| 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_bias0.2,0,_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_bias0.2_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_bias0.2/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_bias0.2_global_step_0_actor_huggingface/results_2025-08-18T16-21-32.723446.json,0.17744916820702403,0.016440744379699786,0.290167865707434,0.00048617417820989814,0.19408502772643252,0.017019380550749422,0.3105515587529976,0.0004876415866282015,0.2727272727272727,0.03173071239071724 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.2,10,_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_bias0.2_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_bias0.2/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_bias0.2_global_step_10_actor_huggingface/results_2025-08-18T16-10-44.289594.json,0.19963031423290203,0.017201316264889924,0.328537170263789,0.000511841898537647,0.22365988909426987,0.01793177105465837,0.35611510791366907,0.0005210750970247321,0.2676767676767677,0.031544498882702866 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.2,20,_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_bias0.2_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_bias0.2/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_bias0.2_global_step_20_actor_huggingface/results_2025-08-18T15-50-59.925267.json,0.19963031423290203,0.01720131626488992,0.34292565947242204,0.0004814028637664121,0.2365988909426987,0.018288827582625643,0.3824940047961631,0.00048680809224631706,0.22727272727272727,0.029857515673386414 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.2,30,_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_bias0.2_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_bias0.2/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_bias0.2_global_step_30_actor_huggingface/results_2025-08-18T15-54-45.139937.json,0.22920517560073936,0.018087757424955345,0.35251798561151076,0.0005083360098454102,0.2680221811460259,0.01906063869163029,0.39928057553956836,0.0005263980103926709,0.2676767676767677,0.03154449888270286 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.2,40,_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_bias0.2_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_bias0.2/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_bias0.2_global_step_40_actor_huggingface/results_2025-08-18T15-47-59.222617.json,0.22181146025878004,0.017878765407944388,0.3513189448441247,0.0005117989534031979,0.26247689463955637,0.018933742876044622,0.39568345323741005,0.0005288825845276972,0.29292929292929293,0.03242497958178816 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.2,50,_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_bias0.2_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_bias0.2/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_bias0.2_global_step_50_actor_huggingface/results_2025-08-18T15-45-31.395283.json,0.25508317929759705,0.01875849195041412,0.3752997601918465,0.0005371033505332842,0.2957486136783734,0.019639405361071123,0.4172661870503597,0.0005372517765562854,0.3181818181818182,0.03318477333845331 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.2,60,_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_bias0.2_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_bias0.2/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_bias0.2_global_step_60_actor_huggingface/results_2025-08-18T15-39-35.865731.json,0.24399260628465805,0.018482234430967866,0.3597122302158273,0.0004956301071835906,0.2975970425138632,0.01967481200441347,0.41007194244604317,0.0005142301550513566,0.2828282828282828,0.032087795587867514 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.2,70,_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_bias0.2_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_bias0.2/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_bias0.2_global_step_70_actor_huggingface/results_2025-08-18T15-45-24.707769.json,0.2513863216266174,0.01866821615224041,0.3752997601918465,0.0005111085586749603,0.3049907578558225,0.01981261268122439,0.4316546762589928,0.0005363750737440186,0.26262626262626265,0.031353050095330855 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.2,80,_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_bias0.2_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_bias0.2/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_bias0.2_global_step_80_actor_huggingface/results_2025-08-18T16-44-03.891235.json,0.24029574861367836,0.01838647358148708,0.37170263788968827,0.0005115562808717749,0.3031423290203327,0.019778732375985472,0.434052757793765,0.0005143507029247541,0.30303030303030304,0.03274287914026868 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.2,90,_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_bias0.2_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_bias0.2/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_bias0.2_global_step_90_actor_huggingface/results_2025-08-18T16-39-21.271295.json,0.2587800369685767,0.018846992560712535,0.3920863309352518,0.0005241019340023833,0.32162661737523107,0.02010080781364547,0.460431654676259,0.0005245124886099761,0.30808080808080807,0.032894773300986155 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.2,100,_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_bias0.2_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_bias0.2/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_bias0.2_global_step_100_actor_huggingface/results_2025-08-18T16-41-52.767998.json,0.26247689463955637,0.018933742876044622,0.4040767386091127,0.0004982089085448825,0.32162661737523107,0.02010080781364547,0.46402877697841727,0.000497799343927761,0.24242424242424243,0.030532892233932036 | |