Files
Qwen3-8B-Material-GRPO-TR/artifacts/wandb-summary.json

1 line
10 KiB
JSON
Raw Normal View History

{"val-aux/sciknoweval/acc/worst@8/mean":0.7248191489361702,"timing_s/agent_loop/generate_sequences/max":7.582470629364252,"val-aux/sciknoweval/score/best@8/std":0.019003297583575607,"critic/score/mean":0.66796875,"perf/max_memory_allocated_gb":128.6631588935852,"val-aux/sciknoweval/reward/best@8/std":0.019003297583575607,"response_length/min":181,"timing_s/agent_loop/slowest/response_length":812,"timing_per_token_ms/gen":0.16229245153852395,"_wandb":{"runtime":3789},"rollout_corr/rollout_is_seq_mean":0.9999960064888,"val-aux/sciknoweval/incorrect_format/best@4/std":0,"rollout_corr/rollout_is_max":1.6438628435134888,"perf/time_per_step":31.49581053480506,"val-aux/sciknoweval/incorrect_format/worst@2/std":0,"critic/score/max":1,"val-aux/sciknoweval/score/best@8/mean":0.8243297872340425,"rollout_corr/rollout_is_ratio_fraction_high":0,"rollout_corr/kl":0.0003323329324800284,"timing_s/agent_loop/slowest/tool_calls":0,"val-aux/sciknoweval/incorrect_format/maj@2/mean":1,"val-aux/sciknoweval/incorrect_format/best@8/mean":1,"num_turns/min":2,"val-aux/sciknoweval/incorrect_format/maj@8/std":0,"val-aux/sciknoweval/acc/maj@8/std":0.02545128949215922,"global_seqlen/balanced_min":24226,"val-aux/sciknoweval/incorrect_format/best@16/mean":1,"perf/mfu/actor":0.09790439064664765,"prompt_length/max":479,"val-aux/sciknoweval/incorrect_format/maj@16/std":0,"timing_s/reward":0.0619368739426136,"training/epoch":3,"val-aux/num_turns/max":2,"rollout_corr/rollout_is_mean":0.9999920725822449,"val-aux/sciknoweval/reward/best@8/mean":0.8243297872340425,"val-aux/sciknoweval/reward/worst@8/std":0.035434813866661946,"response/aborted_ratio":0,"rollout_corr/log_ppl_abs_diff":0.0012461927544791251,"val-aux/sciknoweval/score/worst@16/std":0.027047818566610645,"val-aux/sciknoweval/reward/worst@8/mean":0.7248191489361702,"timing_s/agent_loop/slowest/prompt_length":439,"val-aux/sciknoweval/incorrect_format/best@4/mean":1,"critic/rewards/mean":0.66796875,"response_length_non_aborted/min":181,"response_length/clip_ratio":0,"val-aux/sciknoweval/acc/best@8/mean":0.8243297872340425,"val-aux/sciknoweval/reward/best@2/mean":0.799936170212766,"timing_per_token_ms/adv":0.0006066613285346271,"val-aux/sciknoweval/reward/worst@2/mean":0.7602021276595745,"global_seqlen/minmax_diff":8023,"global_seqlen/max":28642,"val-aux/sciknoweval/acc/maj@2/mean":0.7798829787234042,"timing_s/gen":15.210048558190465,"rollout_corr/rollout_is_seq_max_deviation":0.006072938442230225,"rollout_corr/log_ppl_diff_max":0.006919294595718384,"timing_s/start_profile":0.00011714361608028412,"val-aux/sciknoweval/reward/std@16":0.052101428059437495,"rollout_corr/rollout_is_seq_fraction_low":0,"actor/pg_loss":-1.95368193089962e-06,"response_length/mean":366.09375,"val-aux/sciknoweval/incorrect_format/best@2/std":0,"val-aux/sciknoweval/reward/worst@16/std":0.027047818566610645,"training/global_step":100,"val-aux/sciknoweval/incorrect_format/maj@4/mean":1,"timing_s/agent_loop/tool_calls/max":0,"val-core/sciknoweval/acc/maj@16/mean":0.7843510638297873,"val-core/sciknoweval/acc/best@16/std":0.013309078916541482,"val-aux/sciknoweval/score/mean@16":0.7799202127659575,"actor/pg_clipfrac_lower":0,"response_length_non_aborted/max":851,"val-aux/sciknoweval/score/maj@4/mean":0.7823297872340426,"val-aux/sciknoweval/score/best@4/mean":0.8139574468085107,"critic/advantages/max":0.375,"_runtime":3789.512184693,"perf/mfu/actor_infer":0,"rollout_corr/log_ppl_diff_min":-0.004684805870056152,"num_turns/max":2,"timing_s/agent_loop/generate_sequences/mean":3.872075782623142,"val-aux/sciknoweval/reward/best@4/std":0.027455777877525876,"val-aux/sciknoweval/score/best@16/std":0.013309078916541482,"rollout_corr/rollout_is_ratio_fraction_low":0,"rollout_corr/rollout_is_eff_sample_size":0.9992487118690302,"val-aux/num_turns/min":2,"timing_s/save_checkpoint":12.71921988017857,"training/rollout_probs_diff_valid":1,"val-aux/sciknoweval/score/worst@2/std":0.0484960635134995,"val-aux/sciknoweval/score/worst@2/mean":0.7602021276595745,"training/rollout_actor_probs_pearson_corr":0.9996216893196106,"prompt_length/mean":391.125,"val-au