0.0005172358772147335,0.33086876155268025,0.020248210903192006,0.46402877697841727,0.0005306531900131056,0.25757575757575757,0.031156269519646836 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3,90,_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3_global_step_90_actor_huggingface,verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3/eval_results_ood/global_step_90/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3_global_step_90_actor_huggingface/results_2025-08-19T03-23-41.379521.json,0.2735674676524954,0.019183727107392815,0.4088729016786571,0.0005079553194066644,0.33826247689463956,0.020359772138166043,0.473621103117506,0.0005171265656828542,0.26262626262626265,0.03135305009533085 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3,100,_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3_global_step_100_actor_huggingface,verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3/eval_results_ood/global_step_100/results/_home_work_compass_innovation_minzijun_checkpoints_verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.3_global_step_100_actor_huggingface/results_2025-08-19T03-23-25.775217.json,0.2846580406654344,0.019418769106486003,0.41127098321342925,0.0005019275943079363,0.36044362292051757,0.020661469669879428,0.4844124700239808,0.0005108788936364846,0.2474747474747475,0.030746300742124505 | |