1 line
11 KiB
JSON
1 line
11 KiB
JSON
{"response_length/mean":279.4140625,"critic/advantages/max":0.75,"val-aux/sciknoweval/acc/best@2/std":0.02918137925697454,"rollout_corr/rollout_log_ppl":0.05304120097935083,"val-aux/sciknoweval/incorrect_format/worst@8/mean":1,"rollout_corr/rollout_is_seq_max":1.0095514059066772,"response_length/clip_ratio":0,"training/rollout_probs_diff_mean":0.00194946164265275,"timing_s/agent_loop/generate_sequences/min":1.194736858829856,"rollout_corr/rollout_is_seq_min":0.996039092540741,"_timestamp":1.7830039799930708e+09,"val-aux/sciknoweval/score/maj@16/mean":0.7758829787234042,"val-aux/sciknoweval/score/maj@16/std":0.014657220600786369,"val-aux/sciknoweval/reward/best@2/mean":0.7854148936170213,"val-aux/sciknoweval/reward/best@16/mean":0.8047021276595744,"rollout_corr/log_ppl_diff":0.0006709559474984417,"critic/returns/min":-0.625,"rollout_corr/rollout_is_seq_fraction_high":0,"actor/entropy":0.052000705152750015,"critic/score/max":1,"val-core/sciknoweval/acc/best@16/std":0.005902094678345516,"val-aux/sciknoweval/score/best@16/mean":0.8047021276595744,"val-aux/sciknoweval/incorrect_format/best@4/std":0,"val-aux/sciknoweval/acc/worst@4/mean":0.7343617021276596,"rollout_corr/rollout_is_mean":0.9999290108680725,"timing_s/start_profile":8.36830586194992e-05,"val-aux/sciknoweval/acc/worst@16/std":0.02298913016887447,"rollout_corr/rollout_is_ratio_fraction_high":1.3980147741676774e-05,"val-aux/sciknoweval/reward/best@4/std":0.01639341752615574,"self_distillation/token_reweight_correct_sample_fraction":0.68359375,"rollout_corr/rollout_is_seq_fraction_low":0,"self_distillation/token_reweight_w_std":431654.97193862277,"val-aux/sciknoweval/score/worst@16/std":0.02298913016887447,"val-aux/sciknoweval/reward/worst@8/std":0.03147132698227977,"critic/rewards/max":1,"perf/mfu/actor_infer":0,"val-aux/sciknoweval/score/best@2/std":0.02918137925697454,"val-aux/sciknoweval/score/maj@4/std":0.031426065862592674,"timing_per_token_ms/update_actor":0.08712704845144265,"val-aux/sciknoweval/incorrect_format/worst@16/mean":1,"val-aux/sciknoweval/incorrect_format/worst@2/std":0,"rollout_corr/rollout_is_seq_std":0.0017657832941040397,"self_distillation/token_reweight_w_clip_frac":0.012708691239822656,"num_turns/min":2,"self_distillation/empty_target_batch":0.28125,"response_length_non_aborted/min":91,"val-aux/num_turns/max":2,"val-aux/num_turns/min":2,"training/rollout_actor_probs_pearson_corr":0.9965116381645203,"critic/score/min":0,"val-aux/sciknoweval/incorrect_format/worst@4/mean":1,"perf/mfu/actor":0.03965654144334235,"self_distillation/token_reweight_active_token_fraction":0.71875,"timing_s/agent_loop/generate_sequences/max":4.990295339375734,"global_seqlen/min":18525,"val-aux/sciknoweval/incorrect_format/best@8/std":0,"num_turns/max":2,"prompt_length/mean":391.125,"val-aux/sciknoweval/incorrect_format/best@16/std":0,"val-aux/sciknoweval/score/worst@8/std":0.03147132698227977,"global_seqlen/max":24595,"num_turns/mean":2,"rollout_corr/ppl_ratio":1.0006739916279912,"val-aux/sciknoweval/acc/maj@2/std":0.04273575303351148,"val-aux/sciknoweval/incorrect_format/worst@16/std":0,"actor/pg_clipfrac_lower":0,"global_seqlen/balanced_max":21459,"actor/grad_norm":0.21034935116767883,"response_length_non_aborted/mean":279.4140625,"val-aux/sciknoweval/score/best@16/std":0.005902094678345516,"val-aux/sciknoweval/incorrect_format/worst@8/std":0,"timing_s/old_log_prob":2.2547264005988836,"val-aux/sciknoweval/reward/worst@8/mean":0.7189574468085107,"actor/pg_loss":0.0006925989873707294,"prompt_length/clip_ratio":0,"val-aux/sciknoweval/reward/best@2/std":0.02918137925697454,"val-aux/sciknoweval/score/std@16":0.04264975776630575,"_runtime":3535.705405695,"perf/max_memory_allocated_gb":117.12247371673584,"val-aux/sciknoweval/acc/maj@2/mean":0.7684468085106384,"val-aux/sciknoweval/acc/worst@8/std":0.03147132698227977,"val-aux/sciknoweval/incorrect_format/best@8/mean":1,"rollout_corr/rollout_is_seq_mean":0.9999463558197021,"val-aux/sciknoweval/incorrect_format/maj@2/mean":1,"prompt_length/min":190,"timing_per_token_ms/adv":0.0029145551329557463,"global_seqlen/balanced_min":21455,"rollout_corr/log_ppl_diff_min":-0.005156993865966797,"val-aux/sciknoweval/reward/best@4/mean":0.7959787234042554,"response_length/min":91,"val-aux/sciknoweval/score/worst@4/std":0.03784947100976082,"val-aux/sciknoweval/incorrect_format/maj@16/std":0,"timing_s/update_actor":14.95605488307774,"critic/advantages/min":-0.625,"timing_s/agent_loop/tool_calls/mean":0,"self_distillation/success_group_fraction":0.71875,"val-aux/sciknoweval/reward/best@16/std":0.005902094678345516,"self_distillation/token_reweight_context_sample_fraction":0.71875,"self_distillation/token_reweight_w_mean":32783.957445921376,"actor/pg_clipfrac":0.00034995409805560485,"val-aux/sciknoweval/incorrect_format/maj@16/mean":1,"rollout_corr/training_log_ppl":0.053712156926849275,"val-aux/sciknoweval/score/worst@2/std":0.042424831149873776,"self_distillation/correct_sample_fraction":0.68359375,"critic/rewards/min":0,"global_seqlen/minmax_diff":6070,"val-aux/sciknoweval/score/best@8/std":0.008774628971156027,"val-aux/sciknoweval/score/best@4/std":0.01639341752615574,"perf/total_num_tokens":171658,"val-aux/sciknoweval/score/maj@2/std":0.04273575303351148,"val-aux/sciknoweval/acc/maj@4/std":0.031426065862592674,"rollout_corr/rollout_is_max":2.706778049468994,"val-aux/sciknoweval/acc/worst@2/mean":0.7512340425531915,"val-aux/sciknoweval/incorrect_format/maj@2/std":0,"val-aux/sciknoweval/score/worst@16/mean":0.705404255319149,"rollout_corr/rollout_is_min":0.4680381119251251,"timing_s/adv":0.5003067050129175,"self_distillation/token_reweight_eps_w":0.20000000000000004,"timing_s/stop_profile":9.810179471969604e-05,"val-aux/sciknoweval/score/maj@4/mean":0.7722765957446808,"val-aux/sciknoweval/incorrect_format/mean@16":1,"rollout_corr/k3_kl":0.0005436672919252317,"val-aux/sciknoweval/acc/std@16":0.04264975776630575,"actor/kl_loss":0,"perf/throughput":747.7238083182178,"rollout_corr/chi2_token":0.0007216199301183224,"rollout_corr/chi2_seq":0.15093162842094898,"_step":100,"val-aux/sciknoweval/incorrect_format/best@16/mean":1,"val-aux/sciknoweval/reward/maj@2/std":0.04273575303351148,"self_distillation/success_sample_fraction":0.71875,"timing_s/testing":36.83901543915272,"timing_s/reward":0.05821913294494152,"rollout_corr/rollout_is_eff_sample_size":0.9992544253530163,"val-core/sciknoweval/acc/maj@16/std":0.014657220600786369,"global_seqlen/mean":21457.25,"rollout_corr/rollout_is_ratio_fraction_low":6.990074325585738e-05,"val-aux/sciknoweval/score/worst@2/mean":0.7512340425531915,"val-aux/sciknoweval/reward/worst@2/mean":0.7512340425531915,"val-aux/sciknoweval/incorrect_format/best@2/mean":1,"val-aux/sciknoweval/acc/maj@8/std":0.021830599810403712,"val-aux/sciknoweval/score/mean@16":0.7686170212765957,"training/rollout_probs_diff_max":0.30163344740867615,"val-aux/sciknoweval/reward/worst@4/mean":0.7343617021276596,"critic/score/mean":0.68359375,"val-aux/sciknoweval/score/maj@2/mean":0.7684468085106384,"timing_s/agent_loop/tool_calls/max":0,"training/rollout_probs_diff_std":0.010225724428892136,"self_distillation/reprompt_sample_fraction":0.71875,"self_distillation/feedback_used_fraction":0,"critic/advantages/mean":-0.0022892493288964033,"val-aux/sciknoweval/incorrect_format/maj@4/std":0,"val-aux/sciknoweval/acc/maj@4/mean":0.7722765957446808,"critic/returns/max":0.75,"val-aux/sciknoweval/score/maj@8/mean":0.7746063829787234,"val-aux/sciknoweval/reward/best@8/mean":0.801563829787234,"val-aux/sciknoweval/score/worst@4/mean":0.7343617021276596,"val-core/sciknoweval/acc/maj@16/mean":0.7758829787234042,"perf/cpu_memory_used_gb":147.16178703308105,"_wandb":{"runtime":3535},"val-aux/sciknoweval/acc/best@4/std":0.01639341752615574,"rollout_corr/kl":0.0006709553122909995,"timing_s/gen":10.843061864376068,"val-aux/sciknoweval/acc/worst@16/mean":0.705404255319149,"response_length_non_aborted/clip_ratio":0,"rollout_corr/rollout_ppl":1.0561046437360346,"rollout_corr/log_ppl_abs_diff":0.001628608099053963,"val-aux/sciknoweval/reward/mean@16":0.7686170212765957,"val-aux/sciknoweval/reward/maj@4/std":0.031426065862592674,"val-aux/sciknoweval/reward/maj@8/std":0.021830599810403712,"val-aux/sciknoweval/reward/worst@16/std":0.02298913016887447,"val-aux/sciknoweval/reward/worst@16/mean":0.705404255319149,"val-aux/sciknoweval/acc/worst@4/std":0.03784947100976082,"val-aux/sciknoweval/incorrect_format/worst@2/mean":1,"val-aux/sciknoweval/score/maj@8/std":0.021830599810403712,"actor/lr":1e-06,"val-aux/sciknoweval/score/best@8/mean":0.801563829787234,"val-aux/sciknoweval/score/best@2/mean":0.7854148936170213,"training/global_step":100,"val-aux/sciknoweval/incorrect_format/maj@8/mean":1,"val-aux/sciknoweval/reward/worst@4/std":0.03784947100976082,"rollout_corr/rollout_is_std":0.027313215658068657,"critic/returns/mean":-0.0022892493288964033,"val-aux/sciknoweval/reward/worst@2/std":0.042424831149873776,"timing_s/agent_loop/slowest/generate_sequences":4.990295339375734,"self_distillation/feedback_available_fraction":0,"val-aux/sciknoweval/acc/best@8/std":0.008774628971156027,"val-aux/sciknoweval/reward/best@8/std":0.008774628971156027,"val-aux/sciknoweval/acc/best@4/mean":0.7959787234042554,"timing_s/agent_loop/slowest/response_length":589,"perf/time_per_step":28.696759099140763,"val-aux/sciknoweval/incorrect_format/worst@4/std":0,"val-aux/sciknoweval/reward/maj@4/mean":0.7722765957446808,"val-aux/sciknoweval/reward/maj@16/std":0.014657220600786369,"val-aux/sciknoweval/incorrect_format/best@2/std":0,"val-core/sciknoweval/acc/mean@16":0.7686170212765957,"val-aux/sciknoweval/acc/worst@2/std":0.042424831149873776,"val-aux/sciknoweval/acc/maj@8/mean":0.7746063829787234,"timing_s/agent_loop/tool_calls/min":0,"val-aux/sciknoweval/incorrect_format/maj@4/mean":1,"self_distillation/token_reweight_lambda":0.5,"val-aux/sciknoweval/score/worst@8/mean":0.7189574468085107,"timing_s/agent_loop/slowest/prompt_length":276,"timing_s/agent_loop/generate_sequences/mean":3.1610305702342885,"critic/rewards/mean":0.68359375,"rollout_corr/log_ppl_diff_max":0.01552504301071167,"actor/ppo_kl":0.00022765648916411863,"perf/max_memory_reserved_gb":121.533203125,"training/rollout_probs_diff_valid":1,"training/epoch":3,"val-aux/sciknoweval/acc/worst@8/mean":0.7189574468085107,"val-core/sciknoweval/acc/best@16/mean":0.8047021276595744,"val-aux/sciknoweval/acc/best@8/mean":0.801563829787234,"timing_s/agent_loop/slowest/tool_calls":0,"timing_per_token_ms/gen":0.15158761169266138,"rollout_corr/training_ppl":1.056842735502869,"val-aux/sciknoweval/incorrect_format/std@16":0,"val-aux/sciknoweval/score/best@4/mean":0.7959787234042554,"val-aux/sciknoweval/incorrect_format/maj@8/std":0,"rollout_corr/rollout_is_seq_max_deviation":0.009551405906677246,"timing_s/step":28.696759099140763,"val-aux/num_turns/mean":2,"val-aux/sciknoweval/incorrect_format/best@4/mean":1,"val-aux/sciknoweval/reward/maj@8/mean":0.7746063829787234,"response_length_non_aborted/max":589,"self_distillation/token_reweight_w_clipped_mean":1.0002633819822222,"response_length/max":589,"val-aux/sciknoweval/reward/maj@2/mean":0.7684468085106384,"val-aux/sciknoweval/acc/best@2/mean":0.7854148936170213,"prompt_length/max":479,"val-aux/sciknoweval/reward/maj@16/mean":0.7758829787234042,"val-aux/sciknoweval/reward/std@16":0.04264975776630575,"response/aborted_ratio":0,"timing_s/save_checkpoint":6.8152069840580225} |