1 line
11 KiB
JSON
1 line
11 KiB
JSON
{"_runtime":5136.926281788,"val-aux/sciknoweval/acc/std@16":0.16309517319018776,"val-aux/sciknoweval/reward/worst@4/mean":0.6136428571428572,"val-aux/sciknoweval/acc/worst@8/mean":0.5821809523809525,"critic/rewards/min":0,"val-aux/sciknoweval/acc/worst@8/std":0.04669533954140184,"val-aux/sciknoweval/incorrect_format/best@2/std":0,"val-aux/sciknoweval/incorrect_format/maj@2/std":0,"rollout_corr/rollout_is_seq_max":1.0135035514831543,"val-aux/sciknoweval/reward/worst@16/mean":0.5673476190476191,"val-aux/sciknoweval/score/maj@2/std":0.16273431336078428,"val-aux/sciknoweval/incorrect_format/best@2/mean":1,"timing_s/start_profile":7.35800713300705e-05,"val-aux/sciknoweval/acc/worst@16/mean":0.5673476190476191,"global_seqlen/max":18719,"timing_s/agent_loop/slowest/tool_calls":0,"perf/mfu/actor_infer":0,"val-aux/sciknoweval/score/best@8/mean":0.9082190476190476,"timing_s/agent_loop/generate_sequences/max":6.079234275966883,"val-aux/sciknoweval/acc/maj@2/std":0.16273431336078428,"val-aux/sciknoweval/score/best@16/mean":0.9349142857142857,"self_distillation/dynamic_weighting_enabled":1,"self_distillation/success_group_fraction":0.9375,"rollout_corr/log_ppl_abs_diff":0.001808215893106535,"rollout_corr/rollout_is_mean":1.0000580549240112,"response_length/min":101,"rollout_corr/rollout_is_seq_mean":1.0000791549682617,"val-aux/num_turns/max":2,"training/rollout_probs_diff_valid":1,"num_turns/mean":2,"response_length_non_aborted/clip_ratio":0,"timing_s/gen":12.888554267585278,"rollout_corr/rollout_is_eff_sample_size":0.9991349319207025,"rollout_corr/chi2_token":0.001111235935240984,"val-aux/sciknoweval/score/maj@16/std":0.08119347263943295,"rollout_corr/kl":0.00039874245554472054,"val-core/sciknoweval/acc/maj@16/std":0.08119347263943295,"val-aux/sciknoweval/score/best@4/mean":0.8652047619047618,"srpo/routed_token_count":220.96875,"val-aux/sciknoweval/incorrect_format/best@4/mean":1,"val-aux/sciknoweval/score/worst@4/std":0.08715504524810262,"val-aux/sciknoweval/incorrect_format/best@4/std":0,"prompt_length/clip_ratio":0,"val-aux/sciknoweval/reward/maj@4/std":0.13821444887569323,"self_distillation/empty_target_batch":0.0703125,"critic/returns/mean":-0.006017094478011131,"rollout_corr/log_ppl_diff":0.0003987445088569075,"timing_s/adv":0.4795156195759773,"val-aux/sciknoweval/incorrect_format/maj@16/std":0,"val-aux/sciknoweval/reward/best@8/mean":0.9082190476190476,"timing_s/agent_loop/generate_sequences/mean":2.3858929471098236,"val-aux/sciknoweval/incorrect_format/worst@2/std":0,"critic/rewards/max":1,"val-aux/sciknoweval/reward/worst@16/std":0.02126535217732578,"srpo/grpo_actor/pg_clipfrac_lower":0,"timing_per_token_ms/adv":0.00362215690397614,"prompt_length/mean":296.15625,"rollout_corr/ppl_ratio":1.0004019325133413,"val-core/sciknoweval/acc/best@16/mean":0.9349142857142857,"rollout_corr/k3_kl":0.0004779839419200016,"rollout_corr/rollout_is_ratio_fraction_low":1.767783942341339e-05,"val-aux/num_turns/min":2,"timing_s/agent_loop/slowest/generate_sequences":6.079234275966883,"srpo/grpo_actor/ppo_kl":0,"val-aux/sciknoweval/score/best@2/mean":0.8065190476190476,"timing_s/save_checkpoint":12.84816437587142,"val-aux/sciknoweval/incorrect_format/worst@16/std":0,"srpo/sdpo_loss":0.0033741780134732835,"val-aux/sciknoweval/reward/maj@2/mean":0.7362714285714287,"rollout_corr/log_ppl_diff_min":-0.0124589204788208,"val-aux/sciknoweval/acc/worst@4/mean":0.6136428571428572,"val-aux/sciknoweval/incorrect_format/worst@16/mean":1,"rollout_corr/rollout_is_std":0.029428821057081223,"critic/advantages/max":0.875,"val-aux/sciknoweval/reward/worst@2/mean":0.6664857142857143,"val-aux/sciknoweval/acc/best@8/mean":0.9082190476190476,"val-aux/sciknoweval/score/maj@16/mean":0.7606666666666667,"val-aux/sciknoweval/score/worst@8/mean":0.5821809523809525,"self_distillation/feedback_available_fraction":0,"val-aux/sciknoweval/reward/maj@2/std":0.16273431336078428,"val-aux/sciknoweval/acc/best@4/std":0.11009573982420097,"srpo/correct_sample_fraction":0.75390625,"val-aux/sciknoweval/acc/best@2/std":0.14511294435717983,"val-aux/sciknoweval/score/worst@16/std":0.02126535217732578,"val-aux/sciknoweval/incorrect_format/worst@8/mean":1,"actor/lr":5e-06,"val-aux/sciknoweval/incorrect_format/worst@8/std":0,"val-aux/sciknoweval/score/mean@16":0.7363095238095239,"val-aux/sciknoweval/score/worst@8/std":0.04669533954140184,"rollout_corr/rollout_is_seq_min":0.9921493530273438,"val-aux/sciknoweval/reward/worst@2/std":0.1327243786388664,"training/rollout_actor_probs_pearson_corr":0.9994076490402222,"prompt_length/min":165,"num_turns/max":2,"val-aux/sciknoweval/incorrect_format/maj@4/mean":1,"srpo/routed_loss":-0.08303024125780212,"val-aux/sciknoweval/reward/best@16/std":0.03906584209067313,"self_distillation/success_sample_fraction":0.9296875,"rollout_corr/rollout_ppl":1.4317505126819015,"timing_s/stop_profile":0.00017225928604602814,"critic/returns/max":0.875,"val-aux/sciknoweval/acc/maj@4/std":0.13821444887569323,"rollout_corr/rollout_is_seq_std":0.0024739254731684923,"val-aux/sciknoweval/reward/maj@4/mean":0.7460904761904762,"val-aux/sciknoweval/acc/maj@8/mean":0.753152380952381,"val-aux/sciknoweval/acc/best@2/mean":0.8065190476190476,"response_length/clip_ratio":0,"critic/rewards/mean":0.75390625,"val-aux/sciknoweval/incorrect_format/maj@8/std":0,"val-aux/sciknoweval/score/worst@16/mean":0.5673476190476191,"self_distillation/correct_sample_fraction":0.75390625,"rollout_corr/rollout_is_seq_fraction_high":0,"critic/score/max":1,"val-aux/sciknoweval/reward/best@2/std":0.14511294435717983,"timing_s/old_log_prob":2.9018276669085026,"critic/returns/min":-0.875,"srpo/grpo_actor/pg_clipfrac":0,"_step":100,"perf/time_per_step":37.528549786657095,"actor/entropy":0.3308325409889221,"response_length_non_aborted/mean":220.96875,"rollout_corr/training_log_ppl":0.35083831640076824,"val-aux/sciknoweval/acc/best@4/mean":0.8652047619047618,"timing_s/agent_loop/tool_calls/max":0,"training/global_step":100,"val-aux/sciknoweval/incorrect_format/maj@2/mean":1,"val-aux/sciknoweval/score/std@16":0.16309517319018776,"val-aux/sciknoweval/reward/std@16":0.16309517319018776,"timing_s/reward":0.05373174883425236,"val-aux/sciknoweval/acc/maj@2/mean":0.7362714285714287,"val-aux/sciknoweval/reward/best@4/std":0.11009573982420097,"response_length_non_aborted/max":767,"val-aux/sciknoweval/score/worst@4/mean":0.6136428571428572,"critic/advantages/mean":-0.006017094478011131,"val-core/sciknoweval/acc/maj@16/mean":0.7606666666666667,"prompt_length/max":625,"val-aux/sciknoweval/incorrect_format/best@8/mean":1,"val-aux/sciknoweval/acc/best@8/std":0.0701292672277101,"actor/kl_loss":0,"_timestamp":1.783031839915186e+09,"perf/cpu_memory_used_gb":198.76100540161133,"val-aux/sciknoweval/score/maj@4/mean":0.7460904761904762,"global_seqlen/min":14190,"val-aux/sciknoweval/reward/maj@16/mean":0.7606666666666667,"timing_per_token_ms/gen":0.22784178807073394,"timing_s/agent_loop/generate_sequences/min":0.81765091791749,"val-aux/sciknoweval/score/best@2/std":0.14511294435717983,"val-aux/sciknoweval/score/maj@8/mean":0.753152380952381,"self_distillation/feedback_used_fraction":0,"global_seqlen/mean":16548,"rollout_corr/rollout_is_min":0.49668699502944946,"val-core/sciknoweval/acc/best@16/std":0.03906584209067313,"val-aux/sciknoweval/score/maj@4/std":0.13821444887569323,"val-aux/sciknoweval/acc/maj@8/std":0.10965244514560699,"training/rollout_probs_diff_max":0.17977115511894226,"val-aux/sciknoweval/reward/worst@4/std":0.08715504524810262,"val-aux/sciknoweval/incorrect_format/worst@4/std":0,"val-aux/sciknoweval/reward/mean@16":0.7363095238095239,"training/rollout_probs_diff_std":0.008875560946762562,"rollout_corr/log_ppl_diff_max":0.0092182457447052,"val-aux/sciknoweval/incorrect_format/maj@8/mean":1,"val-aux/sciknoweval/incorrect_format/maj@4/std":0,"perf/mfu/actor":0.04076109142587246,"rollout_corr/rollout_is_seq_fraction_low":0,"response_length_non_aborted/min":101,"timing_s/agent_loop/tool_calls/min":0,"perf/total_num_tokens":132384,"srpo/grpo_sample_fraction":0.81640625,"rollout_corr/rollout_log_ppl":0.35043957189191133,"val-aux/sciknoweval/incorrect_format/maj@16/mean":1,"val-core/sciknoweval/acc/mean@16":0.7363095238095239,"val-aux/sciknoweval/score/best@16/std":0.03906584209067313,"srpo/sdpo_sample_fraction":0.18359375,"timing_s/step":37.528549786657095,"val-aux/sciknoweval/reward/best@2/mean":0.8065190476190476,"rollout_corr/chi2_seq":0.3724006221164018,"timing_s/update_actor":21.113714333623648,"val-aux/sciknoweval/incorrect_format/worst@4/mean":1,"critic/advantages/min":-0.875,"self_distillation/dynamic_weight_max":1.6724038124084473,"response_length/mean":220.96875,"actor/grad_norm":0.10703474283218384,"val-aux/sciknoweval/incorrect_format/best@16/std":0,"val-aux/sciknoweval/score/maj@2/mean":0.7362714285714287,"val-aux/sciknoweval/score/best@4/std":0.11009573982420097,"global_seqlen/minmax_diff":4529,"val-aux/sciknoweval/acc/worst@2/std":0.1327243786388664,"val-aux/sciknoweval/acc/worst@4/std":0.08715504524810262,"val-aux/sciknoweval/reward/maj@8/std":0.10965244514560699,"rollout_corr/rollout_is_max":1.4527522325515747,"self_distillation/dynamic_weight_entropy_mean":0.06372180819744244,"val-aux/sciknoweval/score/worst@2/std":0.1327243786388664,"val-aux/sciknoweval/reward/worst@8/mean":0.5821809523809525,"rollout_corr/training_ppl":1.432360712904483,"val-aux/sciknoweval/acc/worst@16/std":0.02126535217732578,"val-aux/num_turns/mean":2,"training/epoch":1,"perf/max_memory_allocated_gb":149.75289154052734,"val-aux/sciknoweval/incorrect_format/mean@16":1,"response_length/max":767,"rollout_corr/rollout_is_ratio_fraction_high":0,"timing_per_token_ms/update_actor":0.1594884150170991,"self_distillation/dynamic_weight_mean":0.18359375512227416,"critic/score/min":0,"val-aux/sciknoweval/incorrect_format/worst@2/mean":1,"perf/throughput":440.9442969172094,"val-aux/sciknoweval/score/best@8/std":0.0701292672277101,"_wandb":{"runtime":5136},"timing_s/agent_loop/slowest/response_length":767,"val-aux/sciknoweval/reward/worst@8/std":0.04669533954140184,"perf/max_memory_reserved_gb":160.31640625,"self_distillation/reprompt_sample_fraction":0.9296875,"val-aux/sciknoweval/acc/maj@4/mean":0.7460904761904762,"critic/score/mean":0.75390625,"val-aux/sciknoweval/acc/worst@2/mean":0.6664857142857143,"val-aux/sciknoweval/incorrect_format/best@16/mean":1,"srpo/grpo_loss":-0.0864044192712754,"timing_s/agent_loop/slowest/prompt_length":190,"global_seqlen/balanced_min":16545,"val-aux/sciknoweval/reward/best@8/std":0.0701292672277101,"num_turns/min":2,"val-aux/sciknoweval/score/worst@2/mean":0.6664857142857143,"val-aux/sciknoweval/score/maj@8/std":0.10965244514560699,"val-aux/sciknoweval/incorrect_format/std@16":0,"val-aux/sciknoweval/reward/best@16/mean":0.9349142857142857,"val-aux/sciknoweval/incorrect_format/best@8/std":0,"timing_s/testing":63.14465852826834,"global_seqlen/balanced_max":16549,"training/rollout_probs_diff_mean":0.003615855472162366,"val-aux/sciknoweval/reward/maj@8/mean":0.753152380952381,"val-aux/sciknoweval/reward/maj@16/std":0.08119347263943295,"actor/pg_loss":-0.08303024125780212,"rollout_corr/rollout_is_seq_max_deviation":0.013503551483154297,"response/aborted_ratio":0,"timing_s/agent_loop/tool_calls/mean":0,"val-aux/sciknoweval/reward/best@4/mean":0.8652047619047618} |