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

1 line
10 KiB
JSON
Raw Permalink Normal View History

{"rollout_corr/rollout_is_std":0.03857421875,"val-aux/sciknoweval/incorrect_format/maj@2/std":0.0023491098553929386,"training/global_step":100,"val-aux/sciknoweval/score/worst@8/std":0.0857140101161578,"val-aux/sciknoweval/incorrect_format/maj@4/mean":0.9998999999999999,"timing_per_token_ms/update_actor":0.050081200018674395,"val-aux/sciknoweval/reward/mean@16":0.6880952380952381,"val-aux/sciknoweval/score/maj@8/std":0.06145786769010523,"actor/grad_norm":0.19856154546141624,"critic/returns/max":0.75,"timing_s/agent_loop/tool_calls/min":0,"timing_s/stop_profile":0.00011307559907436371,"val-aux/sciknoweval/incorrect_format/maj@8/mean":1,"val-aux/sciknoweval/reward/best@2/mean":0.733652380952381,"val-aux/sciknoweval/score/best@8/std":0.03693927335257077,"val-aux/sciknoweval/reward/best@16/std":0.022230541736786433,"timing_s/agent_loop/generate_sequences/mean":4.972953367498121,"critic/score/min":0,"val-aux/num_turns/min":2,"prompt_length/clip_ratio":0,"val-aux/sciknoweval/incorrect_format/worst@16/std":0.004551356214978213,"val-aux/sciknoweval/score/maj@8/mean":0.6996714285714285,"val-aux/sciknoweval/reward/maj@2/mean":0.6873380952380953,"val-aux/sciknoweval/acc/best@2/std":0.08603291821062148,"val-aux/sciknoweval/reward/maj@8/std":0.06145786769010523,"val-aux/sciknoweval/reward/worst@4/std":0.10105952090465425,"val-aux/sciknoweval/incorrect_format/maj@4/std":0.0009636106375423148,"val-aux/sciknoweval/incorrect_format/best@8/mean":1,"actor/pg_clipfrac":8.191586903194548e-05,"val-aux/sciknoweval/score/best@16/mean":0.8012238095238096,"rollout_corr/chi2_token":0.0015719085931777954,"prompt_length/max":625,"val-aux/sciknoweval/reward/best@4/mean":0.7654619047619048,"val-aux/sciknoweval/reward/best@16/mean":0.8012238095238096,"val-aux/sciknoweval/incorrect_format/best@8/std":0,"val-aux/sciknoweval/acc/best@4/std":0.05713592644762481,"val-aux/sciknoweval/score/best@4/mean":0.7654619047619048,"rollout_corr/rollout_is_ratio_fraction_low":5.7171793741872534e-05,"val-aux/sciknoweval/reward/worst@16/std":0.06804166756222384,"val-aux/sciknoweval/incorrect_format/maj@16/std":0,"val-aux/sciknoweval/acc/maj@8/std":0.06145786769010523,"timing_s/agent_loop/generate_sequences/max":8.91172250919044,"val-aux/sciknoweval/score/maj@4/std":0.0900863886286685,"training/epoch":1,"val-aux/sciknoweval/score/best@2/std":0.08603291821062148,"val-aux/sciknoweval/acc/worst@4/std":0.10105952090465425,"val-aux/sciknoweval/score/maj@2/std":0.11903344827555573,"val-core/sciknoweval/acc/best@16/std":0.022230541736786433,"val-core/sciknoweval/acc/maj@16/mean":0.6995095238095238,"rollout_corr/rollout_is_seq_min":0.9923958778381348,"val-aux/sciknoweval/incorrect_format/worst@2/std":0.003080186451482751,"critic/rewards/max":1,"_wandb":{"runtime":4227},"val-aux/sciknoweval/reward/maj@4/std":0.0900863886286685,"perf/total_num_tokens":198254,"rollout_corr/rollout_ppl":1.5293380408547819,"rollout_corr/rollout_is_seq_std":0.0019207203295081854,"val-aux/sciknoweval/acc/std@16":0.11930428433281097,"response_length_non_aborted/min":194,"val-aux/sciknoweval/acc/worst@16/std":0.06804166756222384,"val-aux/sciknoweval/reward/worst@2/mean":0.6426714285714287,"val-aux/sciknoweval/acc/maj@2/mean":0.6873380952380953,"val-aux/sciknoweval/reward/worst@16/mean":0.5196142857142857,"global_seqlen/balanced_max":24784,"critic/score/mean":0.62890625,"rollout_corr/rollout_is_min":0.3767692446708679,"perf/cpu_memory_used_gb":104.02334213256836,"prompt_length/min":165,"perf/time_per_step":27.04090141505003,"training/rollout_probs_diff_std":0.01128289382904768,"global_seqlen/minmax_diff":7976,"actor/pg_clipfrac_lower":0,"rollout_corr/rollout_is_seq_max_deviation":0.007774233818054199,"val-aux/sciknoweval/score/maj@16/mean":0.6995095238095238,"val-core/sciknoweval/acc/maj@16/std":0.046993204377870766,"val-aux/sciknoweval/acc/worst@16/mean":0.5196142857142857,"rollout_corr/training_log_ppl":0.41552202677121386,"val-aux/sciknoweval/incorrect_format/best@2/mean":0.999952380952381,"rollout_corr/k3_kl":0.0007973912350962564,"val-aux/sciknoweval/reward/worst@8/mean":0.5585714285714286,"training/