Files
Qwen3-4B-Chemistry-SDPO-TR/artifacts/wandb-summary.json

1 line
11 KiB
JSON
Raw Permalink Normal View History

{"critic/advantages/max":0.875,"response_length_non_aborted/max":8192,"prompt_length/mean":296.15625,"val-aux/sciknoweval/acc/best@8/mean":0.7998666666666667,"val-aux/sciknoweval/reward/maj@2/mean":0.7489761904761905,"val-aux/sciknoweval/reward/worst@8/std":0.02011337982070672,"val-aux/sciknoweval/score/maj@2/mean":0.7489761904761905,"val-aux/sciknoweval/acc/worst@2/std":0.046990424566282364,"timing_s/agent_loop/slowest/prompt_length":198,"training/rollout_actor_probs_pearson_corr":0.9987240433692932,"timing_s/agent_loop/slowest/generate_sequences":48.49703206680715,"rollout_corr/k3_kl":0.0001086825423562062,"rollout_corr/rollout_is_seq_min":0.9954840540885925,"val-aux/sciknoweval/reward/maj@16/std":0.028181641738688423,"val-aux/sciknoweval/reward/maj@4/mean":0.7514238095238096,"self_distillation/correct_sample_fraction":0.7265625,"num_turns/min":2,"rollout_corr/rollout_is_max":1.601178526878357,"timing_s/agent_loop/slowest/tool_calls":0,"val-aux/sciknoweval/acc/worst@2/mean":0.7262095238095239,"val-aux/sciknoweval/reward/mean@16":0.7491071428571429,"prompt_length/clip_ratio":0,"rollout_corr/rollout_is_seq_mean":0.9999521374702454,"val-aux/sciknoweval/incorrect_format/worst@16/std":0.002274256999100845,"self_distillation/success_group_fraction":0.875,"val-core/sciknoweval/acc/best@16/std":0.008797318267434817,"val-aux/sciknoweval/reward/best@16/std":0.008797318267434817,"global_seqlen/min":11694,"val-aux/sciknoweval/reward/maj@16/mean":0.7554761904761905,"val-aux/sciknoweval/reward/best@8/mean":0.7998666666666667,"timing_per_token_ms/gen":0.9042675971594261,"val-aux/sciknoweval/acc/best@8/std":0.017334238666592075,"val-aux/sciknoweval/incorrect_format/worst@16/mean":0.9969142857142858,"rollout_corr/rollout_is_seq_std":0.0012089080410078168,"val-aux/sciknoweval/score/worst@16/std":0.010972447210233164,"critic/advantages/mean":-0.09708639234304428,"rollout_corr/rollout_is_std":0.014016308821737766,"val-aux/sciknoweval/reward/worst@2/std":0.046990424566282364,"val-aux/sciknoweval/acc/worst@8/mean":0.6937857142857142,"timing_s/gen":54.36728074401617,"global_seqlen/mean":16992.375,"val-core/sciknoweval/acc/best@16/mean":0.8060142857142858,"perf/throughput":229.551173302718,"timing_s/agent_loop/generate_sequences/max":48.49703206680715,"timing_s/old_log_prob":2.4735527355223894,"val-aux/sciknoweval/incorrect_format/maj@4/std":0.0005594782149609339,"val-aux/sciknoweval/score/best@16/std":0.008797318267434817,"critic/score/mean":0.7265625,"self_distillation/success_sample_fraction":0.8671875,"val-aux/sciknoweval/incorrect_format/best@4/mean":1,"rollout_corr/rollout_is_seq_fraction_high":0,"val-aux/sciknoweval/incorrect_format/best@8/mean":1,"rollout_corr/rollout_is_seq_max_deviation":0.005106925964355469,"val-aux/num_turns/mean":2,"response_length_non_aborted/clip_ratio":0.0078125,"val-aux/sciknoweval/score/worst@4/mean":0.706704761904762,"val-aux/sciknoweval/acc/best@2/mean":0.771904761904762,"rollout_corr/rollout_is_seq_fraction_low":0,"val-aux/sciknoweval/score/maj@2/std":0.05319068044795249,"response_length/min":89,"actor/lr":1e-05,"critic/rewards/mean":0.7265625,"val-aux/sciknoweval/score/worst@16/mean":0.6871380952380951,"val-aux/sciknoweval/incorrect_format/maj@16/std":0,"perf/mfu/actor":0.03008261803486565,"rollout_corr/rollout_is_ratio_fraction_high":0,"perf/max_memory_reserved_gb":142.96875,"val-aux/sciknoweval/incorrect_format/best@2/mean":0.9999857142857144,"self_distillation/dynamic_weighting_enabled":0,"rollout_corr/chi2_seq":0.08643001760356128,"timing_s/agent_loop/tool_calls/max":0,"timing_s/testing":99.34055501781404,"val-aux/sciknoweval/score/best@4/mean":0.7891285714285714,"val-aux/sciknoweval/score/worst@2/mean":0.7262095238095239,"val-aux/sciknoweval/incorrect_format/maj@8/mean":1,"val-aux/sciknoweval/acc/best@2/std":0.0438263744703264,"perf/time_per_step":74.02434392087162,"actor/pg_loss":0.012388480715372907,"val-aux/sciknoweval/reward/maj@8/mean":0.7533142857142858,"timing_s/adv":0.4935148283839226,"response_length/mean":234.85546875,"_timestamp":1.7829831001197822e+09,"perf/mfu/actor_infer":0,"val-aux/s