Files
Qwen3-4B-Material-GRPO-TR/artifacts/wandb-summary.json

1 line
11 KiB
JSON
Raw Normal View History

{"val-aux/num_turns/mean":2,"val-aux/sciknoweval/reward/maj@16/mean":0.7719148936170213,"val-aux/sciknoweval/score/maj@16/std":0.018618881588821194,"response_length_non_aborted/max":1110,"val-core/sciknoweval/acc/best@16/std":0.00848440146040565,"actor/entropy":0.05979691818356514,"rollout_corr/chi2_token":0.0007009264081716537,"val-aux/sciknoweval/incorrect_format/worst@4/mean":0.9895957446808511,"timing_s/agent_loop/tool_calls/max":0,"val-aux/sciknoweval/incorrect_format/maj@8/std":0.0012050425965809124,"perf/max_memory_reserved_gb":129.208984375,"val-aux/sciknoweval/incorrect_format/maj@16/mean":1,"critic/advantages/min":-0.875,"val-aux/sciknoweval/incorrect_format/worst@2/std":0.011671809930422904,"val-core/sciknoweval/acc/mean@16":0.7626329787234043,"val-aux/sciknoweval/reward/std@16":0.06729335213589911,"global_seqlen/minmax_diff":12650,"rollout_corr/rollout_is_eff_sample_size":0.9990782896419121,"val-aux/sciknoweval/reward/best@8/std":0.014096974337829643,"val-aux/sciknoweval/reward/maj@4/std":0.04686638157004486,"val-aux/sciknoweval/acc/maj@2/std":0.06820989645423031,"val-aux/sciknoweval/incorrect_format/best@8/mean":1,"val-aux/sciknoweval/reward/best@16/std":0.00848440146040565,"val-aux/sciknoweval/reward/best@2/mean":0.7876382978723404,"val-aux/sciknoweval/incorrect_format/maj@2/mean":0.996872340425532,"val-aux/sciknoweval/reward/best@2/std":0.041836077372629385,"prompt_length/max":479,"timing_s/agent_loop/slowest/generate_sequences":8.524729896336794,"val-aux/sciknoweval/incorrect_format/maj@4/mean":0.9992021276595746,"val-aux/sciknoweval/score/best@2/std":0.041836077372629385,"val-aux/sciknoweval/score/worst@2/std":0.07115878526747459,"val-aux/sciknoweval/reward/maj@8/std":0.026416313160677763,"timing_s/agent_loop/slowest/prompt_length":276,"critic/returns/min":-0.875,"val-aux/sciknoweval/incorrect_format/best@16/std":0,"val-aux/sciknoweval/score/worst@2/mean":0.7377765957446808,"critic/score/max":1,"critic/score/min":0,"val-aux/sciknoweval/reward/maj@8/mean":0.7723617021276596,"rollout_corr/rollout_is_ratio_fraction_low":0.0001169390743598342,"val-aux/sciknoweval/reward/maj@16/std":0.018618881588821194,"prompt_length/min":190,"val-aux/sciknoweval/acc/best@8/mean":0.8101914893617022,"val-aux/sciknoweval/score/best@4/mean":0.801627659574468,"val-aux/sciknoweval/incorrect_format/worst@2/mean":0.9942765957446809,"global_seqlen/max":33657,"val-aux/sciknoweval/incorrect_format/best@2/mean":0.9995851063829787,"val-aux/sciknoweval/acc/worst@2/std":0.07115878526747459,"critic/rewards/min":0,"val-aux/sciknoweval/score/best@16/mean":0.8149574468085106,"actor/pg_clipfrac_lower":0,"val-aux/sciknoweval/score/maj@2/mean":0.7628297872340426,"timing_s/start_profile":9.438768029212952e-05,"response_length/mean":434.25390625,"rollout_corr/rollout_is_ratio_fraction_high":4.497656846069731e-05,"_wandb":{"runtime":3382},"val-aux/sciknoweval/score/maj@4/mean":0.770308510638298,"val-aux/num_turns/max":2,"timing_s/agent_loop/generate_sequences/mean":4.727881735328992,"val-aux/sciknoweval/score/maj@16/mean":0.7719148936170213,"val-aux/sciknoweval/score/best@8/std":0.014096974337829643,"perf/cpu_memory_used_gb":103.06706619262695,"timing_s/old_log_prob":2.3261021357029676,"timing_s/gen":14.453305700793862,"rollout_corr/k3_kl":0.000666843313986476,"val-aux/sciknoweval/reward/maj@4/mean":0.770308510638298,"val-aux/sciknoweval/reward/best@16/mean":0.8149574468085106,"timing_s/step":26.7111904527992,"rollout_corr/kl":0.0009433553804747419,"perf/time_per_step":26.7111904527992,"val-aux/sciknoweval/reward/worst@16/mean":0.6536170212765957,"val-aux/sciknoweval/incorrect_format/std@16":0.009302503118891777,"val-core/sciknoweval/acc/best@16/mean":0.8149574468085106,"val-aux/sciknoweval/reward/best@4/mean":0.801627659574468,"timing_s/stop_profile":0.00014218315482139587,"val-aux/sciknoweval/score/maj@8/mean":0.7723617021276596,"val-aux/sciknoweval/incorrect_format/worst@8/mean":0.9830851063829789,"timing_s/testing":94.38914828561246,"prompt_length/mean":391.125,"timing_s/agent_loop/slowest/tool_calls":0,"response_length_non_aborted/min":197,"v