| 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_bias0.1,0,_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_bias0.1_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_bias0.1/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_bias0.1_global_step_0_actor_huggingface/results_2025-08-15T13-35-49.816341.json,0.18853974121996303,0.016832096060176906,0.3009592326139089,0.0004899455552453011,0.2033271719038817,0.01731971864183472,0.3225419664268585,0.0004964143868406819,0.23232323232323232,0.030088629490217487 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.1,10,_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_bias0.1_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_bias0.1/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_bias0.1_global_step_10_actor_huggingface/results_2025-08-15T13-31-29.716693.json,0.21072088724584104,0.017549801883664243,0.3381294964028777,0.0005089530151644017,0.23105360443622922,0.01813875717052341,0.35851318944844124,0.0005147978170810502,0.22727272727272727,0.02985751567338641 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.1,20,_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_bias0.1_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_bias0.1/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_bias0.1_global_step_20_actor_huggingface/results_2025-08-15T13-21-09.066792.json,0.20517560073937152,0.017378071196759656,0.3393285371702638,0.00046903577339524614,0.2476894639556377,0.018576139285185186,0.3800959232613909,0.00048843229295727,0.2727272727272727,0.03173071239071724 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.1,30,_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_bias0.1_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_bias0.1/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_bias0.1_global_step_30_actor_huggingface/results_2025-08-15T13-17-38.270755.json,0.25693160813308685,0.01880296257563685,0.37050359712230213,0.0005133237544627958,0.28650646950092423,0.019456528583211702,0.40047961630695444,0.0005263192969937805,0.3181818181818182,0.03318477333845331 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.1,40,_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_bias0.1_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_bias0.1/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_bias0.1_global_step_40_actor_huggingface/results_2025-08-15T14-12-48.635371.json,0.24214417744916822,0.018434587800223085,0.3645083932853717,0.0004930768428297287,0.27171903881700554,0.019143116099594026,0.39928057553956836,0.0005111941678858706,0.20707070707070707,0.028869778460267045 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.1,50,_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_bias0.1_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_bias0.1/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_bias0.1_global_step_50_actor_huggingface/results_2025-08-15T14-05-24.807346.json,0.24214417744916822,0.018434587800223085,0.36330935251798563,0.0005408070774318331,0.27171903881700554,0.019143116099594026,0.40047961630695444,0.0005509574174339303,0.2727272727272727,0.03173071239071724 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.1,60,_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_bias0.1_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_bias0.1/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_bias0.1_global_step_60_actor_huggingface/results_2025-08-15T14-09-25.201221.json,0.23844731977818853,0.01833788809424395,0.35611510791366907,0.0005017596379285598,0.2902033271719039,0.019530856691222523,0.41007194244604317,0.0005197435333584474,0.2676767676767677,0.03154449888270286 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.1,70,_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_bias0.1_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_bias0.1/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_bias0.1_global_step_70_actor_huggingface/results_2025-08-15T14-07-38.503008.json,0.2680221811460259,0.01906063869163028,0.3980815347721823,0.000511821154378924,0.30129390018484287,0.019744473483514297,0.4352517985611511,0.0005169147730323216,0.2828282828282828,0.032087795587867514 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.1,80,_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_bias0.1_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_bias0.1/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_bias0.1_global_step_80_actor_huggingface/results_2025-08-15T14-34-26.588600.json,0.2365988909426987,0.01828882758262565,0.37170263788968827,0.000528537398534992,0.28835489833641403,0.0194938903506547,0.4172661870503597,0.0005291456563994543,0.2727272727272727,0.03173071239071724 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.1,90,_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_bias0.1_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_bias0.1/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_bias0.1_global_step_90_actor_huggingface/results_2025-08-15T14-39-05.603495.json,0.2365988909426987,0.01828882758262565,0.3824940047961631,0.0004959498026712708,0.27911275415896486,0.019303080958497216,0.4244604316546763,0.0005096410732030472,0.2727272727272727,0.03173071239071724 | |
| verl_role_sft_grpo_deepmath_train_sample_6144_context_4k_Qwen3-1.7B-base_max_response4096_batch1024_rollout8_vllm_True_bias0.1,100,_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_bias0.1_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_bias0.1/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_bias0.1_global_step_100_actor_huggingface/results_2025-08-15T14-37-42.123261.json,0.266173752310536,0.0190187668472906,0.3980815347721823,0.0005310616858035166,0.3123844731977819,0.019944386293758908,0.4556354916067146,0.0005307551589885664,0.2777777777777778,0.031911782267135445 | |