1 line
10 KiB
JSON
1 line
10 KiB
JSON
{"val-aux/sciknoweval/reward/worst@2/mean":0.6398875,"val-aux/sciknoweval/incorrect_format/maj@2/mean":0.9178625,"perf/mfu/actor":0.1938518522795187,"training/rollout_actor_probs_pearson_corr":0.9995552897453308,"timing_s/start_profile":9.544752538204193e-05,"val-aux/sciknoweval/score/best@8/mean":0.94765,"response_length_non_aborted/mean":1854.16015625,"val-aux/sciknoweval/acc/best@4/mean":0.8962625000000001,"_timestamp":1.7830577481316543e+09,"rollout_corr/rollout_is_seq_fraction_high":0,"val-aux/sciknoweval/score/maj@16/std":0.11988946359651878,"timing_s/agent_loop/generate_sequences/min":3.829353081062436,"val-aux/sciknoweval/incorrect_format/best@8/std":0.006519985280372976,"num_turns/mean":2,"val-aux/sciknoweval/acc/worst@8/std":0.07452105831546266,"timing_s/agent_loop/slowest/prompt_length":342,"val-aux/sciknoweval/score/maj@16/mean":0.8134375,"val-aux/sciknoweval/acc/worst@4/mean":0.5704125,"rollout_corr/rollout_is_eff_sample_size":0.9992375231521491,"rollout_corr/rollout_is_seq_max":1.0046027898788452,"rollout_corr/rollout_is_mean":1.0000216960906982,"val-aux/sciknoweval/score/maj@8/mean":0.7921125,"val-aux/sciknoweval/reward/best@8/std":0.07747202634384016,"training/rollout_probs_diff_std":0.007731531746685505,"perf/max_memory_reserved_gb":144.7109375,"rollout_corr/kl":0.0003901403655550695,"val-aux/sciknoweval/reward/maj@4/std":0.18390036772261203,"val-aux/sciknoweval/reward/worst@4/mean":0.5704125,"val-aux/sciknoweval/acc/maj@4/mean":0.76555,"val-aux/sciknoweval/score/maj@2/mean":0.7299875,"actor/lr":1e-06,"val-aux/sciknoweval/reward/best@4/mean":0.8962625000000001,"critic/returns/min":-0.875,"val-aux/sciknoweval/reward/best@4/std":0.1329636020160485,"num_turns/max":2,"val-aux/sciknoweval/score/worst@16/mean":0.49838750000000004,"global_seqlen/balanced_min":67861,"val-aux/sciknoweval/incorrect_format/maj@2/std":0.1290617057662198,"timing_s/agent_loop/tool_calls/min":0,"response/aborted_ratio":0,"timing_s/update_actor":19.037666445598006,"timing_s/agent_loop/generate_sequences/mean":17.904436235730827,"training/rollout_probs_diff_max":0.5070421695709229,"timing_s/gen":74.55190750397742,"timing_s/old_log_prob":4.5896848775446415,"rollout_corr/rollout_is_ratio_fraction_high":0,"val-aux/sciknoweval/score/mean@16":0.7296875,"actor/pg_clipfrac":4.500477859892271e-05,"num_turns/min":2,"timing_s/agent_loop/slowest/generate_sequences":67.18781132809818,"val-aux/sciknoweval/incorrect_format/mean@16":0.91796875,"val-aux/sciknoweval/reward/best@16/std":0.03523755515622687,"timing_s/agent_loop/generate_sequences/max":67.18781132809818,"val-aux/sciknoweval/reward/std@16":0.21113028788197769,"val-aux/sciknoweval/acc/worst@8/mean":0.52575,"val-aux/sciknoweval/incorrect_format/maj@4/std":0.10365239649604263,"timing_per_token_ms/update_actor":0.035021397027595616,"val-aux/sciknoweval/incorrect_format/worst@2/mean":0.8673375,"val-aux/sciknoweval/incorrect_format/worst@4/std":0.13896621816207683,"val-aux/sciknoweval/incorrect_format/maj@8/mean":0.9701000000000001,"val-aux/sciknoweval/acc/best@8/mean":0.94765,"rollout_corr/training_log_ppl":0.35900981855229475,"actor/entropy":0.45818907022476196,"val-aux/sciknoweval/incorrect_format/best@4/std":0.02834928965541194,"val-aux/sciknoweval/incorrect_format/maj@8/std":0.07166964518321907,"rollout_corr/rollout_is_seq_mean":1.000037670135498,"val-aux/sciknoweval/acc/worst@4/std":0.12181204475036858,"val-aux/sciknoweval/incorrect_format/std@16":0.12921982287160344,"val-aux/sciknoweval/incorrect_format/best@4/mean":0.99245,"rollout_corr/log_ppl_diff_min":-0.003899216651916504,"critic/advantages/max":0.875,"val-core/sciknoweval/acc/maj@16/mean":0.8134375,"rollout_corr/log_ppl_abs_diff":0.0008009577286429703,"rollout_corr/rollout_is_seq_fraction_low":0,"timing_s/reward":0.11913642473518848,"rollout_corr/rollout_is_seq_max_deviation":0.004602789878845215,"val-aux/sciknoweval/incorrect_format/best@2/mean":0.969875,"_wandb":{"runtime":8906},"timing_per_token_ms/adv":0.000223674584299434,"val-aux/sciknoweval/reward/maj@16/mean":0.8134375,"val-aux/sciknoweval/reward/best@8/mean":0.94765,"perf/mfu/actor_infer":0,"_runtime":8906.629129377,"val-aux/sciknoweval/reward/maj@2/std":0.21049045089342963,"timing_s/agent_loop/tool_calls/max":0,"val-aux/sciknoweval/reward/worst@8/std":0.07452105831546266,"rollout_corr/log_ppl_diff_max":0.003635212779045105,"timing_s/testing":211.57785162888467,"val-aux/sciknoweval/score/maj@4/std":0.18390036772261203,"val-aux/sciknoweval/score/worst@2/mean":0.6398875,"rollout_corr/rollout_log_ppl":0.3586196781543549,"val-aux/sciknoweval/score/best@4/mean":0.8962625000000001,"global_seqlen/mean":67950.125,"val-aux/sciknoweval/reward/worst@2/std":0.1769580686853267,"timing_s/adv":0.12158972769975662,"rollout_corr/rollout_is_seq_min":0.9973651766777039,"val-aux/sciknoweval/reward/worst@16/std":0.045061773818715276,"val-aux/sciknoweval/score/maj@4/mean":0.76555,"val-aux/sciknoweval/reward/maj@8/std":0.15178369995448104,"perf/max_memory_allocated_gb":135.608256816864,"val-aux/sciknoweval/incorrect_format/maj@4/mean":0.9458125,"critic/score/min":0,"val-aux/sciknoweval/reward/mean@16":0.7296875,"val-core/sciknoweval/acc/maj@16/std":0.11988946359651878,"response_length_non_aborted/min":336,"rollout_corr/ppl_ratio":1.00039070751518,"val-aux/sciknoweval/acc/worst@2/std":0.1769580686853267,"val-core/sciknoweval/acc/best@16/std":0.03523755515622687,"timing_s/agent_loop/slowest/response_length":8192,"timing_s/stop_profile":0.00014124438166618347,"global_seqlen/max":110235,"global_seqlen/min":33724,"rollout_corr/rollout_is_seq_std":0.0009224399691447616,"val-aux/sciknoweval/acc/maj@2/mean":0.7299875,"critic/returns/max":0.875,"rollout_corr/rollout_is_ratio_fraction_low":4.213497959426604e-06,"val-aux/sciknoweval/acc/maj@8/std":0.15178369995448104,"val-aux/sciknoweval/incorrect_format/worst@2/std":0.14334823308252115,"rollout_corr/rollout_ppl":1.4562692898325622,"response_length/max":8192,"critic/advantages/mean":-0.023091811686754227,"val-aux/sciknoweval/acc/best@2/mean":0.8195625,"val-aux/sciknoweval/incorrect_format/worst@16/mean":0.695975,"val-aux/sciknoweval/reward/worst@8/mean":0.52575,"rollout_corr/rollout_is_min":0.46327918767929077,"val-aux/sciknoweval/acc/maj@4/std":0.18390036772261203,"response_length/mean":1854.16015625,"training/epoch":4,"val-aux/sciknoweval/score/maj@2/std":0.21049045089342963,"val-aux/sciknoweval/reward/maj@2/mean":0.7299875,"val-aux/sciknoweval/incorrect_format/best@16/mean":0.9999500000000001,"val-aux/sciknoweval/score/worst@8/mean":0.52575,"perf/cpu_memory_used_gb":107.14533233642578,"val-aux/sciknoweval/acc/worst@16/mean":0.49838750000000004,"rollout_corr/rollout_is_max":1.5396182537078857,"response_length_non_aborted/clip_ratio":0.01171875,"val-aux/sciknoweval/score/best@4/std":0.1329636020160485,"actor/kl_loss":0,"val-aux/sciknoweval/reward/best@2/std":0.18312693662059415,"timing_per_token_ms/gen":0.15706215436987647,"prompt_length/clip_ratio":0,"val-aux/sciknoweval/reward/maj@4/mean":0.76555,"timing_s/save_checkpoint":12.600243583321571,"rollout_corr/chi2_seq":3.438023897819221,"timing_s/step":98.50110033527017,"val-aux/sciknoweval/score/maj@8/std":0.15178369995448104,"critic/advantages/min":-0.875,"perf/total_num_tokens":543601,"val-aux/sciknoweval/incorrect_format/worst@16/std":0.07060170833571836,"actor/pg_clipfrac_lower":0,"val-aux/sciknoweval/score/best@2/mean":0.8195625,"val-aux/sciknoweval/score/std@16":0.21113028788197769,"val-aux/num_turns/max":2,"val-aux/sciknoweval/incorrect_format/best@16/std":0.0010787124614257437,"timing_s/agent_loop/tool_calls/mean":0,"val-aux/num_turns/mean":2,"timing_s/agent_loop/slowest/tool_calls":0,"val-aux/sciknoweval/incorrect_format/maj@16/mean":0.9865999999999999,"critic/returns/mean":-0.023091811686754227,"val-aux/sciknoweval/reward/maj@16/std":0.11988946359651878,"val-aux/sciknoweval/score/best@2/std":0.18312693662059415,"val-aux/sciknoweval/incorrect_format/maj@16/std":0.03874947109834628,"response_length/min":336,"actor/pg_loss":-1.3369019143283367e-05,"val-aux/sciknoweval/score/best@8/std":0.07747202634384016,"response_length_non_aborted/max":8192,"rollout_corr/k3_kl":0.00036932118109689327,"critic/rewards/mean":0.7734375,"prompt_length/max":383,"val-aux/sciknoweval/reward/maj@8/mean":0.7921125,"critic/score/max":1,"actor/grad_norm":0.04646363575011492,"rollout_corr/rollout_is_std":0.02762351557612419,"global_seqlen/minmax_diff":76511,"val-aux/sciknoweval/incorrect_format/best@8/mean":0.9992375000000001,"rollout_corr/training_ppl":1.4568524220958352,"global_seqlen/balanced_max":68027,"training/rollout_probs_diff_mean":0.0032882497180253267,"val-aux/sciknoweval/score/worst@4/std":0.12181204475036858,"val-aux/sciknoweval/incorrect_format/worst@8/std":0.112524959047603,"val-aux/sciknoweval/reward/worst@16/mean":0.49838750000000004,"val-aux/sciknoweval/acc/best@2/std":0.18312693662059415,"val-aux/sciknoweval/reward/worst@4/std":0.12181204475036858,"val-aux/sciknoweval/score/worst@8/std":0.07452105831546266,"val-core/sciknoweval/acc/mean@16":0.7296875,"rollout_corr/chi2_token":0.0006887984927743673,"prompt_length/mean":269.28125,"training/rollout_probs_diff_valid":1,"val-aux/sciknoweval/incorrect_format/best@2/std":0.07402285219911191,"actor/ppo_kl":6.788543978900052e-05,"val-aux/sciknoweval/reward/best@2/mean":0.8195625,"val-aux/sciknoweval/acc/std@16":0.21113028788197769,"_step":100,"rollout_corr/log_ppl_diff":0.000390140397939831,"val-aux/sciknoweval/acc/best@4/std":0.1329636020160485,"val-aux/sciknoweval/acc/maj@8/mean":0.7921125,"response_length/clip_ratio":0.01171875,"val-aux/sciknoweval/reward/best@16/mean":0.9732,"prompt_length/min":191,"critic/rewards/min":0,"training/global_step":100,"val-aux/sciknoweval/score/best@16/std":0.03523755515622687,"val-aux/sciknoweval/acc/best@8/std":0.07747202634384016,"val-aux/sciknoweval/score/worst@16/std":0.045061773818715276,"critic/score/mean":0.7734375,"critic/rewards/max":1,"val-aux/sciknoweval/incorrect_format/worst@4/mean":0.80435,"val-aux/sciknoweval/score/worst@2/std":0.1769580686853267,"val-core/sciknoweval/acc/best@16/mean":0.9732,"val-aux/sciknoweval/acc/worst@2/mean":0.6398875,"val-aux/sciknoweval/acc/maj@2/std":0.21049045089342963,"val-aux/sciknoweval/incorrect_format/worst@8/mean":0.7432874999999999,"val-aux/num_turns/min":2,"val-aux/sciknoweval/score/best@16/mean":0.9732,"perf/time_per_step":98.50110033527017,"val-aux/sciknoweval/score/worst@4/mean":0.5704125,"perf/throughput":689.8412786122876,"val-aux/sciknoweval/acc/worst@16/std":0.045061773818715276} |