1 line
11 KiB
JSON
1 line
11 KiB
JSON
{"val-aux/tooluse/score/worst@8/std":0.04295702328078124,"rollout_corr/rollout_log_ppl":0.13506622198246987,"global_seqlen/max":26968,"val-aux/tooluse/incorrect_format/best@16/mean":0,"_wandb":{"runtime":3606},"val-aux/tooluse/acc/best@4/mean":0.6149852941176471,"val-aux/tooluse/score/best@4/std":0.008533867314311838,"rollout_corr/rollout_is_seq_max_deviation":0.03070974349975586,"training/rollout_probs_diff_mean":0.0043131280690431595,"val-aux/tooluse/incorrect_format/maj@16/mean":0,"timing_s/gen":8.898955885320902,"prompt_length/clip_ratio":0,"val-aux/tooluse/acc/std@16":0.0397091660129265,"critic/score/min":0,"self_distillation/empty_target_batch":0.0078125,"val-aux/tooluse/reward/best@16/mean":0.6176323529411765,"self_distillation/reprompt_sample_fraction":0.9921875,"num_turns/mean":2,"rollout_corr/rollout_is_seq_max":1.0307097434997559,"val-core/tooluse/acc/best@16/std":0.00046480825380232674,"self_distillation/token_reweight_eps_w":0.20000000000000004,"val-aux/tooluse/reward/worst@4/std":0.04688580916450452,"val-aux/num_turns/mean":2,"response_length_non_aborted/clip_ratio":0,"val-aux/tooluse/score/maj@8/std":0.016895760456278525,"val-aux/num_turns/min":2,"perf/time_per_step":27.41770182736218,"rollout_corr/rollout_is_ratio_fraction_high":8.322236681124195e-05,"val-aux/tooluse/reward/best@4/std":0.008533867314311838,"val-aux/tooluse/acc/worst@16/std":0.03331555519855917,"val-aux/tooluse/reward/best@2/std":0.021252889471013774,"val-aux/tooluse/acc/worst@2/mean":0.5792499999999999,"val-aux/tooluse/score/maj@4/std":0.028165057960484432,"critic/returns/mean":-0.008946404792368412,"val-aux/tooluse/score/std@16":0.0397091660129265,"critic/rewards/mean":0.43359375,"global_seqlen/balanced_min":21525,"val-aux/tooluse/acc/worst@16/mean":0.5220294117647059,"val-aux/tooluse/acc/best@4/std":0.008533867314311838,"self_distillation/token_reweight_w_clipped_mean":1.0005437803920358,"val-aux/tooluse/incorrect_format/worst@8/mean":0,"val-aux/tooluse/acc/maj@4/std":0.028165057960484432,"timing_s/testing":21.686641938984394,"training/rollout_probs_diff_max":0.41022178530693054,"training/rollout_probs_diff_std":0.013580762781202793,"val-aux/tooluse/reward/mean@16":0.59375,"val-core/tooluse/acc/maj@16/std":0.010892157021494882,"actor/grad_norm":0.2933488469570875,"perf/throughput":785.1861594947975,"val-aux/tooluse/score/worst@2/mean":0.5792499999999999,"rollout_corr/log_ppl_diff_min":-0.01413802057504654,"rollout_corr/rollout_is_max":2.820589303970337,"rollout_corr/log_ppl_abs_diff":0.0032595000756145964,"critic/rewards/max":1,"val-aux/tooluse/acc/maj@8/std":0.016895760456278525,"val-aux/tooluse/incorrect_format/worst@16/std":0,"val-aux/tooluse/reward/maj@8/mean":0.6039705882352941,"timing_s/agent_loop/generate_sequences/min":0.5163722541183233,"val-core/tooluse/acc/best@16/mean":0.6176323529411765,"rollout_corr/chi2_token":0.0012444618623703718,"val-aux/tooluse/incorrect_format/std@16":0,"rollout_corr/rollout_is_seq_fraction_high":0,"prompt_length/max":934,"val-aux/tooluse/incorrect_format/worst@8/std":0,"rollout_corr/training_ppl":1.150276112370193,"rollout_corr/rollout_is_mean":1.0004314184188843,"val-aux/tooluse/acc/worst@4/mean":0.5617647058823529,"timing_per_token_ms/update_actor":0.09024581746989048,"timing_s/agent_loop/tool_calls/mean":0,"critic/advantages/mean":-0.008946404792368412,"val-aux/tooluse/acc/maj@2/mean":0.5937794117647059,"critic/returns/min":-0.875,"timing_s/step":27.41770182736218,"val-aux/tooluse/score/maj@2/mean":0.5937794117647059,"response/aborted_ratio":0,"timing_s/reward":0.0715668685734272,"prompt_length/mean":578.875,"rollout_corr/rollout_is_std":0.03966965898871422,"val-aux/tooluse/incorrect_format/best@8/std":0,"timing_s/agent_loop/tool_calls/max":0,"response_length_non_aborted/mean":93.875,"val-aux/tooluse/score/worst@4/mean":0.5617647058823529,"val-aux/tooluse/incorrect_format/maj@4/mean":0,"val-aux/tooluse/incorrect_format/best@2/std":0,"rollout_corr/rollout_is_min":0.5896446108818054,"_runtime":3606.405012169,"val-aux/tooluse/reward/maj@2/std":0.04051488817390775,"self_distillation/success_sample_fraction":0.4921875,"val-aux/tooluse/incorrect_format/maj@8/mean":0,"timing_s/agent_loop/tool_calls/min":0,"val-aux/tooluse/reward/worst@2/mean":0.5792499999999999,"val-aux/tooluse/score/maj@8/mean":0.6039705882352941,"val-aux/tooluse/acc/best@8/std":0.0024199717661105916,"rollout_corr/rollout_is_ratio_fraction_low":0,"val-aux/tooluse/reward/worst@16/mean":0.5220294117647059,"timing_per_token_ms/adv":0.0034544185582314423,"response_length/mean":93.875,"actor/pg_loss":0.0003957020817324519,"critic/score/mean":0.43359375,"timing_s/save_checkpoint":6.582349684089422,"prompt_length/min":251,"timing_s/agent_loop/slowest/tool_calls":0,"response_length/clip_ratio":0,"perf/mfu/actor_infer":0,"training/rollout_actor_probs_pearson_corr":0.9975739121437073,"val-aux/tooluse/score/mean@16":0.59375,"timing_s/start_profile":0.00013494491577148438,"rollout_corr/k3_kl":0.0007612622766463763,"val-aux/tooluse/score/best@16/std":0.00046480825380232674,"global_seqlen/min":18236,"val-aux/tooluse/reward/std@16":0.0397091660129265,"self_distillation/success_group_fraction":0.5,"actor/pg_clipfrac":0.0003417537127461401,"val-aux/tooluse/incorrect_format/mean@16":0,"self_distillation/correct_sample_fraction":0.43359375,"val-aux/tooluse/reward/worst@4/mean":0.5617647058823529,"val-aux/tooluse/score/best@16/mean":0.6176323529411765,"val-aux/tooluse/reward/worst@16/std":0.03331555519855917,"timing_per_token_ms/gen":0.37029610042114275,"actor/kl_loss":0,"timing_s/agent_loop/slowest/generate_sequences":2.9513198621571064,"val-core/tooluse/acc/maj@16/mean":0.6051029411764706,"timing_s/agent_loop/slowest/response_length":450,"val-aux/tooluse/incorrect_format/maj@2/mean":0,"val-aux/tooluse/reward/maj@4/mean":0.6007058823529412,"training/epoch":0,"rollout_corr/ppl_ratio":1.000922082690522,"rollout_corr/training_log_ppl":0.13597888073957165,"val-aux/tooluse/reward/maj@4/std":0.028165057960484432,"val-aux/tooluse/incorrect_format/maj@4/std":0,"val-aux/tooluse/acc/best@8/mean":0.6173676470588235,"val-aux/tooluse/incorrect_format/worst@4/std":0,"val-aux/tooluse/incorrect_format/best@4/mean":0,"num_turns/min":2,"val-aux/tooluse/score/best@8/std":0.0024199717661105916,"rollout_corr/rollout_ppl":1.1492430758662522,"val-aux/tooluse/reward/maj@16/std":0.010892157021494882,"rollout_corr/chi2_seq":0.2823219452984631,"val-aux/tooluse/incorrect_format/maj@8/std":0,"val-aux/tooluse/incorrect_format/best@8/mean":0,"rollout_corr/rollout_is_seq_mean":1.0004057884216309,"val-aux/tooluse/acc/worst@2/std":0.04534421397949325,"response_length_non_aborted/min":38,"self_distillation/token_reweight_w_clip_frac":0.0204948254104238,"val-aux/tooluse/score/worst@16/std":0.03331555519855917,"val-aux/tooluse/acc/worst@4/std":0.04688580916450452,"timing_s/stop_profile":0.00013438984751701355,"val-aux/tooluse/score/worst@8/mean":0.5418382352941178,"critic/advantages/min":-0.875,"_timestamp":1.783010844235158e+09,"val-aux/tooluse/score/worst@4/std":0.04688580916450452,"val-aux/tooluse/acc/maj@8/mean":0.6039705882352941,"val-aux/tooluse/incorrect_format/maj@16/std":0,"val-aux/tooluse/score/best@4/mean":0.6149852941176471,"timing_s/adv":0.5949337817728519,"val-aux/tooluse/score/maj@4/mean":0.6007058823529412,"response_length/min":38,"actor/pg_clipfrac_lower":0,"val-aux/tooluse/score/maj@2/std":0.04051488817390775,"val-aux/tooluse/reward/best@2/mean":0.6084705882352941,"self_distillation/token_reweight_correct_sample_fraction":0.43359375,"rollout_corr/rollout_is_seq_fraction_low":0,"timing_s/agent_loop/slowest/prompt_length":565,"val-aux/tooluse/score/worst@2/std":0.04534421397949325,"val-aux/tooluse/score/best@2/mean":0.6084705882352941,"val-aux/tooluse/acc/worst@8/mean":0.5418382352941178,"perf/max_memory_allocated_gb":125.62322854995728,"val-aux/tooluse/score/best@8/mean":0.6173676470588235,"critic/rewards/min":0,"rollout_corr/kl":0.0009126573687510131,"training/global_step":100,"self_distillation/feedback_available_fraction":0.56640625,"val-aux/tooluse/acc/best@2/std":0.021252889471013774,"self_distillation/token_reweight_lambda":0.5,"global_seqlen/mean":21528,"val-aux/tooluse/score/maj@16/mean":0.6051029411764706,"timing_s/agent_loop/generate_sequences/max":2.9513198621571064,"actor/ppo_kl":0.0005588802749123545,"critic/returns/max":0.875,"global_seqlen/balanced_max":21530,"rollout_corr/log_ppl_diff":0.0009126587571017808,"timing_s/agent_loop/generate_sequences/mean":1.0657720906820032,"response_length/max":450,"val-aux/tooluse/incorrect_format/best@4/std":0,"critic/advantages/max":0.875,"val-aux/tooluse/score/best@2/std":0.021252889471013774,"self_distillation/token_reweight_active_token_fraction":0.9921875,"timing_s/old_log_prob":2.2224635891616344,"val-aux/tooluse/score/worst@16/mean":0.5220294117647059,"self_distillation/feedback_used_fraction":0.5,"critic/score/max":1,"val-aux/tooluse/score/maj@16/std":0.010892157021494882,"training/rollout_probs_diff_valid":1,"val-aux/tooluse/acc/worst@8/std":0.04295702328078124,"val-aux/tooluse/reward/best@8/std":0.0024199717661105916,"num_turns/max":2,"val-aux/tooluse/acc/best@2/mean":0.6084705882352941,"val-aux/tooluse/reward/worst@8/std":0.04295702328078124,"actor/lr":1e-06,"timing_s/update_actor":15.542495667934418,"self_distillation/token_reweight_w_std":3.821444055784081e+06,"val-aux/tooluse/reward/best@16/std":0.00046480825380232674,"_step":100,"rollout_corr/log_ppl_diff_max":0.014425680041313171,"val-core/tooluse/acc/mean@16":0.59375,"val-aux/tooluse/incorrect_format/best@16/std":0,"rollout_corr/rollout_is_eff_sample_size":0.9984299791032849,"perf/cpu_memory_used_gb":146.36181640625,"val-aux/tooluse/incorrect_format/worst@2/mean":0,"perf/max_memory_reserved_gb":132.13671875,"val-aux/tooluse/incorrect_format/worst@2/std":0,"val-aux/tooluse/reward/worst@8/mean":0.5418382352941178,"val-aux/tooluse/reward/maj@16/mean":0.6051029411764706,"self_distillation/token_reweight_w_mean":532096.5092833575,"val-aux/tooluse/reward/maj@2/mean":0.5937794117647059,"val-aux/tooluse/reward/maj@8/std":0.016895760456278525,"val-aux/tooluse/reward/worst@2/std":0.04534421397949325,"self_distillation/token_reweight_context_sample_fraction":0.9921875,"actor/entropy":0.15581010282039642,"perf/total_num_tokens":172224,"rollout_corr/rollout_is_seq_std":0.004751189611852169,"val-aux/tooluse/incorrect_format/maj@2/std":0,"response_length_non_aborted/max":450,"val-aux/tooluse/reward/best@8/mean":0.6173676470588235,"val-aux/tooluse/incorrect_format/best@2/mean":0,"val-aux/num_turns/max":2,"val-aux/tooluse/incorrect_format/worst@16/mean":0,"val-aux/tooluse/incorrect_format/worst@4/mean":0,"global_seqlen/minmax_diff":8732,"val-aux/tooluse/reward/best@4/mean":0.6149852941176471,"val-aux/tooluse/acc/maj@2/std":0.04051488817390775,"perf/mfu/actor":0.03961015113542468,"rollout_corr/rollout_is_seq_min":0.9896077513694763,"val-aux/tooluse/acc/maj@4/mean":0.6007058823529412} |