Files
Qwen3-8B-Material-RLSD-TR/artifacts/wandb-summary.json
ModelHub XC 7f2f12b8da 初始化项目,由ModelHub XC社区提供模型
Model: SeongryongJung/Qwen3-8B-Material-RLSD-TR
Source: Original Platform
2026-08-12 14:13:18 +08:00

1 line
11 KiB
JSON

{"val-aux/sciknoweval/incorrect_format/best@8/std":0,"val-aux/num_turns/max":2,"self_distillation/feedback_used_fraction":0,"val-aux/sciknoweval/score/best@4/std":0.027759320424956824,"val-aux/sciknoweval/reward/best@2/mean":0.7809255319148936,"rollout_corr/rollout_log_ppl":0.15177409329044167,"val-aux/sciknoweval/incorrect_format/worst@4/std":0,"val-aux/sciknoweval/score/best@16/mean":0.813808510638298,"val-aux/sciknoweval/score/maj@4/std":0.031638291528035434,"val-aux/sciknoweval/score/maj@16/std":0.013016076917955125,"val-aux/sciknoweval/acc/worst@2/mean":0.7494468085106384,"val-aux/sciknoweval/score/maj@4/mean":0.764436170212766,"val-aux/sciknoweval/incorrect_format/worst@4/mean":1,"val-aux/sciknoweval/score/best@4/mean":0.7944468085106383,"response_length/clip_ratio":0,"self_distillation/token_reweight_eps_w":0.20000000000000004,"rollout_corr/log_ppl_diff_max":0.008212178945541382,"self_distillation/token_reweight_w_clipped_mean":0.9991688379086554,"val-aux/sciknoweval/score/maj@16/mean":0.7633617021276595,"timing_s/agent_loop/tool_calls/min":0,"critic/score/min":0,"critic/rewards/mean":0.67578125,"response_length_non_aborted/clip_ratio":0,"timing_s/save_checkpoint":12.537891793996096,"val-aux/sciknoweval/reward/maj@8/std":0.020384144360752216,"timing_s/agent_loop/slowest/generate_sequences":6.074218215420842,"global_seqlen/max":28053,"critic/advantages/min":-0.875,"val-aux/sciknoweval/acc/best@4/mean":0.7944468085106383,"self_distillation/reprompt_sample_fraction":0.6875,"val-core/sciknoweval/acc/best@16/mean":0.813808510638298,"val-aux/sciknoweval/acc/best@2/std":0.03462769102759328,"val-core/sciknoweval/acc/mean@16":0.7646276595744681,"rollout_corr/rollout_is_std":0.029670871794223785,"val-aux/sciknoweval/score/worst@2/mean":0.7494468085106384,"rollout_corr/log_ppl_diff":0.000486337645270396,"val-aux/sciknoweval/acc/worst@8/std":0.018341796156463988,"rollout_corr/log_ppl_abs_diff":0.0015185297597781755,"training/rollout_probs_diff_mean":0.002963797887787223,"critic/returns/mean":-0.0012260939693078399,"val-aux/sciknoweval/score/best@16/std":0.011791035148083585,"prompt_length/min":190,"val-aux/sciknoweval/reward/best@16/std":0.011791035148083585,"val-aux/sciknoweval/score/worst@16/mean":0.7182659574468084,"rollout_corr/log_ppl_diff_min":-0.004179961979389191,"val-aux/sciknoweval/reward/best@4/std":0.027759320424956824,"global_seqlen/balanced_max":23681,"timing_s/update_actor":23.16192770190537,"training/global_step":100,"self_distillation/token_reweight_w_mean":1.0763714532367885,"val-aux/sciknoweval/incorrect_format/maj@4/std":0,"self_distillation/empty_target_batch":0.3125,"val-core/sciknoweval/acc/best@16/std":0.011791035148083585,"critic/score/mean":0.67578125,"val-aux/sciknoweval/incorrect_format/best@16/mean":1,"_timestamp":1.7830844827750568e+09,"perf/max_memory_reserved_gb":140.86328125,"val-aux/sciknoweval/reward/maj@4/std":0.031638291528035434,"val-aux/sciknoweval/incorrect_format/std@16":0,"critic/rewards/max":1,"val-aux/sciknoweval/score/maj@2/mean":0.7651063829787235,"self_distillation/token_reweight_context_sample_fraction":0.6875,"val-aux/num_turns/min":2,"val-core/sciknoweval/acc/maj@16/std":0.013016076917955125,"global_seqlen/minmax_diff":7733,"response_length_non_aborted/mean":348.859375,"val-aux/sciknoweval/incorrect_format/maj@16/mean":1,"val-aux/sciknoweval/score/mean@16":0.7646276595744681,"rollout_corr/rollout_is_seq_max_deviation":0.005880475044250488,"global_seqlen/balanced_min":23678,"timing_s/agent_loop/generate_sequences/max":6.074218215420842,"rollout_corr/training_log_ppl":0.15226043093571207,"rollout_corr/rollout_is_ratio_fraction_low":0,"val-aux/sciknoweval/incorrect_format/best@2/mean":1,"timing_s/testing":40.37994212470949,"training/epoch":3,"self_distillation/success_group_fraction":0.6875,"actor/grad_norm":0.08581123873591423,"val-aux/sciknoweval/acc/std@16":0.0401737640202366,"val-aux/sciknoweval/incorrect_format/worst@16/std":0,"val-aux/sciknoweval/reward/maj@16/std":0.013016076917955125,"self_distillation/feedback_available_fraction":0,"perf/time_per_step":40.37044422328472,"val-aux/sciknoweval/incorrect_format/worst@16/mean":1,"rollout_corr/chi2_token":0.0010083636734634638,"val-aux/sciknoweval/incorrect_format/maj@4/mean":1,"val-aux/sciknoweval/score/best@2/std":0.03462769102759328,"timing_s/reward":0.06225111149251461,"val-aux/sciknoweval/score/best@8/std":0.019636408581214976,"rollout_corr/rollout_is_seq_std":0.0018322683172300458,"val-aux/sciknoweval/acc/worst@4/mean":0.7366170212765957,"rollout_corr/ppl_ratio":1.0004883385263383,"val-aux/sciknoweval/incorrect_format/best@8/mean":1,"val-aux/sciknoweval/incorrect_format/best@2/std":0,"val-aux/sciknoweval/acc/best@2/mean":0.7809255319148936,"rollout_corr/rollout_is_eff_sample_size":0.9991205327494357,"val-aux/sciknoweval/reward/worst@2/mean":0.7494468085106384,"rollout_corr/k3_kl":0.0005004820688583322,"val-aux/sciknoweval/acc/worst@16/std":0.01148574534900323,"rollout_corr/rollout_is_ratio_fraction_high":1.119720491260523e-05,"val-aux/sciknoweval/reward/best@2/std":0.03462769102759328,"val-aux/sciknoweval/reward/worst@8/mean":0.726340425531915,"val-aux/sciknoweval/score/worst@8/mean":0.726340425531915,"self_distillation/token_reweight_w_std":0.7491657065111212,"val-aux/sciknoweval/reward/worst@2/std":0.03367836377537821,"critic/advantages/mean":-0.0012260939693078399,"val-aux/sciknoweval/score/best@2/mean":0.7809255319148936,"val-aux/sciknoweval/score/worst@4/std":0.025172389625263896,"_runtime":4643.638902204,"rollout_corr/rollout_is_seq_max":1.0058804750442505,"response_length/max":607,"_wandb":{"runtime":4643},"perf/max_memory_allocated_gb":130.98332166671753,"val-aux/sciknoweval/reward/maj@16/mean":0.7633617021276595,"timing_s/start_profile":9.56319272518158e-05,"val-aux/sciknoweval/acc/worst@2/std":0.03367836377537821,"rollout_corr/rollout_is_min":0.567776620388031,"val-aux/sciknoweval/reward/worst@16/std":0.01148574534900323,"rollout_corr/rollout_is_seq_min":0.9950834512710571,"critic/advantages/max":0.125,"val-aux/sciknoweval/score/worst@4/mean":0.7366170212765957,"prompt_length/max":479,"val-aux/sciknoweval/score/worst@8/std":0.018341796156463988,"self_distillation/token_reweight_lambda":0.5,"val-aux/sciknoweval/incorrect_format/maj@8/std":0,"response_length/mean":348.859375,"val-aux/sciknoweval/acc/best@8/std":0.019636408581214976,"perf/cpu_memory_used_gb":196.12500381469727,"val-aux/sciknoweval/reward/best@8/mean":0.8061702127659575,"val-aux/sciknoweval/incorrect_format/best@16/std":0,"perf/mfu/actor":0.05333022759720618,"val-aux/sciknoweval/acc/best@8/mean":0.8061702127659575,"val-aux/sciknoweval/score/worst@16/std":0.01148574534900323,"prompt_length/clip_ratio":0,"actor/pg_loss":0.0005346338730305433,"timing_s/stop_profile":0.00013117678463459015,"self_distillation/token_reweight_active_token_fraction":0.6875,"timing_s/agent_loop/slowest/response_length":607,"val-aux/sciknoweval/reward/mean@16":0.7646276595744681,"self_distillation/token_reweight_w_clip_frac":0.016922423528740183,"global_seqlen/mean":23679.5,"val-aux/sciknoweval/incorrect_format/maj@2/mean":1,"val-aux/sciknoweval/incorrect_format/worst@8/mean":1,"rollout_corr/chi2_seq":0.34577386314049363,"val-aux/sciknoweval/reward/best@4/mean":0.7944468085106383,"val-aux/sciknoweval/acc/maj@2/std":0.040366939474519616,"val-aux/sciknoweval/acc/worst@8/mean":0.726340425531915,"val-core/sciknoweval/acc/maj@16/mean":0.7633617021276595,"val-aux/sciknoweval/reward/best@8/std":0.019636408581214976,"self_distillation/success_sample_fraction":0.6875,"timing_per_token_ms/update_actor":0.12226782502747825,"global_seqlen/min":20320,"timing_s/agent_loop/generate_sequences/min":2.0379578452557325,"rollout_corr/rollout_ppl":1.170548481401056,"perf/total_num_tokens":189436,"critic/returns/max":0.125,"actor/ppo_kl":0.00010043691768402141,"rollout_corr/rollout_is_max":2.0033769607543945,"val-aux/sciknoweval/acc/worst@16/mean":0.7182659574468084,"val-aux/sciknoweval/score/maj@2/std":0.040366939474519616,"val-aux/sciknoweval/acc/maj@4/std":0.031638291528035434,"val-aux/sciknoweval/reward/worst@16/mean":0.7182659574468084,"val-aux/sciknoweval/acc/maj@8/mean":0.764223404255319,"timing_s/agent_loop/slowest/tool_calls":0,"perf/mfu/actor_infer":0,"timing_s/agent_loop/tool_calls/max":0,"actor/pg_clipfrac":5.5688376050966326e-05,"_step":100,"rollout_corr/rollout_is_seq_fraction_high":0,"val-aux/sciknoweval/reward/worst@4/mean":0.7366170212765957,"response/aborted_ratio":0,"val-aux/sciknoweval/acc/maj@8/std":0.020384144360752216,"val-aux/sciknoweval/incorrect_format/maj@8/mean":1,"val-aux/sciknoweval/reward/worst@8/std":0.018341796156463988,"rollout_corr/rollout_is_seq_mean":1.0000578165054321,"training/rollout_probs_diff_max":0.25462448596954346,"critic/rewards/min":0,"training/rollout_probs_diff_std":0.010244213975965977,"response_length/min":161,"val-aux/sciknoweval/acc/best@4/std":0.027759320424956824,"actor/entropy":0.15241530537605286,"val-aux/sciknoweval/incorrect_format/best@4/std":0,"val-aux/sciknoweval/acc/maj@4/mean":0.764436170212766,"val-aux/sciknoweval/incorrect_format/worst@2/std":0,"val-aux/sciknoweval/incorrect_format/maj@2/std":0,"response_length_non_aborted/min":161,"val-aux/sciknoweval/incorrect_format/best@4/mean":1,"num_turns/mean":2,"val-aux/sciknoweval/score/maj@8/mean":0.764223404255319,"response_length_non_aborted/max":607,"val-aux/sciknoweval/reward/maj@4/mean":0.764436170212766,"val-aux/sciknoweval/reward/maj@8/mean":0.764223404255319,"critic/score/max":1,"timing_s/adv":0.5430567041039467,"val-aux/sciknoweval/incorrect_format/mean@16":1,"rollout_corr/training_ppl":1.1711667915806174,"perf/throughput":586.5553489833095,"timing_s/gen":13.520746255293489,"val-aux/sciknoweval/reward/maj@2/std":0.040366939474519616,"val-aux/num_turns/mean":2,"actor/pg_clipfrac_lower":0,"val-aux/sciknoweval/reward/best@16/mean":0.813808510638298,"val-aux/sciknoweval/incorrect_format/worst@8/std":0,"val-aux/sciknoweval/reward/worst@4/std":0.025172389625263896,"self_distillation/correct_sample_fraction":0.67578125,"val-aux/sciknoweval/reward/maj@2/mean":0.7651063829787235,"training/rollout_probs_diff_valid":1,"timing_s/old_log_prob":2.993232659995556,"val-aux/sciknoweval/acc/maj@2/mean":0.7651063829787235,"timing_s/agent_loop/tool_calls/mean":0,"training/rollout_actor_probs_pearson_corr":0.9986335039138794,"rollout_corr/rollout_is_seq_fraction_low":0,"val-aux/sciknoweval/score/maj@8/std":0.020384144360752216,"actor/kl_loss":0,"num_turns/max":2,"val-aux/sciknoweval/score/worst@2/std":0.03367836377537821,"val-aux/sciknoweval/incorrect_format/worst@2/mean":1,"timing_s/agent_loop/slowest/prompt_length":414,"num_turns/min":2,"timing_per_token_ms/gen":0.15139456997462142,"val-aux/sciknoweval/incorrect_format/maj@16/std":0,"actor/lr":1e-06,"prompt_length/mean":391.125,"timing_per_token_ms/adv":0.0028667027603198267,"val-aux/sciknoweval/reward/std@16":0.0401737640202366,"val-aux/sciknoweval/score/best@8/mean":0.8061702127659575,"timing_s/step":40.37044422328472,"self_distillation/token_reweight_correct_sample_fraction":0.67578125,"val-aux/sciknoweval/score/std@16":0.0401737640202366,"rollout_corr/rollout_is_mean":1.0000261068344116,"timing_s/agent_loop/generate_sequences/mean":4.135219432209851,"val-aux/sciknoweval/acc/worst@4/std":0.025172389625263896,"rollout_corr/kl":0.00048633673415032774,"critic/returns/min":-0.875}