1 line
11 KiB
JSON
1 line
11 KiB
JSON
{"global_seqlen/balanced_min":20404,"perf/mfu/actor":0.05015731000074418,"rollout_corr/rollout_is_max":1.4396264553070068,"critic/returns/max":0.875,"val-aux/sciknoweval/score/worst@4/std":0.06781975137735766,"val-aux/sciknoweval/score/worst@16/std":0.02453756899133995,"critic/rewards/max":1,"_wandb":{"runtime":5598},"critic/returns/min":-0.875,"timing_s/agent_loop/tool_calls/mean":0,"prompt_length/clip_ratio":0,"val-core/sciknoweval/acc/best@16/std":0.02643611186706914,"val-aux/sciknoweval/reward/worst@4/std":0.06781975137735766,"val-aux/sciknoweval/acc/maj@8/std":0.07759319570379096,"rollout_corr/rollout_is_seq_max":1.005556344985962,"val-aux/sciknoweval/score/maj@16/mean":0.7659095238095238,"val-aux/sciknoweval/incorrect_format/best@2/std":0.0005486204557414609,"self_distillation/correct_sample_fraction":0.74609375,"val-aux/sciknoweval/acc/worst@4/mean":0.6645857142857142,"val-aux/sciknoweval/incorrect_format/worst@16/std":0.004610765367099621,"rollout_corr/rollout_is_mean":1.0000677108764648,"val-aux/num_turns/max":2,"timing_s/agent_loop/generate_sequences/min":1.463550142943859,"critic/returns/mean":-0.1100277230143547,"rollout_corr/ppl_ratio":1.0003064163029194,"timing_s/agent_loop/generate_sequences/max":59.862675316631794,"rollout_corr/kl":0.0003049511138506489,"val-aux/sciknoweval/incorrect_format/maj@4/std":0.0015374570050248517,"val-aux/sciknoweval/acc/maj@8/mean":0.7632142857142856,"response_length_non_aborted/mean":426.9609375,"perf/mfu/actor_infer":0,"response_length/min":111,"rollout_corr/rollout_is_seq_min":0.9953861832618713,"critic/score/max":1,"val-aux/sciknoweval/reward/best@4/std":0.08027809906180897,"val-aux/sciknoweval/incorrect_format/worst@8/std":0.004640859533041082,"training/rollout_actor_probs_pearson_corr":0.9996629357337952,"timing_s/update_actor":24.305801395326853,"val-aux/sciknoweval/incorrect_format/maj@16/std":0.0002604287412166727,"global_seqlen/mean":23139.75,"val-aux/sciknoweval/score/worst@8/std":0.04212482341518254,"val-aux/sciknoweval/acc/worst@8/std":0.04212482341518254,"prompt_length/max":625,"val-aux/sciknoweval/reward/maj@8/mean":0.7632142857142856,"val-aux/sciknoweval/score/best@8/std":0.05177293853368067,"val-aux/sciknoweval/score/worst@8/mean":0.6388952380952381,"self_distillation/success_sample_fraction":0.93359375,"perf/cpu_memory_used_gb":198.26213359832764,"response_length/clip_ratio":0.01171875,"val-aux/sciknoweval/reward/maj@2/mean":0.7537380952380953,"training/rollout_probs_diff_mean":0.002571779303252697,"rollout_corr/rollout_is_ratio_fraction_high":0,"rollout_corr/training_log_ppl":0.3963073937320587,"val-aux/sciknoweval/incorrect_format/maj@8/mean":0.9999476190476191,"response_length_non_aborted/max":8192,"val-aux/sciknoweval/reward/best@4/mean":0.8485619047619049,"val-aux/sciknoweval/reward/worst@4/mean":0.6645857142857142,"val-aux/sciknoweval/incorrect_format/maj@4/mean":0.9997333333333334,"val-aux/sciknoweval/reward/best@8/std":0.05177293853368067,"val-aux/sciknoweval/acc/best@2/std":0.10681452617322888,"rollout_corr/rollout_is_seq_fraction_low":0,"timing_per_token_ms/gen":0.6090940510229154,"self_distillation/success_group_fraction":0.9375,"timing_s/testing":128.6522780675441,"rollout_corr/rollout_is_min":0.7055256962776184,"training/global_step":100,"response_length/max":8192,"val-aux/sciknoweval/acc/best@8/std":0.05177293853368067,"val-aux/sciknoweval/reward/maj@4/std":0.10127914193936502,"val-aux/sciknoweval/incorrect_format/best@4/std":0,"val-aux/sciknoweval/incorrect_format/worst@2/std":0.0030314321460128213,"num_turns/mean":2,"val-aux/sciknoweval/reward/maj@2/std":0.12035272201748512,"val-aux/sciknoweval/incorrect_format/worst@16/mean":0.994052380952381,"rollout_corr/rollout_is_seq_max_deviation":0.005556344985961914,"training/rollout_probs_diff_max":0.12101149559020996,"self_distillation/dynamic_weighting_enabled":0,"timing_s/agent_loop/slowest/tool_calls":0,"val-aux/sciknoweval/score/maj@4/mean":0.7588809523809524,"val-aux/sciknoweval/incorrect_format/worst@4/mean":0.9978380952380952,"rollout_corr/rollout_is_seq_mean":1.0000479221343994,"global_seqlen/max":41519,"val-aux/sciknoweval/reward/best@16/std":0.02643611186706914,"timing_s/adv":0.5594744142144918,"_runtime":5598.712326179,"perf/max_memory_reserved_gb":161.068359375,"val-aux/sciknoweval/score/std@16":0.12122068576933157,"val-aux/sciknoweval/incorrect_format/std@16":0.0023053472298853674,"val-aux/sciknoweval/incorrect_format/best@8/std":0,"training/epoch":1,"val-aux/sciknoweval/acc/worst@16/mean":0.623247619047619,"critic/rewards/mean":0.74609375,"val-aux/sciknoweval/incorrect_format/worst@8/mean":0.9963047619047618,"val-aux/sciknoweval/incorrect_format/worst@2/mean":0.9989095238095239,"num_turns/max":2,"val-aux/sciknoweval/reward/worst@8/std":0.04212482341518254,"rollout_corr/rollout_is_std":0.022850310429930687,"val-aux/sciknoweval/score/maj@8/std":0.07759319570379096,"val-aux/sciknoweval/reward/best@2/mean":0.8051285714285714,"global_seqlen/min":18884,"timing_s/stop_profile":0.0001576654613018036,"val-aux/sciknoweval/incorrect_format/maj@16/mean":0.9999857142857144,"rollout_corr/rollout_is_ratio_fraction_low":0,"val-aux/sciknoweval/score/maj@16/std":0.05543722468721845,"val-aux/sciknoweval/reward/worst@8/mean":0.6388952380952381,"val-aux/sciknoweval/acc/worst@8/mean":0.6388952380952381,"global_seqlen/balanced_max":27699,"actor/entropy":0.2955058515071869,"rollout_corr/log_ppl_abs_diff":0.0012992577630939195,"val-core/sciknoweval/acc/mean@16":0.7535714285714286,"timing_s/old_log_prob":3.303921103477478,"val-aux/sciknoweval/reward/worst@16/mean":0.623247619047619,"num_turns/min":2,"val-aux/sciknoweval/reward/maj@16/mean":0.7659095238095238,"val-aux/sciknoweval/score/maj@4/std":0.10127914193936502,"rollout_corr/log_ppl_diff":0.0003049547322007129,"val-aux/sciknoweval/reward/maj@4/mean":0.7588809523809524,"val-aux/sciknoweval/incorrect_format/best@16/std":0,"val-aux/sciknoweval/score/worst@4/mean":0.6645857142857142,"val-aux/sciknoweval/incorrect_format/best@8/mean":1,"val-core/sciknoweval/acc/maj@16/mean":0.7659095238095238,"timing_s/step":94.8968273177743,"val-aux/sciknoweval/reward/maj@16/std":0.05543722468721845,"timing_per_token_ms/update_actor":0.13129896279846828,"response/aborted_ratio":0,"critic/score/mean":0.74609375,"response_length_non_aborted/min":111,"timing_s/save_checkpoint":12.790537357330322,"val-aux/sciknoweval/score/best@4/mean":0.8485619047619049,"rollout_corr/rollout_log_ppl":0.396002438999858,"val-aux/sciknoweval/score/best@8/mean":0.8807,"timing_s/agent_loop/generate_sequences/mean":4.768673313919862,"critic/score/min":0,"_timestamp":1.7830265440559719e+09,"self_distillation/feedback_available_fraction":0,"timing_s/agent_loop/slowest/prompt_length":326,"val-aux/sciknoweval/incorrect_format/worst@4/std":0.003989388429962291,"val-aux/num_turns/min":2,"self_distillation/reprompt_sample_fraction":0.93359375,"val-aux/sciknoweval/acc/worst@2/mean":0.7031761904761905,"val-aux/sciknoweval/score/maj@8/mean":0.7632142857142856,"val-aux/sciknoweval/incorrect_format/maj@2/mean":0.9994619047619048,"val-aux/num_turns/mean":2,"rollout_corr/rollout_ppl":1.5010556285269558,"response_length/mean":426.9609375,"perf/throughput":243.84113414575552,"actor/pg_loss":0.023008882750787052,"val-aux/sciknoweval/score/maj@2/std":0.12035272201748512,"val-aux/sciknoweval/score/best@16/mean":0.9000428571428573,"val-aux/sciknoweval/score/worst@16/mean":0.623247619047619,"val-aux/sciknoweval/score/best@2/mean":0.8051285714285714,"perf/time_per_step":94.8968273177743,"critic/advantages/max":0.875,"rollout_corr/k3_kl":0.0003528280437024023,"_step":100,"val-aux/sciknoweval/incorrect_format/maj@8/std":0.0004966783087710074,"val-aux/sciknoweval/reward/mean@16":0.7535714285714286,"rollout_corr/log_ppl_diff_max":0.00518268346786499,"val-aux/sciknoweval/reward/maj@8/std":0.07759319570379096,"self_distillation/feedback_used_fraction":0,"val-core/sciknoweval/acc/best@16/mean":0.9000428571428573,"val-aux/sciknoweval/reward/std@16":0.12122068576933157,"val-aux/sciknoweval/reward/best@2/std":0.10681452617322888,"rollout_corr/rollout_is_seq_fraction_high":0,"val-aux/sciknoweval/incorrect_format/best@16/mean":1,"val-aux/sciknoweval/reward/worst@2/mean":0.7031761904761905,"val-aux/sciknoweval/reward/best@8/mean":0.8807,"rollout_corr/rollout_is_eff_sample_size":0.999478016711329,"timing_per_token_ms/adv":0.003022258312073876,"timing_s/agent_loop/slowest/response_length":8192,"rollout_corr/training_ppl":1.5015285341069102,"rollout_corr/chi2_seq":0.38230782863684,"perf/max_memory_allocated_gb":150.15545225143433,"val-aux/sciknoweval/incorrect_format/mean@16":0.9994047619047619,"val-aux/sciknoweval/acc/worst@2/std":0.09916615120061645,"prompt_length/mean":296.15625,"val-aux/sciknoweval/acc/std@16":0.12122068576933157,"timing_s/start_profile":0.00015718117356300354,"val-aux/sciknoweval/acc/worst@4/std":0.06781975137735766,"val-aux/sciknoweval/score/maj@2/mean":0.7537380952380953,"val-core/sciknoweval/acc/maj@16/std":0.05543722468721845,"global_seqlen/minmax_diff":22635,"critic/advantages/min":-0.875,"val-aux/sciknoweval/acc/worst@16/std":0.02453756899133995,"timing_s/gen":66.57519796490669,"val-aux/sciknoweval/score/worst@2/mean":0.7031761904761905,"val-aux/sciknoweval/acc/best@4/std":0.08027809906180897,"training/rollout_probs_diff_std":0.006634686142206192,"val-aux/sciknoweval/score/best@2/std":0.10681452617322888,"response_length_non_aborted/clip_ratio":0.01171875,"val-aux/sciknoweval/reward/best@16/mean":0.9000428571428573,"val-aux/sciknoweval/acc/maj@2/std":0.12035272201748512,"timing_s/agent_loop/slowest/generate_sequences":59.862675316631794,"val-aux/sciknoweval/acc/maj@2/mean":0.7537380952380953,"val-aux/sciknoweval/score/worst@2/std":0.09916615120061645,"actor/grad_norm":0.10394406318664551,"val-aux/sciknoweval/score/best@16/std":0.02643611186706914,"val-aux/sciknoweval/incorrect_format/best@4/mean":1,"critic/advantages/mean":-0.1100277230143547,"val-aux/sciknoweval/reward/worst@2/std":0.09916615120061645,"self_distillation/empty_target_batch":0.06640625,"val-aux/sciknoweval/acc/maj@4/mean":0.7588809523809524,"val-aux/sciknoweval/score/best@4/std":0.08027809906180897,"rollout_corr/log_ppl_diff_min":-0.00502324104309082,"prompt_length/min":165,"val-aux/sciknoweval/reward/worst@16/std":0.02453756899133995,"val-aux/sciknoweval/acc/best@4/mean":0.8485619047619049,"val-aux/sciknoweval/acc/best@2/mean":0.8051285714285714,"val-aux/sciknoweval/score/mean@16":0.7535714285714286,"critic/rewards/min":0,"val-aux/sciknoweval/incorrect_format/best@2/mean":0.9999666666666667,"val-aux/sciknoweval/incorrect_format/maj@2/std":0.0021986843400458376,"timing_s/reward":0.06458244659006596,"val-aux/sciknoweval/acc/best@8/mean":0.8807,"rollout_corr/chi2_token":0.0008032168261706829,"timing_s/agent_loop/tool_calls/max":0,"timing_s/agent_loop/tool_calls/min":0,"perf/total_num_tokens":185118,"actor/kl_loss":0,"rollout_corr/rollout_is_seq_std":0.0016783673781901598,"training/rollout_probs_diff_valid":1,"actor/lr":1e-05,"val-aux/sciknoweval/acc/maj@4/std":0.10127914193936502} |