1 line
11 KiB
JSON
1 line
11 KiB
JSON
{"critic/advantages/max":0.875,"response_length_non_aborted/max":8192,"prompt_length/mean":296.15625,"val-aux/sciknoweval/acc/best@8/mean":0.7998666666666667,"val-aux/sciknoweval/reward/maj@2/mean":0.7489761904761905,"val-aux/sciknoweval/reward/worst@8/std":0.02011337982070672,"val-aux/sciknoweval/score/maj@2/mean":0.7489761904761905,"val-aux/sciknoweval/acc/worst@2/std":0.046990424566282364,"timing_s/agent_loop/slowest/prompt_length":198,"training/rollout_actor_probs_pearson_corr":0.9987240433692932,"timing_s/agent_loop/slowest/generate_sequences":48.49703206680715,"rollout_corr/k3_kl":0.0001086825423562062,"rollout_corr/rollout_is_seq_min":0.9954840540885925,"val-aux/sciknoweval/reward/maj@16/std":0.028181641738688423,"val-aux/sciknoweval/reward/maj@4/mean":0.7514238095238096,"self_distillation/correct_sample_fraction":0.7265625,"num_turns/min":2,"rollout_corr/rollout_is_max":1.601178526878357,"timing_s/agent_loop/slowest/tool_calls":0,"val-aux/sciknoweval/acc/worst@2/mean":0.7262095238095239,"val-aux/sciknoweval/reward/mean@16":0.7491071428571429,"prompt_length/clip_ratio":0,"rollout_corr/rollout_is_seq_mean":0.9999521374702454,"val-aux/sciknoweval/incorrect_format/worst@16/std":0.002274256999100845,"self_distillation/success_group_fraction":0.875,"val-core/sciknoweval/acc/best@16/std":0.008797318267434817,"val-aux/sciknoweval/reward/best@16/std":0.008797318267434817,"global_seqlen/min":11694,"val-aux/sciknoweval/reward/maj@16/mean":0.7554761904761905,"val-aux/sciknoweval/reward/best@8/mean":0.7998666666666667,"timing_per_token_ms/gen":0.9042675971594261,"val-aux/sciknoweval/acc/best@8/std":0.017334238666592075,"val-aux/sciknoweval/incorrect_format/worst@16/mean":0.9969142857142858,"rollout_corr/rollout_is_seq_std":0.0012089080410078168,"val-aux/sciknoweval/score/worst@16/std":0.010972447210233164,"critic/advantages/mean":-0.09708639234304428,"rollout_corr/rollout_is_std":0.014016308821737766,"val-aux/sciknoweval/reward/worst@2/std":0.046990424566282364,"val-aux/sciknoweval/acc/worst@8/mean":0.6937857142857142,"timing_s/gen":54.36728074401617,"global_seqlen/mean":16992.375,"val-core/sciknoweval/acc/best@16/mean":0.8060142857142858,"perf/throughput":229.551173302718,"timing_s/agent_loop/generate_sequences/max":48.49703206680715,"timing_s/old_log_prob":2.4735527355223894,"val-aux/sciknoweval/incorrect_format/maj@4/std":0.0005594782149609339,"val-aux/sciknoweval/score/best@16/std":0.008797318267434817,"critic/score/mean":0.7265625,"self_distillation/success_sample_fraction":0.8671875,"val-aux/sciknoweval/incorrect_format/best@4/mean":1,"rollout_corr/rollout_is_seq_fraction_high":0,"val-aux/sciknoweval/incorrect_format/best@8/mean":1,"rollout_corr/rollout_is_seq_max_deviation":0.005106925964355469,"val-aux/num_turns/mean":2,"response_length_non_aborted/clip_ratio":0.0078125,"val-aux/sciknoweval/score/worst@4/mean":0.706704761904762,"val-aux/sciknoweval/acc/best@2/mean":0.771904761904762,"rollout_corr/rollout_is_seq_fraction_low":0,"val-aux/sciknoweval/score/maj@2/std":0.05319068044795249,"response_length/min":89,"actor/lr":1e-05,"critic/rewards/mean":0.7265625,"val-aux/sciknoweval/score/worst@16/mean":0.6871380952380951,"val-aux/sciknoweval/incorrect_format/maj@16/std":0,"perf/mfu/actor":0.03008261803486565,"rollout_corr/rollout_is_ratio_fraction_high":0,"perf/max_memory_reserved_gb":142.96875,"val-aux/sciknoweval/incorrect_format/best@2/mean":0.9999857142857144,"self_distillation/dynamic_weighting_enabled":0,"rollout_corr/chi2_seq":0.08643001760356128,"timing_s/agent_loop/tool_calls/max":0,"timing_s/testing":99.34055501781404,"val-aux/sciknoweval/score/best@4/mean":0.7891285714285714,"val-aux/sciknoweval/score/worst@2/mean":0.7262095238095239,"val-aux/sciknoweval/incorrect_format/maj@8/mean":1,"val-aux/sciknoweval/acc/best@2/std":0.0438263744703264,"perf/time_per_step":74.02434392087162,"actor/pg_loss":0.012388480715372907,"val-aux/sciknoweval/reward/maj@8/mean":0.7533142857142858,"timing_s/adv":0.4935148283839226,"response_length/mean":234.85546875,"_timestamp":1.7829831001197822e+09,"perf/mfu/actor_infer":0,"val-aux/sciknoweval/incorrect_format/worst@8/std":0.002330883067212054,"val-aux/sciknoweval/acc/worst@8/std":0.02011337982070672,"val-aux/sciknoweval/incorrect_format/maj@8/std":0,"rollout_corr/rollout_is_min":0.620167076587677,"val-aux/sciknoweval/score/best@2/mean":0.771904761904762,"response_length/max":8192,"actor/grad_norm":0.1816544234752655,"val-aux/sciknoweval/score/best@4/std":0.029400016901477092,"critic/returns/min":-0.75,"timing_s/stop_profile":0.00012635812163352966,"rollout_corr/rollout_log_ppl":0.035416119580048644,"val-aux/sciknoweval/acc/worst@4/mean":0.706704761904762,"val-aux/sciknoweval/acc/best@4/mean":0.7891285714285714,"training/rollout_probs_diff_valid":1,"val-aux/sciknoweval/incorrect_format/best@8/std":0,"val-aux/sciknoweval/score/maj@8/mean":0.7533142857142858,"val-core/sciknoweval/acc/maj@16/std":0.028181641738688423,"response/aborted_ratio":0,"val-aux/sciknoweval/score/std@16":0.053113895809534975,"val-aux/sciknoweval/incorrect_format/worst@8/mean":0.9981047619047619,"val-aux/sciknoweval/incorrect_format/worst@2/mean":0.9994238095238095,"val-aux/sciknoweval/acc/std@16":0.053113895809534975,"val-aux/sciknoweval/reward/worst@16/mean":0.6871380952380951,"val-aux/sciknoweval/incorrect_format/maj@2/mean":0.999652380952381,"timing_s/step":74.02434392087162,"critic/score/min":0,"val-aux/sciknoweval/incorrect_format/maj@4/mean":0.9999333333333333,"critic/rewards/min":0,"val-aux/sciknoweval/reward/best@4/std":0.029400016901477092,"timing_s/agent_loop/tool_calls/min":0,"val-aux/sciknoweval/reward/best@8/std":0.017334238666592075,"rollout_corr/rollout_is_ratio_fraction_low":0,"rollout_corr/training_ppl":1.0374733386561275,"timing_per_token_ms/adv":0.003630413850211658,"val-aux/sciknoweval/incorrect_format/worst@2/std":0.00155298702746771,"self_distillation/feedback_available_fraction":0,"rollout_corr/rollout_is_mean":1.0000271797180176,"timing_s/reward":0.05386519245803356,"val-aux/sciknoweval/reward/worst@8/mean":0.6937857142857142,"val-aux/sciknoweval/incorrect_format/mean@16":0.999702380952381,"timing_s/start_profile":0.00018905289471149445,"rollout_corr/rollout_is_seq_max":1.0051069259643555,"perf/total_num_tokens":135939,"rollout_corr/log_ppl_abs_diff":0.0008119762612466275,"perf/cpu_memory_used_gb":148.59503936767578,"val-aux/sciknoweval/score/maj@4/std":0.04465209687151754,"val-aux/sciknoweval/reward/best@2/mean":0.771904761904762,"timing_s/agent_loop/tool_calls/mean":0,"val-aux/sciknoweval/acc/maj@8/mean":0.7533142857142858,"val-aux/sciknoweval/incorrect_format/best@2/std":0.0002604287412166728,"val-aux/sciknoweval/acc/best@4/std":0.029400016901477092,"val-aux/sciknoweval/score/maj@16/std":0.028181641738688423,"val-aux/sciknoweval/score/worst@4/std":0.03441618216443395,"critic/score/max":1,"training/epoch":1,"timing_s/agent_loop/generate_sequences/min":1.0307863671332598,"perf/max_memory_allocated_gb":138.14347791671753,"global_seqlen/balanced_max":22508,"response_length/clip_ratio":0.0078125,"val-aux/sciknoweval/score/mean@16":0.7491071428571429,"val-aux/sciknoweval/reward/worst@4/mean":0.706704761904762,"_step":100,"val-core/sciknoweval/acc/mean@16":0.7491071428571429,"val-aux/sciknoweval/reward/worst@2/mean":0.7262095238095239,"val-aux/sciknoweval/score/best@16/mean":0.8060142857142858,"val-aux/sciknoweval/reward/best@2/std":0.0438263744703264,"critic/advantages/min":-0.75,"val-aux/sciknoweval/incorrect_format/best@4/std":0,"val-aux/sciknoweval/acc/maj@4/std":0.04465209687151754,"global_seqlen/minmax_diff":20485,"self_distillation/feedback_used_fraction":0,"val-aux/num_turns/min":2,"val-aux/sciknoweval/acc/worst@16/std":0.010972447210233164,"prompt_length/max":625,"rollout_corr/chi2_token":0.00012734439224004745,"num_turns/max":2,"val-aux/sciknoweval/incorrect_format/maj@16/mean":1,"actor/entropy":0.04114176332950592,"val-aux/sciknoweval/score/best@8/std":0.017334238666592075,"rollout_corr/rollout_ppl":1.0373146124184132,"_wandb":{"runtime":4229},"_runtime":4229.750378638,"rollout_corr/training_log_ppl":0.03557254603532556,"critic/rewards/max":1,"prompt_length/min":165,"val-aux/sciknoweval/acc/worst@16/mean":0.6871380952380951,"training/rollout_probs_diff_std":0.005446349270641804,"response_length_non_aborted/mean":234.85546875,"val-aux/sciknoweval/incorrect_format/worst@4/std":0.0019947266079247076,"val-aux/sciknoweval/score/worst@8/mean":0.6937857142857142,"timing_s/agent_loop/generate_sequences/mean":2.187226355308667,"self_distillation/reprompt_sample_fraction":0.8671875,"global_seqlen/balanced_min":15112,"timing_s/save_checkpoint":6.613632952794433,"val-aux/sciknoweval/score/maj@4/mean":0.7514238095238096,"rollout_corr/kl":0.00015642605447674374,"val-aux/sciknoweval/reward/worst@4/std":0.03441618216443395,"val-aux/num_turns/max":2,"val-aux/sciknoweval/reward/maj@2/std":0.05319068044795249,"rollout_corr/log_ppl_diff_max":0.004914548248052597,"rollout_corr/rollout_is_eff_sample_size":0.9998034625161065,"timing_s/update_actor":16.550736574456096,"val-aux/sciknoweval/score/worst@8/std":0.02011337982070672,"rollout_corr/log_ppl_diff_min":-0.0037901997566223145,"val-aux/sciknoweval/acc/worst@4/std":0.03441618216443395,"critic/returns/max":0.875,"val-aux/sciknoweval/acc/maj@8/std":0.03611226257903519,"training/rollout_probs_diff_mean":0.000801601039711386,"response_length_non_aborted/min":89,"val-aux/sciknoweval/reward/worst@16/std":0.010972447210233164,"val-aux/sciknoweval/reward/best@16/mean":0.8060142857142858,"rollout_corr/log_ppl_diff":0.00015642645527691457,"val-aux/sciknoweval/score/maj@8/std":0.03611226257903519,"training/rollout_probs_diff_max":0.23105812072753906,"num_turns/mean":2,"val-aux/sciknoweval/score/best@8/mean":0.7998666666666667,"val-aux/sciknoweval/incorrect_format/std@16":0.0011526736149426837,"timing_per_token_ms/update_actor":0.12175120145400581,"val-aux/sciknoweval/incorrect_format/best@16/std":0,"val-aux/sciknoweval/score/worst@2/std":0.046990424566282364,"val-aux/sciknoweval/acc/maj@2/std":0.05319068044795249,"critic/returns/mean":-0.09708639234304428,"val-aux/sciknoweval/score/maj@16/mean":0.7554761904761905,"val-aux/sciknoweval/incorrect_format/best@16/mean":1,"val-aux/sciknoweval/reward/std@16":0.053113895809534975,"training/global_step":100,"val-core/sciknoweval/acc/maj@16/mean":0.7554761904761905,"val-aux/sciknoweval/reward/maj@8/std":0.03611226257903519,"actor/kl_loss":0,"rollout_corr/ppl_ratio":1.0001571618486196,"val-aux/sciknoweval/reward/maj@4/std":0.04465209687151754,"val-aux/sciknoweval/incorrect_format/maj@2/std":0.001238745250613849,"val-aux/sciknoweval/incorrect_format/worst@4/mean":0.9989190476190476,"val-aux/sciknoweval/reward/best@4/mean":0.7891285714285714,"timing_s/agent_loop/slowest/response_length":8192,"val-aux/sciknoweval/acc/maj@4/mean":0.7514238095238096,"val-aux/sciknoweval/acc/maj@2/mean":0.7489761904761905,"val-aux/sciknoweval/score/best@2/std":0.0438263744703264,"global_seqlen/max":32179,"self_distillation/empty_target_batch":0.1328125} |