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

1 line
10 KiB
JSON
Raw Normal View History

{"timing_s/testing":66.87851952016354,"response_length_non_aborted/clip_ratio":0,"critic/advantages/max":0.875,"val-aux/sciknoweval/acc/worst@2/mean":0.34018,"val-aux/sciknoweval/score/worst@16/std":0.032643145410378605,"rollout_corr/rollout_is_seq_max":1.0129314661026,"val-aux/sciknoweval/incorrect_format/maj@2/std":0.008749306906094262,"critic/score/min":0,"timing_s/agent_loop/tool_calls/min":0,"timing_s/agent_loop/tool_calls/mean":0,"training/rollout_actor_probs_pearson_corr":0.9986383318901062,"val-aux/sciknoweval/incorrect_format/best@8/std":0,"timing_s/agent_loop/generate_sequences/max":6.653986681252718,"global_seqlen/mean":19150.75,"val-core/sciknoweval/acc/best@16/mean":0.6543800000000001,"prompt_length/clip_ratio":0,"rollout_corr/rollout_is_seq_min":0.9921380877494812,"val-aux/sciknoweval/incorrect_format/maj@8/std":0.002504474311076614,"critic/advantages/min":-0.875,"val-aux/sciknoweval/acc/worst@8/mean":0.21248,"_timestamp":1.782992219032823e+09,"actor/pg_loss":5.068804603070021e-05,"rollout_corr/rollout_is_std":0.040638964623212814,"rollout_corr/rollout_is_seq_max_deviation":0.012931466102600098,"response_length/max":977,"rollout_corr/rollout_is_max":4.838179588317871,"val-aux/sciknoweval/reward/worst@16/std":0.032643145410378605,"val-aux/sciknoweval/reward/worst@8/std":0.07648178927728648,"timing_s/agent_loop/tool_calls/max":0,"val-aux/sciknoweval/reward/best@8/std":0.08028605088755732,"critic/rewards/max":1,"val-aux/sciknoweval/reward/best@2/std":0.1717994688879882,"val-core/sciknoweval/acc/mean@16":0.42625,"response_length/mean":248.5546875,"val-aux/sciknoweval/reward/mean@16":0.42625,"val-aux/sciknoweval/reward/maj@16/mean":0.44676000000000005,"val-aux/sciknoweval/acc/best@8/std":0.08028605088755732,"critic/returns/max":0.875,"_runtime":4362.153699158,"timing_per_token_ms/adv":0.000782728193439709,"val-aux/sciknoweval/score/best@8/std":0.08028605088755732,"val-aux/sciknoweval/reward/best@4/std":0.12262009307652777,"val-aux/sciknoweval/score/maj@8/mean":0.4453000000000001,"response_length_non_aborted/mean":248.5546875,"perf/max_memory_allocated_gb":123.67111110687256,"val-aux/sciknoweval/reward/maj@8/std":0.1155248617088842,"actor/ppo_kl":0.0007291194974783366,"val-aux/sciknoweval/acc/maj@4/mean":0.43562000000000006,"perf/max_memory_reserved_gb":129.48046875,"val-aux/num_turns/max":2,"global_seqlen/max":24171,"training/rollout_probs_diff_mean":0.005042023025453091,"val-aux/sciknoweval/acc/worst@16/mean":0.18634,"val-core/sciknoweval/acc/best@16/std":0.0472167585635409,"val-aux/sciknoweval/reward/maj@8/mean":0.4453000000000001,"perf/cpu_memory_used_gb":103.95842933654785,"global_seqlen/balanced_max":19154,"val-aux/sciknoweval/reward/worst@16/mean":0.18634,"timing_s/agent_loop/generate_sequences/min":0.7739909812808037,"val-aux/sciknoweval/reward/maj@4/std":0.16364220520548403,"val-aux/sciknoweval/reward/best@4/mean":0.5764,"response_length_non_aborted/max":977,"actor/grad_norm":0.496917188167572,"val-aux/sciknoweval/score/worst@8/mean":0.21248,"_wandb":{"runtime":4362},"timing_s/agent_loop/generate_sequences/mean":2.1721716843530885,"val-aux/sciknoweval/reward/best@8/mean":0.62372,"val-aux/sciknoweval/incorrect_format/best@16/std":0,"val-aux/sciknoweval/score/best@16/std":0.0472167585635409,"prompt_length/max":499,"actor/pg_clipfrac_lower":0,"val-aux/sciknoweval/score/maj@16/mean":0.44676000000000005,"val-core/sciknoweval/acc/maj@16/std":0.07340249583082686,"val-aux/sciknoweval/reward/worst@4/std":0.12852696069401104,"actor/pg_clipfrac":0.0010578534675005358,"rollout_corr/training_log_ppl":0.3114306655479595,"val-aux/sciknoweval/reward/best@16/mean":0.6543800000000001,"rollout_corr/rollout_is_eff_sample_size":0.9983516728081235,"val-aux/sciknoweval/incorrect_format/best@8/mean":1,"training/rollout_probs_diff_valid":1,"rollout_corr/kl":0.0015098723376403278,"val-aux/sciknoweval/score/best@16/mean":0.6543800000000001,"rollout_corr/rollout_is_seq_fraction_low":0,"actor/entropy":0.2835403382778168,"rollout_corr/log_ppl_diff_max":0.014475628733634949,"prompt_length/min":180,"val-aux/sciknoweval/reward/maj@2/me