1 line
9.8 KiB
JSON
1 line
9.8 KiB
JSON
|
|
{"timing_s/agent_loop/tool_calls/min":0,"timing_s/agent_loop/tool_calls/mean":0,"actor/kl_loss":0,"val-aux/tooluse/reward/worst@2/mean":0.5376176470588235,"timing_s/adv":0.1153049636632204,"val-aux/tooluse/score/std@16":0.07216799159273436,"val-aux/tooluse/score/maj@2/std":0.07217664475658733,"val-aux/tooluse/incorrect_format/worst@8/std":0,"val-aux/tooluse/acc/worst@2/std":0.065873083983826,"timing_s/save_checkpoint":6.70261013135314,"val-aux/tooluse/incorrect_format/maj@16/std":0,"rollout_corr/rollout_is_min":0.6138291954994202,"critic/returns/max":0.875,"critic/score/max":1,"critic/score/mean":0.19140625,"val-aux/tooluse/reward/mean@16":0.5689338235294118,"response_length/mean":78.55859375,"critic/rewards/min":0,"val-aux/tooluse/acc/std@16":0.07216799159273436,"timing_s/reward":0.07088929787278175,"val-aux/tooluse/acc/maj@2/std":0.07217664475658733,"rollout_corr/rollout_is_seq_fraction_high":0,"timing_s/step":20.5681207254529,"val-core/tooluse/acc/mean@16":0.5689338235294118,"critic/score/min":0,"val-aux/tooluse/incorrect_format/best@8/mean":0,"timing_s/agent_loop/generate_sequences/max":2.4420138355344534,"training/rollout_probs_diff_max":0.12089437246322632,"val-aux/tooluse/reward/maj@4/std":0.06112092824371511,"_step":100,"timing_s/testing":19.794103000313044,"val-aux/tooluse/acc/best@4/mean":0.6216029411764705,"num_turns/min":2,"val-aux/tooluse/acc/worst@8/std":0.03169698340721522,"actor/pg_clipfrac_lower":0,"val-aux/tooluse/reward/maj@4/mean":0.571764705882353,"rollout_corr/rollout_is_mean":0.9998654127120972,"val-aux/tooluse/score/maj@8/mean":0.5750294117647058,"val-aux/tooluse/score/worst@16/mean":0.47960294117647057,"timing_per_token_ms/update_actor":0.058253035554515414,"perf/max_memory_allocated_gb":124.25863409042358,"val-aux/tooluse/score/worst@4/mean":0.5104117647058823,"timing_per_token_ms/adv":0.0006851685731115321,"global_seqlen/min":18523,"response_length/max":403,"val-core/tooluse/acc/maj@16/mean":0.5770294117647059,"val-aux/tooluse/score/best@16/std":0.011314008550944105,"val-aux/tooluse/acc/best@4/std":0.03612363935746805,"rollout_corr/rollout_is_seq_max_deviation":0.01654648780822754,"rollout_corr/rollout_is_ratio_fraction_low":0,"val-aux/tooluse/score/best@16/mean":0.6409558823529412,"val-aux/tooluse/score/maj@8/std":0.04699356851985658,"val-aux/tooluse/reward/worst@2/std":0.065873083983826,"perf/max_memory_reserved_gb":130.046875,"training/rollout_probs_diff_mean":0.002164518227800727,"timing_s/agent_loop/tool_calls/max":0,"val-aux/tooluse/acc/maj@4/std":0.06112092824371511,"training/rollout_actor_probs_pearson_corr":0.9992616772651672,"val-aux/tooluse/score/maj@4/std":0.06112092824371511,"val-aux/tooluse/incorrect_format/maj@2/std":0,"training/rollout_probs_diff_valid":1,"perf/time_per_step":20.5681207254529,"rollout_corr/chi2_token":0.0007952849846333265,"val-aux/tooluse/reward/worst@4/mean":0.5104117647058823,"timing_s/start_profile":0.00012426264584064484,"perf/mfu/actor_infer":0,"val-aux/tooluse/incorrect_format/mean@16":0,"timing_s/agent_loop/generate_sequences/min":0.25598333589732647,"val-aux/tooluse/score/best@8/std":0.021009683207312644,"val-aux/tooluse/score/best@8/mean":0.6337205882352942,"timing_per_token_ms/gen":0.40311589798174624,"rollout_corr/rollout_is_seq_fraction_low":0,"val-aux/tooluse/score/maj@4/mean":0.571764705882353,"val-aux/tooluse/acc/best@2/std":0.05681896001528811,"val-aux/tooluse/score/best@4/std":0.03612363935746805,"val-aux/tooluse/reward/maj@8/mean":0.5750294117647058,"global_seqlen/mean":21035.875,"prompt_length/mean":578.8125,"val-aux/tooluse/reward/maj@8/std":0.04699356851985658,"val-aux/tooluse/incorrect_format/std@16":0,"val-aux/tooluse/reward/best@2/mean":0.599735294117647,"val-aux/tooluse/reward/worst@8/mean":0.4913382352941176,"val-aux/tooluse/incorrect_format/worst@16/mean":0,"val-aux/tooluse/acc/worst@16/std":0.018056061970488427,"critic/advantages/mean":0.0050159115344285965,"val-aux/tooluse/reward/maj@2/std":0.07217664475658733,"val-aux/tooluse/incorrect_format/best@16/std":0,"val-aux/tooluse/incorrect_format/maj@8/std":0,"val-aux/tooluse/reward/bes
|