1 line
11 KiB
JSON
1 line
11 KiB
JSON
{"_timestamp":1.7830428622245336e+09,"response_length/clip_ratio":0,"training/rollout_actor_probs_pearson_corr":0.9996163845062256,"training/epoch":1,"timing_s/adv":0.11797017231583595,"critic/score/min":0,"val-aux/sciknoweval/reward/maj@8/std":0.13381993498072356,"rollout_corr/rollout_is_std":0.03385905548930168,"prompt_length/clip_ratio":0,"critic/score/max":1,"val-aux/sciknoweval/reward/std@16":0.21497342883717385,"response_length/mean":832.765625,"_step":100,"val-aux/sciknoweval/reward/maj@4/std":0.17575431812173817,"rollout_corr/log_ppl_abs_diff":0.0010970584698952734,"val-core/sciknoweval/acc/best@16/mean":0.9295428571428572,"global_seqlen/max":40888,"training/global_step":100,"rollout_corr/chi2_token":0.0011783752124756575,"timing_s/agent_loop/generate_sequences/mean":9.964753172069322,"val-aux/sciknoweval/reward/best@2/mean":0.7771380952380953,"global_seqlen/min":32181,"rollout_corr/rollout_is_ratio_fraction_low":4.690695732278982e-06,"val-aux/sciknoweval/acc/worst@2/std":0.18719434569533389,"val-aux/sciknoweval/incorrect_format/maj@2/mean":0.9648952380952381,"val-aux/sciknoweval/reward/maj@8/mean":0.7290142857142858,"rollout_corr/log_ppl_diff":0.0006403211154974997,"critic/returns/mean":0.0025611198507249355,"val-aux/sciknoweval/incorrect_format/maj@8/std":0.03401010083743055,"val-aux/sciknoweval/incorrect_format/best@8/mean":1,"val-aux/sciknoweval/reward/best@8/mean":0.8984857142857143,"val-aux/sciknoweval/reward/maj@2/mean":0.6898333333333334,"val-aux/sciknoweval/reward/best@2/std":0.17605374457438094,"val-aux/sciknoweval/incorrect_format/maj@2/std":0.08777577522757704,"val-aux/sciknoweval/incorrect_format/worst@8/std":0.12414222258147589,"val-aux/sciknoweval/acc/maj@2/std":0.21523251642703473,"val-aux/sciknoweval/score/worst@16/std":0.07212858795272402,"timing_per_token_ms/adv":0.00040819563852346666,"val-aux/num_turns/max":2,"val-aux/sciknoweval/incorrect_format/std@16":0.08812713073175192,"val-aux/sciknoweval/score/maj@16/mean":0.7341666666666667,"prompt_length/min":165,"val-aux/sciknoweval/acc/worst@8/std":0.10546043434417916,"val-aux/sciknoweval/score/best@16/mean":0.9295428571428572,"val-aux/sciknoweval/acc/std@16":0.21497342883717385,"response_length/max":1661,"timing_s/agent_loop/generate_sequences/min":3.3696037754416466,"num_turns/min":2,"actor/pg_clipfrac_lower":0,"response_length_non_aborted/mean":832.765625,"val-aux/sciknoweval/reward/worst@8/std":0.10546043434417916,"num_turns/mean":2,"val-aux/sciknoweval/incorrect_format/best@16/std":0,"rollout_corr/training_log_ppl":0.7637257019523531,"rollout_corr/kl":0.0006403183531915602,"val-aux/sciknoweval/reward/best@16/mean":0.9295428571428572,"rollout_corr/rollout_is_seq_min":0.9932000041007996,"global_seqlen/balanced_max":36131,"prompt_length/max":625,"actor/pg_loss":1.9038328900933266e-05,"val-aux/sciknoweval/reward/worst@8/mean":0.47553809523809526,"timing_s/start_profile":0.00019051693379878998,"val-aux/sciknoweval/score/best@16/std":0.046578487700934236,"val-aux/sciknoweval/incorrect_format/maj@16/mean":0.9985285714285715,"val-aux/sciknoweval/acc/best@8/mean":0.8984857142857143,"val-aux/sciknoweval/incorrect_format/worst@2/std":0.11103403939640785,"response_length_non_aborted/min":260,"rollout_corr/chi2_seq":1.0062491311691701,"timing_s/agent_loop/slowest/response_length":1661,"actor/entropy":0.7533097267150879,"val-aux/sciknoweval/acc/best@4/std":0.128790863116223,"val-aux/sciknoweval/acc/best@4/mean":0.8483380952380953,"val-aux/sciknoweval/acc/best@2/std":0.17605374457438094,"rollout_corr/rollout_is_eff_sample_size":0.9988548770722385,"rollout_corr/rollout_is_seq_max":1.0046179294586182,"rollout_corr/rollout_is_seq_std":0.0013772835955023766,"val-aux/sciknoweval/score/worst@8/mean":0.47553809523809526,"critic/advantages/mean":0.0025611198507249355,"response_length_non_aborted/max":1661,"timing_s/reward":0.08236124739050865,"perf/total_num_tokens":289004,"training/rollout_probs_diff_max":0.2046983242034912,"timing_s/gen":22.220765927806497,"prompt_length/mean":296.15625,"critic/returns/max":0.875,"val-aux/sciknoweval/acc/worst@2/mean":0.6025857142857143,"val-aux/sciknoweval/acc/worst@4/mean":0.5304333333333333,"timing_s/agent_loop/tool_calls/min":0,"val-aux/sciknoweval/incorrect_format/best@2/std":0.03432708478331659,"val-aux/sciknoweval/acc/maj@8/std":0.13381993498072356,"val-aux/sciknoweval/reward/mean@16":0.6901785714285714,"actor/lr":1e-06,"rollout_corr/log_ppl_diff_min":-0.0030992627143859863,"val-aux/sciknoweval/acc/best@8/std":0.08310585797508012,"num_turns/max":2,"critic/rewards/mean":0.67578125,"val-aux/sciknoweval/score/maj@8/mean":0.7290142857142858,"critic/rewards/min":0,"timing_s/testing":247.1615258976817,"critic/advantages/max":0.875,"val-aux/sciknoweval/score/best@4/mean":0.8483380952380953,"val-aux/sciknoweval/score/maj@16/std":0.10114866635456188,"global_seqlen/minmax_diff":8707,"val-aux/sciknoweval/acc/worst@4/std":0.14423738321883778,"val-aux/sciknoweval/reward/worst@2/mean":0.6025857142857143,"rollout_corr/k3_kl":0.0006129170444637566,"timing_s/stop_profile":6.954558193683624e-05,"rollout_corr/rollout_is_seq_mean":0.9999972581863403,"val-aux/sciknoweval/incorrect_format/best@4/std":0.004692755954297982,"val-core/sciknoweval/acc/mean@16":0.6901785714285714,"timing_s/agent_loop/slowest/tool_calls":0,"timing_s/agent_loop/tool_calls/max":0,"val-aux/sciknoweval/acc/worst@8/mean":0.47553809523809526,"perf/max_memory_reserved_gb":144.970703125,"val-aux/sciknoweval/score/worst@8/std":0.10546043434417916,"training/rollout_probs_diff_valid":1,"val-aux/sciknoweval/incorrect_format/worst@2/mean":0.9356666666666668,"actor/grad_norm":0.20593710616230965,"timing_s/old_log_prob":3.2945325896143913,"_wandb":{"runtime":6166},"val-aux/sciknoweval/reward/worst@4/std":0.14423738321883778,"val-aux/sciknoweval/acc/maj@4/std":0.17575431812173817,"response/aborted_ratio":0,"val-aux/sciknoweval/score/std@16":0.21497342883717385,"val-aux/sciknoweval/acc/maj@4/mean":0.7146857142857143,"rollout_corr/rollout_is_mean":0.999994695186615,"val-aux/sciknoweval/acc/maj@8/mean":0.7290142857142858,"val-aux/sciknoweval/incorrect_format/maj@16/std":0.011901839678247213,"val-aux/sciknoweval/score/best@2/std":0.17605374457438094,"val-aux/sciknoweval/score/worst@2/std":0.18719434569533389,"training/rollout_probs_diff_std":0.00816031452268362,"timing_s/agent_loop/slowest/prompt_length":255,"timing_per_token_ms/update_actor":0.047571522798793175,"val-aux/sciknoweval/reward/worst@4/mean":0.5304333333333333,"critic/rewards/max":1,"val-aux/sciknoweval/reward/worst@2/std":0.18719434569533389,"val-aux/sciknoweval/incorrect_format/best@2/mean":0.9939428571428572,"perf/time_per_step":39.5498822350055,"rollout_corr/rollout_is_seq_fraction_low":0,"val-aux/sciknoweval/score/worst@4/std":0.14423738321883778,"global_seqlen/balanced_min":36112,"val-aux/sciknoweval/acc/worst@16/std":0.07212858795272402,"val-aux/sciknoweval/incorrect_format/worst@4/std":0.12713323596686327,"val-aux/sciknoweval/reward/best@4/std":0.128790863116223,"val-aux/sciknoweval/score/maj@4/mean":0.7146857142857143,"training/rollout_probs_diff_mean":0.0041552940383553505,"val-aux/sciknoweval/score/best@2/mean":0.7771380952380953,"rollout_corr/rollout_is_max":1.4722437858581543,"val-aux/sciknoweval/incorrect_format/mean@16":0.9645833333333333,"val-aux/sciknoweval/score/best@8/mean":0.8984857142857143,"val-aux/sciknoweval/score/mean@16":0.6901785714285714,"val-aux/sciknoweval/reward/worst@16/std":0.07212858795272402,"val-aux/sciknoweval/incorrect_format/worst@16/mean":0.7765571428571428,"val-aux/sciknoweval/reward/maj@16/std":0.10114866635456188,"val-aux/sciknoweval/incorrect_format/best@8/std":0,"critic/returns/min":-0.875,"rollout_corr/rollout_is_seq_fraction_high":0,"val-aux/sciknoweval/score/maj@2/std":0.21523251642703473,"timing_s/update_actor":13.748360374942422,"val-aux/sciknoweval/reward/worst@16/mean":0.4326952380952381,"val-aux/sciknoweval/incorrect_format/best@16/mean":1,"val-aux/sciknoweval/incorrect_format/maj@8/mean":0.9934761904761904,"val-aux/sciknoweval/incorrect_format/worst@4/mean":0.890790476190476,"rollout_corr/log_ppl_diff_max":0.007486581802368164,"val-aux/num_turns/min":2,"val-core/sciknoweval/acc/maj@16/mean":0.7341666666666667,"val-aux/sciknoweval/score/worst@2/mean":0.6025857142857143,"global_seqlen/mean":36125.5,"val-aux/sciknoweval/score/best@4/std":0.128790863116223,"response_length/min":260,"val-aux/sciknoweval/incorrect_format/maj@4/std":0.06141628384478388,"val-aux/sciknoweval/score/worst@4/mean":0.5304333333333333,"val-aux/sciknoweval/reward/best@8/std":0.08310585797508012,"rollout_corr/rollout_log_ppl":0.7630853808368556,"rollout_corr/rollout_ppl":2.194364957511425,"_runtime":6166.310807234,"val-aux/sciknoweval/acc/maj@2/mean":0.6898333333333334,"val-aux/num_turns/mean":2,"actor/kl_loss":0,"rollout_corr/rollout_is_seq_max_deviation":0.0067999958992004395,"val-aux/sciknoweval/reward/best@16/std":0.046578487700934236,"val-aux/sciknoweval/score/maj@8/std":0.13381993498072356,"val-aux/sciknoweval/incorrect_format/worst@16/std":0.0965096914106855,"rollout_corr/ppl_ratio":1.000641372287646,"perf/mfu/actor_infer":0,"perf/throughput":913.4161205674948,"val-aux/sciknoweval/incorrect_format/worst@8/mean":0.8332238095238094,"val-aux/sciknoweval/incorrect_format/maj@4/mean":0.9825095238095238,"rollout_corr/rollout_is_ratio_fraction_high":0,"val-aux/sciknoweval/score/worst@16/mean":0.4326952380952381,"val-aux/sciknoweval/reward/maj@16/mean":0.7341666666666667,"actor/ppo_kl":4.644049759150448e-05,"val-aux/sciknoweval/reward/maj@2/std":0.21523251642703473,"actor/pg_clipfrac":0.00016279506189675885,"response_length_non_aborted/clip_ratio":0,"val-aux/sciknoweval/score/maj@4/std":0.17575431812173817,"val-aux/sciknoweval/reward/best@4/mean":0.8483380952380953,"perf/cpu_memory_used_gb":105.93215942382812,"critic/score/mean":0.67578125,"val-aux/sciknoweval/acc/best@2/mean":0.7771380952380953,"timing_per_token_ms/gen":0.1042308475514874,"val-aux/sciknoweval/incorrect_format/best@4/mean":0.9996333333333334,"val-aux/sciknoweval/acc/worst@16/mean":0.4326952380952381,"perf/max_memory_allocated_gb":132.26526165008545,"timing_s/agent_loop/slowest/generate_sequences":15.491487443447113,"val-aux/sciknoweval/score/best@8/std":0.08310585797508012,"val-aux/sciknoweval/reward/maj@4/mean":0.7146857142857143,"timing_s/step":39.5498822350055,"val-core/sciknoweval/acc/best@16/std":0.046578487700934236,"timing_s/save_checkpoint":12.889757573604584,"rollout_corr/training_ppl":2.1957929017953575,"val-aux/sciknoweval/score/maj@2/mean":0.6898333333333334,"critic/advantages/min":-0.875,"val-core/sciknoweval/acc/maj@16/std":0.10114866635456188,"perf/mfu/actor":0.13613751711674105,"rollout_corr/rollout_is_min":0.08076818287372589,"timing_s/agent_loop/tool_calls/mean":0,"timing_s/agent_loop/generate_sequences/max":15.491487443447113} |