Qwen3-4B-Chemistry-SRPO-TR/artifacts/output.log

wandb: Detected [openai] in use.
wandb: Use W&B Weave for improved LLM call tracing. Install Weave with `pip install weave` then add `import weave` to the top of your script.
wandb: For more information, check out the docs at: https://weave-docs.wandb.ai/
Checkpoint tracker file does not exist: /workspace/L2T/checkpoints/datasets/sciknoweval/chemistry/qwen3gen-chemistry-SRPO_TR-Qwen-Qwen3-4B-mbs32-tr0.1-dwtrue-train32-rollout8-lr5e-6-vllm0.8/latest_checkpointed_iteration.txt
Training from scratch
Training Progress:   9%|▉         | 9/100 [06:41<1:08:06, 44.90s/it]
step:1 - global_seqlen/min:15337 - global_seqlen/max:19940 - global_seqlen/minmax_diff:4603 - global_seqlen/balanced_min:18633 - global_seqlen/balanced_max:19042 - global_seqlen/mean:18688.0 - actor/entropy:0.21700496971607208 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.353939026594162 - training/rollout_probs_diff_mean:0.005942696705460548 - training/rollout_probs_diff_std:0.01577228680253029 - training/rollout_actor_probs_pearson_corr:0.9973705410957336 - self_distillation/success_group_fraction:0.78125 - self_distillation/success_sample_fraction:0.77734375 - self_distillation/correct_sample_fraction:0.59375 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.77734375 - rollout_corr/rollout_is_mean:1.000277042388916 - rollout_corr/rollout_is_ratio_fraction_high:3.594708687160164e-05 - rollout_corr/rollout_is_ratio_fraction_low:0.00015577070007566363 - rollout_corr/rollout_is_max:3.89467716217041 - rollout_corr/rollout_is_min:0.29519152641296387 - rollout_corr/rollout_is_std:0.04623226821422577 - rollout_corr/rollout_is_eff_sample_size:0.997868085880857 - rollout_corr/rollout_is_seq_mean:1.0003846883773804 - rollout_corr/rollout_is_seq_std:0.0028691221959888935 - rollout_corr/rollout_is_seq_max:1.0124605894088745 - rollout_corr/rollout_is_seq_min:0.9909685254096985 - rollout_corr/rollout_is_seq_max_deviation:0.012460589408874512 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.2600004188716412) - rollout_corr/training_log_ppl:np.float64(0.22832959592051338) - rollout_corr/kl:np.float64(0.0008263703017519219) - rollout_corr/k3_kl:np.float64(0.001211077764651236) - rollout_corr/rollout_ppl:np.float64(1.2589329704642296) - rollout_corr/rollout_log_ppl:np.float64(0.22750322484353092) - rollout_corr/log_ppl_diff:np.float64(0.0008263710769824684) - rollout_corr/log_ppl_abs_diff:np.float64(0.002128550666384399) - rollout_corr/log_ppl_diff_max:np.float64(0.012260347604751587) - rollout_corr/log_ppl_diff_min:np.float64(-0.007211625576019287) - rollout_corr/ppl_ratio:np.float64(1.0008303739596158) - rollout_corr/chi2_token:np.float64(0.0033761742524802685) - rollout_corr/chi2_seq:np.float64(0.6491703353822231) - actor/pg_loss:np.float64(-0.0703192461260187) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.07713406294351444) - srpo/sdpo_loss:np.float64(0.006814816817495739) - srpo/routed_loss:np.float64(-0.0703192461260187) - srpo/routed_token_count:np.float64(326.0) - srpo/correct_sample_fraction:np.float64(0.59375) - srpo/sdpo_sample_fraction:np.float64(0.1875) - srpo/grpo_sample_fraction:np.float64(0.8125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.03272323397686705) - self_distillation/dynamic_weight_mean:np.float64(0.18750000023283064) - self_distillation/dynamic_weight_max:np.float64(1.2397534847259521) - self_distillation/empty_target_batch:np.float64(0.22265625) - actor/grad_norm:np.float64(0.31313756108283997) - perf/mfu/actor:np.float64(0.027500897822924846) - perf/max_memory_allocated_gb:np.float64(115.95331954956055) - perf/max_memory_reserved_gb:np.float64(121.171875) - perf/cpu_memory_used_gb:np.float64(155.59004163742065) - actor/lr:np.float64(0.0) - training/global_step:1 - training/epoch:0 - critic/score/mean:0.59375 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.59375 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.0067655411548912525 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.0067655411548912525 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:326.0 - response_length/max:1352.0 - response_length/min:147.0 - response_length/clip_r
step:2 - global_seqlen/min:16609 - global_seqlen/max:27639 - global_seqlen/minmax_diff:11030 - global_seqlen/balanced_min:20813 - global_seqlen/balanced_max:20818 - global_seqlen/mean:20815.625 - actor/entropy:0.21829853951931 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.3394761085510254 - training/rollout_probs_diff_mean:0.006095316726714373 - training/rollout_probs_diff_std:0.01596890576183796 - training/rollout_actor_probs_pearson_corr:0.9973609447479248 - self_distillation/success_group_fraction:0.65625 - self_distillation/success_sample_fraction:0.6328125 - self_distillation/correct_sample_fraction:0.37890625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.6328125 - rollout_corr/rollout_is_mean:1.0002561807632446 - rollout_corr/rollout_is_ratio_fraction_high:5.900331598240882e-05 - rollout_corr/rollout_is_ratio_fraction_low:0.00011800663196481764 - rollout_corr/rollout_is_max:4.452920913696289 - rollout_corr/rollout_is_min:0.31313827633857727 - rollout_corr/rollout_is_std:0.04850232973694801 - rollout_corr/rollout_is_eff_sample_size:0.9976541128901532 - rollout_corr/rollout_is_seq_mean:1.00030517578125 - rollout_corr/rollout_is_seq_std:0.003281550481915474 - rollout_corr/rollout_is_seq_max:1.0109145641326904 - rollout_corr/rollout_is_seq_min:0.9904547929763794 - rollout_corr/rollout_is_seq_max_deviation:0.01091456413269043 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.2722020936198533) - rollout_corr/training_log_ppl:np.float64(0.23766601431998424) - rollout_corr/kl:np.float64(0.0009567292685428441) - rollout_corr/k3_kl:np.float64(0.0012618661846772738) - rollout_corr/rollout_ppl:np.float64(1.2709640711545944) - rollout_corr/rollout_log_ppl:np.float64(0.23670928270439617) - rollout_corr/log_ppl_diff:np.float64(0.000956731615588069) - rollout_corr/log_ppl_abs_diff:np.float64(0.002635535958688706) - rollout_corr/log_ppl_diff_max:np.float64(0.012460693717002869) - rollout_corr/log_ppl_diff_min:np.float64(-0.007930487394332886) - rollout_corr/ppl_ratio:np.float64(1.000962538877502) - rollout_corr/chi2_token:np.float64(0.003215277101844549) - rollout_corr/chi2_seq:np.float64(2.0966858628671616) - actor/pg_loss:np.float64(-0.06847139411547687) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.077676738234004) - srpo/sdpo_loss:np.float64(0.009205344118527137) - srpo/routed_loss:np.float64(-0.06847139411547687) - srpo/routed_token_count:np.float64(331.01953125) - srpo/correct_sample_fraction:np.float64(0.37890625) - srpo/sdpo_sample_fraction:np.float64(0.27734375) - srpo/grpo_sample_fraction:np.float64(0.72265625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.06322825237293728) - self_distillation/dynamic_weight_mean:np.float64(0.27734375023283064) - self_distillation/dynamic_weight_max:np.float64(1.316468358039856) - self_distillation/empty_target_batch:np.float64(0.3671875) - actor/grad_norm:np.float64(0.3211483955383301) - perf/mfu/actor:np.float64(0.037297831652202976) - perf/max_memory_allocated_gb:np.float64(118.93788957595825) - perf/max_memory_reserved_gb:np.float64(123.7890625) - perf/cpu_memory_used_gb:np.float64(151.38793563842773) - actor/lr:np.float64(5.000000000000001e-07) - training/global_step:2 - training/epoch:0 - critic/score/mean:0.37890625 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.37890625 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.0014131293864920735 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.0014131293864920735 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:331.01953125 - response_length/max:824.0 - respo
step:3 - global_seqlen/min:16908 - global_seqlen/max:24427 - global_seqlen/minmax_diff:7519 - global_seqlen/balanced_min:20103 - global_seqlen/balanced_max:20113 - global_seqlen/mean:20109.875 - actor/entropy:0.22853685915470123 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.7399731874465942 - training/rollout_probs_diff_mean:0.005943723022937775 - training/rollout_probs_diff_std:0.01592753827571869 - training/rollout_actor_probs_pearson_corr:0.9974951148033142 - self_distillation/success_group_fraction:0.71875 - self_distillation/success_sample_fraction:0.70703125 - self_distillation/correct_sample_fraction:0.41796875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.70703125 - rollout_corr/rollout_is_mean:1.000004768371582 - rollout_corr/rollout_is_ratio_fraction_high:3.5950961319031194e-05 - rollout_corr/rollout_is_ratio_fraction_low:0.0001318201975664124 - rollout_corr/rollout_is_max:2.521167755126953 - rollout_corr/rollout_is_min:0.010440251789987087 - rollout_corr/rollout_is_std:0.04662127047777176 - rollout_corr/rollout_is_eff_sample_size:0.997831171066516 - rollout_corr/rollout_is_seq_mean:1.0000784397125244 - rollout_corr/rollout_is_seq_std:0.003086655866354704 - rollout_corr/rollout_is_seq_max:1.0102639198303223 - rollout_corr/rollout_is_seq_min:0.9887522459030151 - rollout_corr/rollout_is_seq_max_deviation:0.011247754096984863 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.2868499774485826) - rollout_corr/training_log_ppl:np.float64(0.24925029833684675) - rollout_corr/kl:np.float64(0.0011279603448315356) - rollout_corr/k3_kl:np.float64(0.0012063818591059317) - rollout_corr/rollout_ppl:np.float64(1.2853895733132958) - rollout_corr/rollout_log_ppl:np.float64(0.2481223378854338) - rollout_corr/log_ppl_diff:np.float64(0.001127960451412946) - rollout_corr/log_ppl_abs_diff:np.float64(0.0025958048063330352) - rollout_corr/log_ppl_diff_max:np.float64(0.01622951030731201) - rollout_corr/log_ppl_diff_min:np.float64(-0.008839517831802368) - rollout_corr/ppl_ratio:np.float64(1.0011336305178702) - rollout_corr/chi2_token:np.float64(0.0025189488660544157) - rollout_corr/chi2_seq:np.float64(1.2722296840511262) - actor/pg_loss:np.float64(-0.09052861935924739) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.10305640715523623) - srpo/sdpo_loss:np.float64(0.012527787795988843) - srpo/routed_loss:np.float64(-0.09052861935924739) - srpo/routed_token_count:np.float64(325.96484375) - srpo/correct_sample_fraction:np.float64(0.41796875) - srpo/sdpo_sample_fraction:np.float64(0.30078125) - srpo/grpo_sample_fraction:np.float64(0.69921875) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.07666321203578264) - self_distillation/dynamic_weight_mean:np.float64(0.3007812546566129) - self_distillation/dynamic_weight_max:np.float64(1.34437096118927) - self_distillation/empty_target_batch:np.float64(0.29296875) - actor/grad_norm:np.float64(0.4573976695537567) - perf/mfu/actor:np.float64(0.036088217792550704) - perf/max_memory_allocated_gb:np.float64(118.93788957595825) - perf/max_memory_reserved_gb:np.float64(123.7890625) - perf/cpu_memory_used_gb:np.float64(149.50705337524414) - actor/lr:np.float64(1.0000000000000002e-06) - training/global_step:3 - training/epoch:0 - critic/score/mean:0.41796875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.41796875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.0046182009391486645 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.0046182009391486645 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:325.96484375 - response_length/max:925.0
step:4 - global_seqlen/min:14893 - global_seqlen/max:22954 - global_seqlen/minmax_diff:8061 - global_seqlen/balanced_min:19463 - global_seqlen/balanced_max:19468 - global_seqlen/mean:19466.125 - actor/entropy:0.2511502504348755 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.7468032836914062 - training/rollout_probs_diff_mean:0.006447272840887308 - training/rollout_probs_diff_std:0.016030525788664818 - training/rollout_actor_probs_pearson_corr:0.9975630044937134 - self_distillation/success_group_fraction:0.625 - self_distillation/success_sample_fraction:0.62109375 - self_distillation/correct_sample_fraction:0.3671875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.62109375 - rollout_corr/rollout_is_mean:0.99993896484375 - rollout_corr/rollout_is_ratio_fraction_high:2.5903715140884742e-05 - rollout_corr/rollout_is_ratio_fraction_low:0.00011656671995297074 - rollout_corr/rollout_is_max:2.1845600605010986 - rollout_corr/rollout_is_min:0.1104939728975296 - rollout_corr/rollout_is_std:0.047994595021009445 - rollout_corr/rollout_is_eff_sample_size:0.997701456743119 - rollout_corr/rollout_is_seq_mean:1.0000107288360596 - rollout_corr/rollout_is_seq_std:0.002924026222899556 - rollout_corr/rollout_is_seq_max:1.0110195875167847 - rollout_corr/rollout_is_seq_min:0.992973268032074 - rollout_corr/rollout_is_seq_max_deviation:0.011019587516784668 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.3049874636344612) - rollout_corr/training_log_ppl:np.float64(0.26412568462546915) - rollout_corr/kl:np.float64(0.0012122795804199171) - rollout_corr/k3_kl:np.float64(0.001223077373651904) - rollout_corr/rollout_ppl:np.float64(1.3033996932208538) - rollout_corr/rollout_log_ppl:np.float64(0.26291340473107994) - rollout_corr/log_ppl_diff:np.float64(0.0012122798943892121) - rollout_corr/log_ppl_abs_diff:np.float64(0.002499745925888419) - rollout_corr/log_ppl_diff_max:np.float64(0.008524119853973389) - rollout_corr/log_ppl_diff_min:np.float64(-0.010081544518470764) - rollout_corr/ppl_ratio:np.float64(1.0012174150906503) - rollout_corr/chi2_token:np.float64(0.0024151464458554983) - rollout_corr/chi2_seq:np.float64(0.7148186564445496) - actor/pg_loss:np.float64(-0.0902680644112479) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.09962156144320033) - srpo/sdpo_loss:np.float64(0.009353497031952429) - srpo/routed_loss:np.float64(-0.0902680644112479) - srpo/routed_token_count:np.float64(301.59765625) - srpo/correct_sample_fraction:np.float64(0.3671875) - srpo/sdpo_sample_fraction:np.float64(0.2578125) - srpo/grpo_sample_fraction:np.float64(0.7421875) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.06475623702863231) - self_distillation/dynamic_weight_mean:np.float64(0.2578125016298145) - self_distillation/dynamic_weight_max:np.float64(1.3603585958480835) - self_distillation/empty_target_batch:np.float64(0.37890625) - actor/grad_norm:np.float64(0.38336315751075745) - perf/mfu/actor:np.float64(0.03515670622823734) - perf/max_memory_allocated_gb:np.float64(118.93788957595825) - perf/max_memory_reserved_gb:np.float64(123.7890625) - perf/cpu_memory_used_gb:np.float64(146.19318389892578) - actor/lr:np.float64(1.5e-06) - training/global_step:4 - training/epoch:0 - critic/score/mean:0.3671875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.3671875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.0017760235350579023 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.0017760235350579023 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:301.59765625 - response_length/max:850.0 - response_length/min
step:5 - global_seqlen/min:15752 - global_seqlen/max:25512 - global_seqlen/minmax_diff:9760 - global_seqlen/balanced_min:19026 - global_seqlen/balanced_max:26130 - global_seqlen/mean:19916.5 - actor/entropy:0.22281140089035034 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.42376136779785156 - training/rollout_probs_diff_mean:0.005535186268389225 - training/rollout_probs_diff_std:0.014674077741801739 - training/rollout_actor_probs_pearson_corr:0.99783855676651 - self_distillation/success_group_fraction:0.875 - self_distillation/success_sample_fraction:0.85546875 - self_distillation/correct_sample_fraction:0.50390625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.85546875 - rollout_corr/rollout_is_mean:0.9997702240943909 - rollout_corr/rollout_is_ratio_fraction_high:3.3603657357161865e-05 - rollout_corr/rollout_is_ratio_fraction_low:4.4804874050896615e-05 - rollout_corr/rollout_is_max:2.3356175422668457 - rollout_corr/rollout_is_min:0.3917287588119507 - rollout_corr/rollout_is_std:0.044353704899549484 - rollout_corr/rollout_is_eff_sample_size:0.9980357206168908 - rollout_corr/rollout_is_seq_mean:0.9997634887695312 - rollout_corr/rollout_is_seq_std:0.0026603336445987225 - rollout_corr/rollout_is_seq_max:1.008905291557312 - rollout_corr/rollout_is_seq_min:0.9922382831573486 - rollout_corr/rollout_is_seq_max_deviation:0.008905291557312012 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.3019596561789513) - rollout_corr/training_log_ppl:np.float64(0.2605670520770218) - rollout_corr/kl:np.float64(0.0013656955257896897) - rollout_corr/k3_kl:np.float64(0.0011291429811706166) - rollout_corr/rollout_ppl:np.float64(1.3001631228253245) - rollout_corr/rollout_log_ppl:np.float64(0.25920135507840314) - rollout_corr/log_ppl_diff:np.float64(0.0013656969986186596) - rollout_corr/log_ppl_abs_diff:np.float64(0.0023622289172635647) - rollout_corr/log_ppl_diff_max:np.float64(0.008890509605407715) - rollout_corr/log_ppl_diff_min:np.float64(-0.006910637021064758) - rollout_corr/ppl_ratio:np.float64(1.0013701494317502) - rollout_corr/chi2_token:np.float64(0.0018071448430418968) - rollout_corr/chi2_seq:np.float64(0.22499786876142025) - actor/pg_loss:np.float64(-0.09476500544792543) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.10982142353896052) - srpo/sdpo_loss:np.float64(0.015056418091035084) - srpo/routed_loss:np.float64(-0.09476500544792543) - srpo/routed_token_count:np.float64(348.734375) - srpo/correct_sample_fraction:np.float64(0.50390625) - srpo/sdpo_sample_fraction:np.float64(0.37109375) - srpo/grpo_sample_fraction:np.float64(0.62890625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.08710912278729666) - self_distillation/dynamic_weight_mean:np.float64(0.37109375139698386) - self_distillation/dynamic_weight_max:np.float64(1.3746187686920166) - self_distillation/empty_target_batch:np.float64(0.14453125) - actor/grad_norm:np.float64(0.46219602227211) - perf/mfu/actor:np.float64(0.03359267382705325) - perf/max_memory_allocated_gb:np.float64(137.8146800994873) - perf/max_memory_reserved_gb:np.float64(142.3046875) - perf/cpu_memory_used_gb:np.float64(144.12317276000977) - actor/lr:np.float64(2.0000000000000003e-06) - training/global_step:5 - training/epoch:0 - critic/score/mean:0.50390625 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.50390625 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.010883384384214878 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.010883384384214878 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:348.734375 - response_length/max:8192.0 -
step:6 - global_seqlen/min:17236 - global_seqlen/max:21216 - global_seqlen/minmax_diff:3980 - global_seqlen/balanced_min:19597 - global_seqlen/balanced_max:19609 - global_seqlen/mean:19603.75 - actor/entropy:0.24815915524959564 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.38898196816444397 - training/rollout_probs_diff_mean:0.005613780114799738 - training/rollout_probs_diff_std:0.01433836854994297 - training/rollout_actor_probs_pearson_corr:0.9980629086494446 - self_distillation/success_group_fraction:0.71875 - self_distillation/success_sample_fraction:0.69921875 - self_distillation/correct_sample_fraction:0.421875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.69921875 - rollout_corr/rollout_is_mean:0.9997754693031311 - rollout_corr/rollout_is_ratio_fraction_high:3.469411240075715e-05 - rollout_corr/rollout_is_ratio_fraction_low:0.00010408234084025025 - rollout_corr/rollout_is_max:3.066223621368408 - rollout_corr/rollout_is_min:0.1570271998643875 - rollout_corr/rollout_is_std:0.044291846454143524 - rollout_corr/rollout_is_eff_sample_size:0.9980410640097951 - rollout_corr/rollout_is_seq_mean:0.9997889399528503 - rollout_corr/rollout_is_seq_std:0.0028244946151971817 - rollout_corr/rollout_is_seq_max:1.0091670751571655 - rollout_corr/rollout_is_seq_min:0.9896945357322693 - rollout_corr/rollout_is_seq_max_deviation:0.010305464267730713 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.3130784477107227) - rollout_corr/training_log_ppl:np.float64(0.2680182019539643) - rollout_corr/kl:np.float64(0.0012861443127256678) - rollout_corr/k3_kl:np.float64(0.0010751346092092717) - rollout_corr/rollout_ppl:np.float64(1.3113522394560277) - rollout_corr/rollout_log_ppl:np.float64(0.2667320551699959) - rollout_corr/log_ppl_diff:np.float64(0.0012861467839684337) - rollout_corr/log_ppl_abs_diff:np.float64(0.0024161361216101795) - rollout_corr/log_ppl_diff_max:np.float64(0.012646734714508057) - rollout_corr/log_ppl_diff_min:np.float64(-0.006797090172767639) - rollout_corr/ppl_ratio:np.float64(1.0012912419624627) - rollout_corr/chi2_token:np.float64(0.0017323305364698172) - rollout_corr/chi2_seq:np.float64(0.554826382547617) - actor/pg_loss:np.float64(-0.0822394966162392) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.09276439799577929) - srpo/sdpo_loss:np.float64(0.010524901379540097) - srpo/routed_loss:np.float64(-0.0822394966162392) - srpo/routed_token_count:np.float64(337.7734375) - srpo/correct_sample_fraction:np.float64(0.421875) - srpo/sdpo_sample_fraction:np.float64(0.296875) - srpo/grpo_sample_fraction:np.float64(0.703125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.07022242833045311) - self_distillation/dynamic_weight_mean:np.float64(0.296875006519258) - self_distillation/dynamic_weight_max:np.float64(1.3328691720962524) - self_distillation/empty_target_batch:np.float64(0.30078125) - actor/grad_norm:np.float64(0.3473239839076996) - perf/mfu/actor:np.float64(0.03534663041616369) - perf/max_memory_allocated_gb:np.float64(137.8146800994873) - perf/max_memory_reserved_gb:np.float64(142.3046875) - perf/cpu_memory_used_gb:np.float64(143.67934799194336) - actor/lr:np.float64(2.5e-06) - training/global_step:6 - training/epoch:0 - critic/score/mean:0.421875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.421875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.005014745052903891 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.005014745052903891 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:337.7734375 - response_length/max:1090.0 - response_length/min:147.0 -
step:7 - global_seqlen/min:15502 - global_seqlen/max:21641 - global_seqlen/minmax_diff:6139 - global_seqlen/balanced_min:18399 - global_seqlen/balanced_max:18401 - global_seqlen/mean:18400.375 - actor/entropy:0.29591771960258484 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.3016374111175537 - training/rollout_probs_diff_mean:0.006198689341545105 - training/rollout_probs_diff_std:0.014233511872589588 - training/rollout_actor_probs_pearson_corr:0.9982411861419678 - self_distillation/success_group_fraction:0.6875 - self_distillation/success_sample_fraction:0.6640625 - self_distillation/correct_sample_fraction:0.36328125 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.6640625 - rollout_corr/rollout_is_mean:0.9998568296432495 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:2.731531458266545e-05 - rollout_corr/rollout_is_max:1.6489237546920776 - rollout_corr/rollout_is_min:0.44956451654434204 - rollout_corr/rollout_is_std:0.04394938051700592 - rollout_corr/rollout_is_eff_sample_size:0.9980717005668774 - rollout_corr/rollout_is_seq_mean:0.9997868537902832 - rollout_corr/rollout_is_seq_std:0.002762353979051113 - rollout_corr/rollout_is_seq_max:1.0084675550460815 - rollout_corr/rollout_is_seq_min:0.9917079210281372 - rollout_corr/rollout_is_seq_max_deviation:0.008467555046081543 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.366286383010447) - rollout_corr/training_log_ppl:np.float64(0.30824938564910553) - rollout_corr/kl:np.float64(0.001253259818801311) - rollout_corr/k3_kl:np.float64(0.0010400572712114808) - rollout_corr/rollout_ppl:np.float64(1.3645314392633736) - rollout_corr/rollout_log_ppl:np.float64(0.30699612337048166) - rollout_corr/log_ppl_diff:np.float64(0.001253262278623879) - rollout_corr/log_ppl_abs_diff:np.float64(0.002334850956685841) - rollout_corr/log_ppl_diff_max:np.float64(0.010306566953659058) - rollout_corr/log_ppl_diff_min:np.float64(-0.006784215569496155) - rollout_corr/ppl_ratio:np.float64(1.0012580519542098) - rollout_corr/chi2_token:np.float64(0.0016352287493646145) - rollout_corr/chi2_seq:np.float64(0.43829257879406214) - actor/pg_loss:np.float64(-0.08196726479945937) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.09418283405830152) - srpo/sdpo_loss:np.float64(0.012215569258842152) - srpo/routed_loss:np.float64(-0.08196726479945937) - srpo/routed_token_count:np.float64(286.01171875) - srpo/correct_sample_fraction:np.float64(0.36328125) - srpo/sdpo_sample_fraction:np.float64(0.32421875) - srpo/grpo_sample_fraction:np.float64(0.67578125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.08215987417497672) - self_distillation/dynamic_weight_mean:np.float64(0.32421875069849193) - self_distillation/dynamic_weight_max:np.float64(1.351878046989441) - self_distillation/empty_target_batch:np.float64(0.3359375) - actor/grad_norm:np.float64(0.35601046681404114) - perf/mfu/actor:np.float64(0.033251132956465215) - perf/max_memory_allocated_gb:np.float64(137.8146800994873) - perf/max_memory_reserved_gb:np.float64(142.3046875) - perf/cpu_memory_used_gb:np.float64(143.80960273742676) - actor/lr:np.float64(3e-06) - training/global_step:7 - training/epoch:0 - critic/score/mean:0.36328125 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.36328125 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.0032197926193475723 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.0032197926193475723 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:286.01171875 - response_length/max:592.0 - response_length/min:117.0 - res
step:8 - global_seqlen/min:15364 - global_seqlen/max:27246 - global_seqlen/minmax_diff:11882 - global_seqlen/balanced_min:18439 - global_seqlen/balanced_max:25670 - global_seqlen/mean:19347.625 - actor/entropy:0.2827216386795044 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.46715983748435974 - training/rollout_probs_diff_mean:0.005191668402403593 - training/rollout_probs_diff_std:0.013087828643620014 - training/rollout_actor_probs_pearson_corr:0.9985535740852356 - self_distillation/success_group_fraction:0.875 - self_distillation/success_sample_fraction:0.85546875 - self_distillation/correct_sample_fraction:0.51953125 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.85546875 - rollout_corr/rollout_is_mean:0.9996800422668457 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:5.0066963012795895e-05 - rollout_corr/rollout_is_max:1.7034646272659302 - rollout_corr/rollout_is_min:0.056605979800224304 - rollout_corr/rollout_is_std:0.040403611958026886 - rollout_corr/rollout_is_eff_sample_size:0.9983692579778656 - rollout_corr/rollout_is_seq_mean:0.9996277093887329 - rollout_corr/rollout_is_seq_std:0.003191681345924735 - rollout_corr/rollout_is_seq_max:1.0135866403579712 - rollout_corr/rollout_is_seq_min:0.9900147318840027 - rollout_corr/rollout_is_seq_max_deviation:0.013586640357971191 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.4152285596355796) - rollout_corr/training_log_ppl:np.float64(0.3390528556137724) - rollout_corr/kl:np.float64(0.0014037047080250886) - rollout_corr/k3_kl:np.float64(0.0010313593394926102) - rollout_corr/rollout_ppl:np.float64(1.413201994728297) - rollout_corr/rollout_log_ppl:np.float64(0.3376491487779276) - rollout_corr/log_ppl_diff:np.float64(0.0014037068358447868) - rollout_corr/log_ppl_abs_diff:np.float64(0.0027630017611954827) - rollout_corr/log_ppl_diff_max:np.float64(0.01595906913280487) - rollout_corr/log_ppl_diff_min:np.float64(-0.01148805022239685) - rollout_corr/ppl_ratio:np.float64(1.0014102652203292) - rollout_corr/chi2_token:np.float64(0.0012436762917786837) - rollout_corr/chi2_seq:np.float64(0.5322157160844654) - actor/pg_loss:np.float64(-0.10188636504562965) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.11465688009047881) - srpo/sdpo_loss:np.float64(0.012770515044849162) - srpo/routed_loss:np.float64(-0.10188636504562965) - srpo/routed_token_count:np.float64(312.08203125) - srpo/correct_sample_fraction:np.float64(0.51953125) - srpo/sdpo_sample_fraction:np.float64(0.35546875) - srpo/grpo_sample_fraction:np.float64(0.64453125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.08229101066581279) - self_distillation/dynamic_weight_mean:np.float64(0.35546875349245965) - self_distillation/dynamic_weight_max:np.float64(1.3635149002075195) - self_distillation/empty_target_batch:np.float64(0.14453125) - actor/grad_norm:np.float64(0.37420737743377686) - perf/mfu/actor:np.float64(0.03284429108631868) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(143.6675682067871) - actor/lr:np.float64(3.5e-06) - training/global_step:8 - training/epoch:0 - critic/score/mean:0.51953125 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.51953125 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.035973113030195236 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.035973113030195236 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:312.08203125 - response_length/max:8192.0 - response_length/min:113.0
step:9 - global_seqlen/min:13932 - global_seqlen/max:18627 - global_seqlen/minmax_diff:4695 - global_seqlen/balanced_min:16168 - global_seqlen/balanced_max:16169 - global_seqlen/mean:16168.625 - actor/entropy:0.360135942697525 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.6739586591720581 - training/rollout_probs_diff_mean:0.006341492757201195 - training/rollout_probs_diff_std:0.014084307476878166 - training/rollout_actor_probs_pearson_corr:0.9985119700431824 - self_distillation/success_group_fraction:0.65625 - self_distillation/success_sample_fraction:0.640625 - self_distillation/correct_sample_fraction:0.33203125 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.640625 - rollout_corr/rollout_is_mean:1.0001614093780518 - rollout_corr/rollout_is_ratio_fraction_high:5.4146737966220826e-05 - rollout_corr/rollout_is_ratio_fraction_low:3.6097826523473486e-05 - rollout_corr/rollout_is_max:2.39359188079834 - rollout_corr/rollout_is_min:0.3225047290325165 - rollout_corr/rollout_is_std:0.045027896761894226 - rollout_corr/rollout_is_eff_sample_size:0.9979771845259078 - rollout_corr/rollout_is_seq_mean:1.0001497268676758 - rollout_corr/rollout_is_seq_std:0.0031596811022609472 - rollout_corr/rollout_is_seq_max:1.0116209983825684 - rollout_corr/rollout_is_seq_min:0.99076247215271 - rollout_corr/rollout_is_seq_max_deviation:0.01162099838256836 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.447497637476772) - rollout_corr/training_log_ppl:np.float64(0.36257378016307484) - rollout_corr/kl:np.float64(0.0008900819554565897) - rollout_corr/k3_kl:np.float64(0.001039842043610406) - rollout_corr/rollout_ppl:np.float64(1.4461712059564888) - rollout_corr/rollout_log_ppl:np.float64(0.36168369596998673) - rollout_corr/log_ppl_diff:np.float64(0.0008900841930881143) - rollout_corr/log_ppl_abs_diff:np.float64(0.0024978589499369264) - rollout_corr/log_ppl_diff_max:np.float64(0.01060187816619873) - rollout_corr/log_ppl_diff_min:np.float64(-0.010052591562271118) - rollout_corr/ppl_ratio:np.float64(1.0008954918012023) - rollout_corr/chi2_token:np.float64(0.0024011670611798763) - rollout_corr/chi2_seq:np.float64(1.2476253274362534) - actor/pg_loss:np.float64(-0.07659202421928057) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.09132937749382108) - srpo/sdpo_loss:np.float64(0.01473735327454051) - srpo/routed_loss:np.float64(-0.07659202421928057) - srpo/routed_token_count:np.float64(216.42578125) - srpo/correct_sample_fraction:np.float64(0.33203125) - srpo/sdpo_sample_fraction:np.float64(0.32421875) - srpo/grpo_sample_fraction:np.float64(0.67578125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.07646568206837401) - self_distillation/dynamic_weight_mean:np.float64(0.32421875023283064) - self_distillation/dynamic_weight_max:np.float64(1.3421391248703003) - self_distillation/empty_target_batch:np.float64(0.359375) - actor/grad_norm:np.float64(0.3978496491909027) - perf/mfu/actor:np.float64(0.029078903174358885) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(143.66810989379883) - actor/lr:np.float64(4.000000000000001e-06) - training/global_step:9 - training/epoch:0 - critic/score/mean:0.33203125 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.33203125 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.015991335734725 - critic/advantages/max:0.875 - critic/advantages/min:-0.5 - critic/returns/mean:0.015991335734725 - critic/returns/max:0.875 - critic/returns/min:-0.5 - response_length/mean:216.42578125 - response_length/max:693.0 - response_length/
test_gen_batch meta info: {'eos_token_id': 151645, 'pad_token_id': 151643, 'recompute_log_prob': False, 'do_sample': True, 'validate': True, 'global_steps': 10}
validation generation end
[prompt] system

Given a question and four options, please select the right answer. Respond in the following format:
<reasoning>
...
</reasoning>
<answer>
...
</answer>

For the answer, only output the letter corresponding to the correct option (A, B, C, or D), and nothing else. Do not restate the answer text. For example, if the answer is "A", just output:
<answer>
A
</answer>

user
Which of the following molecules is the only correct reactant used in the synthesis of the product "Nc1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1" ?

A: O=C1c2ccccc2C(=O)N1c1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1
B: O=C1c2ccccc2C(Cl)N1c1ccc2ccc(Cl)nc2n1
C: O=C1c2ccccc2C(=O)N1Cc1ccc(Oc2ccc(Cl)c(C(F)(F)F)c2)cc1
D: O=C1c2ccccc2C(=O)N1CC=C(Cc1ccccc1)C(F)(F)F
Please reason step by step.
assistant
<think>

</think>


[response] <reasoning>
The product "Nc1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1" contains a substituted aryl group with a chlorine atom and a trifluoromethyl group. The correct reactant must include these substituents. Option A includes a ketone group and a substituted amine, which matches the structure of the product. Option B has a chlorine but lacks the trifluoromethyl group. Option C includes a hydroxyl group and a trifluoromethyl group, but not the correct amine structure. Option D includes a trifluoromethyl group but lacks the chlorine and correct amine structure. Therefore, the correct answer is A.
</reasoning>
<answer>
A
</answer>
[ground_truth] A
[score] 1.0
[acc] 1.0
[pred] A
[incorrect_format] 1
[feedback]
local_global_step_folder: /workspace/L2T/checkpoints/datasets/sciknoweval/chemistry/qwen3gen-chemistry-SRPO_TR-Qwen-Qwen3-4B-mbs32-tr0.1-dwtrue-train32-rollout8-lr5e-6-vllm0.8/global_step_10
step:10 - global_seqlen/min:13442 - global_seqlen/max:17402 - global_seqlen/minmax_diff:3960 - global_seqlen/balanced_min:14624 - global_seqlen/balanced_max:14628 - global_seqlen/mean:14627.0 - actor/entropy:0.3017302453517914 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.2099539041519165 - training/rollout_probs_diff_mean:0.006223808508366346 - training/rollout_probs_diff_std:0.014369786716997623 - training/rollout_actor_probs_pearson_corr:0.9982905983924866 - self_distillation/success_group_fraction:0.59375 - self_distillation/success_sample_fraction:0.5859375 - self_distillation/correct_sample_fraction:0.4453125 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.5859375 - rollout_corr/rollout_is_mean:1.0000187158584595 - rollout_corr/rollout_is_ratio_fraction_high:4.161811375524849e-05 - rollout_corr/rollout_is_ratio_fraction_low:2.0809056877624243e-05 - rollout_corr/rollout_is_max:2.3723411560058594 - rollout_corr/rollout_is_min:0.46146368980407715 - rollout_corr/rollout_is_std:0.04663916677236557 - rollout_corr/rollout_is_eff_sample_size:0.9978295093706268 - rollout_corr/rollout_is_seq_mean:0.9999683499336243 - rollout_corr/rollout_is_seq_std:0.0038052601739764214 - rollout_corr/rollout_is_seq_max:1.0122159719467163 - rollout_corr/rollout_is_seq_min:0.987821638584137 - rollout_corr/rollout_is_seq_max_deviation:0.012215971946716309 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.373092225752771) - rollout_corr/training_log_ppl:np.float64(0.3124894078646321) - rollout_corr/kl:np.float64(0.001139784962125745) - rollout_corr/k3_kl:np.float64(0.0011081025107841924) - rollout_corr/rollout_ppl:np.float64(1.3715080502443016) - rollout_corr/rollout_log_ppl:np.float64(0.31134961851057597) - rollout_corr/log_ppl_diff:np.float64(0.00113978935405612) - rollout_corr/log_ppl_abs_diff:np.float64(0.002987924905028194) - rollout_corr/log_ppl_diff_max:np.float64(0.013416081666946411) - rollout_corr/log_ppl_diff_min:np.float64(-0.010176807641983032) - rollout_corr/ppl_ratio:np.float64(1.0011475218925625) - rollout_corr/chi2_token:np.float64(0.002209543716162443) - rollout_corr/chi2_seq:np.float64(0.6262747284490615) - actor/pg_loss:np.float64(-0.05378324627963593) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.06053450555191375) - srpo/sdpo_loss:np.float64(0.006751259272277821) - srpo/routed_loss:np.float64(-0.05378324627963593) - srpo/routed_token_count:np.float64(187.71875) - srpo/correct_sample_fraction:np.float64(0.4453125) - srpo/sdpo_sample_fraction:np.float64(0.1484375) - srpo/grpo_sample_fraction:np.float64(0.8515625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.02956272999290377) - self_distillation/dynamic_weight_mean:np.float64(0.14843750069849193) - self_distillation/dynamic_weight_max:np.float64(1.2933728694915771) - self_distillation/empty_target_batch:np.float64(0.4140625) - actor/grad_norm:np.float64(0.2651667892932892) - perf/mfu/actor:np.float64(0.026753218633881963) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(143.65354919433594) - actor/lr:np.float64(4.5e-06) - val-aux/sciknoweval/reward/mean@16:np.float64(0.49851190476190477) - val-aux/sciknoweval/reward/std@16:np.float64(0.16970373138029815) - val-aux/sciknoweval/reward/best@2/mean:np.float64(0.5651380952380952) - val-aux/sciknoweval/reward/best@2/std:np.float64(0.1466642144887248) - val-aux/sciknoweval/reward/worst@2/mean:np.float64(0.43117619047619044) - val-aux/sciknoweval/reward/worst@2/std:np.float64(0.13917358209541247) - val-aux/sciknoweval/reward/maj@2/mean:np.float64(0.49825714285714284)
step:11 - global_seqlen/min:12444 - global_seqlen/max:16379 - global_seqlen/minmax_diff:3935 - global_seqlen/balanced_min:14789 - global_seqlen/balanced_max:14865 - global_seqlen/mean:14800.625 - actor/entropy:0.28668051958084106 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.378137469291687 - training/rollout_probs_diff_mean:0.006458899471908808 - training/rollout_probs_diff_std:0.015101846307516098 - training/rollout_actor_probs_pearson_corr:0.9980108737945557 - self_distillation/success_group_fraction:0.75 - self_distillation/success_sample_fraction:0.734375 - self_distillation/correct_sample_fraction:0.50390625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.734375 - rollout_corr/rollout_is_mean:0.9999337196350098 - rollout_corr/rollout_is_ratio_fraction_high:2.3902286557131447e-05 - rollout_corr/rollout_is_ratio_fraction_low:0.0001434137229807675 - rollout_corr/rollout_is_max:2.070037603378296 - rollout_corr/rollout_is_min:0.24098028242588043 - rollout_corr/rollout_is_std:0.047629326581954956 - rollout_corr/rollout_is_eff_sample_size:0.9977363445628024 - rollout_corr/rollout_is_seq_mean:0.9997848868370056 - rollout_corr/rollout_is_seq_std:0.004454692825675011 - rollout_corr/rollout_is_seq_max:1.011418342590332 - rollout_corr/rollout_is_seq_min:0.9831334352493286 - rollout_corr/rollout_is_seq_max_deviation:0.016866564750671387 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.3632012782618403) - rollout_corr/training_log_ppl:np.float64(0.30577456433093175) - rollout_corr/kl:np.float64(0.001455732750443417) - rollout_corr/k3_kl:np.float64(0.0012406106384048599) - rollout_corr/rollout_ppl:np.float64(1.3611590014770627) - rollout_corr/rollout_log_ppl:np.float64(0.30431882984703407) - rollout_corr/log_ppl_diff:np.float64(0.001455734483897686) - rollout_corr/log_ppl_abs_diff:np.float64(0.0037959616747684777) - rollout_corr/log_ppl_diff_max:np.float64(0.022277116775512695) - rollout_corr/log_ppl_diff_min:np.float64(-0.009778529405593872) - rollout_corr/ppl_ratio:np.float64(1.0014676633290946) - rollout_corr/chi2_token:np.float64(0.0019878384191542864) - rollout_corr/chi2_seq:np.float64(0.5885442295111716) - actor/pg_loss:np.float64(-0.06822081007703673) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.0805482406867668) - srpo/sdpo_loss:np.float64(0.012327430609730072) - srpo/routed_loss:np.float64(-0.06822081007703673) - srpo/routed_token_count:np.float64(163.42578125) - srpo/correct_sample_fraction:np.float64(0.50390625) - srpo/sdpo_sample_fraction:np.float64(0.24609375) - srpo/grpo_sample_fraction:np.float64(0.75390625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.05149039445677772) - self_distillation/dynamic_weight_mean:np.float64(0.24609375139698386) - self_distillation/dynamic_weight_max:np.float64(1.2955985069274902) - self_distillation/empty_target_batch:np.float64(0.265625) - actor/grad_norm:np.float64(0.37790244817733765) - perf/mfu/actor:np.float64(0.02697562051436024) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(147.03709411621094) - actor/lr:np.float64(5e-06) - training/global_step:11 - training/epoch:0 - critic/score/mean:0.50390625 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.50390625 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.0027158975135535 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.0027158975135535 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:163.42578125 - response_length/max:832.0 - response_length/min:9
step:12 - global_seqlen/min:11748 - global_seqlen/max:15391 - global_seqlen/minmax_diff:3643 - global_seqlen/balanced_min:13327 - global_seqlen/balanced_max:13416 - global_seqlen/mean:13355.125 - actor/entropy:0.23876447975635529 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.41412875056266785 - training/rollout_probs_diff_mean:0.005941879935562611 - training/rollout_probs_diff_std:0.015210720710456371 - training/rollout_actor_probs_pearson_corr:0.9977619647979736 - self_distillation/success_group_fraction:0.78125 - self_distillation/success_sample_fraction:0.76171875 - self_distillation/correct_sample_fraction:0.55078125 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.76171875 - rollout_corr/rollout_is_mean:1.0003182888031006 - rollout_corr/rollout_is_ratio_fraction_high:2.5825778720900416e-05 - rollout_corr/rollout_is_ratio_fraction_low:0.00010330311488360167 - rollout_corr/rollout_is_max:2.0163207054138184 - rollout_corr/rollout_is_min:0.35228511691093445 - rollout_corr/rollout_is_std:0.047272585332393646 - rollout_corr/rollout_is_eff_sample_size:0.9977717095367359 - rollout_corr/rollout_is_seq_mean:1.0003490447998047 - rollout_corr/rollout_is_seq_std:0.004676243755966425 - rollout_corr/rollout_is_seq_max:1.0201698541641235 - rollout_corr/rollout_is_seq_min:0.9847532510757446 - rollout_corr/rollout_is_seq_max_deviation:0.020169854164123535 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.2908877073787153) - rollout_corr/training_log_ppl:np.float64(0.2518373990606051) - rollout_corr/kl:np.float64(0.0008109549114863057) - rollout_corr/k3_kl:np.float64(0.0011599612378176971) - rollout_corr/rollout_ppl:np.float64(1.2897977489046752) - rollout_corr/rollout_log_ppl:np.float64(0.2510264429729432) - rollout_corr/log_ppl_diff:np.float64(0.0008109560876619071) - rollout_corr/log_ppl_abs_diff:np.float64(0.003409657481824979) - rollout_corr/log_ppl_diff_max:np.float64(0.019825011491775513) - rollout_corr/log_ppl_diff_min:np.float64(-0.01660427451133728) - rollout_corr/ppl_ratio:np.float64(1.0008219359442592) - rollout_corr/chi2_token:np.float64(0.003050214843824506) - rollout_corr/chi2_seq:np.float64(0.9592279253993183) - actor/pg_loss:np.float64(-0.050964499707333744) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.06204456492559984) - srpo/sdpo_loss:np.float64(0.0110800652182661) - srpo/routed_loss:np.float64(-0.050964499707333744) - srpo/routed_token_count:np.float64(151.25390625) - srpo/correct_sample_fraction:np.float64(0.55078125) - srpo/sdpo_sample_fraction:np.float64(0.23046875) - srpo/grpo_sample_fraction:np.float64(0.76953125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.038228451667237096) - self_distillation/dynamic_weight_mean:np.float64(0.230468753259629) - self_distillation/dynamic_weight_max:np.float64(1.2636548280715942) - self_distillation/empty_target_batch:np.float64(0.23828125) - actor/grad_norm:np.float64(0.38662582635879517) - perf/mfu/actor:np.float64(0.024355847918892164) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(146.7984161376953) - actor/lr:np.float64(5e-06) - training/global_step:12 - training/epoch:0 - critic/score/mean:0.55078125 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.55078125 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.004952093120664358 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.004952093120664358 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:151.25390625 - response_length/max:835.0 - response
step:13 - global_seqlen/min:10920 - global_seqlen/max:14735 - global_seqlen/minmax_diff:3815 - global_seqlen/balanced_min:13190 - global_seqlen/balanced_max:13191 - global_seqlen/mean:13190.125 - actor/entropy:0.24556629359722137 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.35802704095840454 - training/rollout_probs_diff_mean:0.006480875425040722 - training/rollout_probs_diff_std:0.015921348705887794 - training/rollout_actor_probs_pearson_corr:0.9975414276123047 - self_distillation/success_group_fraction:0.78125 - self_distillation/success_sample_fraction:0.76953125 - self_distillation/correct_sample_fraction:0.54296875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.76953125 - rollout_corr/rollout_is_mean:1.0001860857009888 - rollout_corr/rollout_is_ratio_fraction_high:2.8907583327963948e-05 - rollout_corr/rollout_is_ratio_fraction_low:0.00020235308329574764 - rollout_corr/rollout_is_max:2.064023733139038 - rollout_corr/rollout_is_min:0.29720383882522583 - rollout_corr/rollout_is_std:0.04786650836467743 - rollout_corr/rollout_is_eff_sample_size:0.9977149843835857 - rollout_corr/rollout_is_seq_mean:1.000301480293274 - rollout_corr/rollout_is_seq_std:0.00467937346547842 - rollout_corr/rollout_is_seq_max:1.01895010471344 - rollout_corr/rollout_is_seq_min:0.983443558216095 - rollout_corr/rollout_is_seq_max_deviation:0.01895010471343994 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.2827758030034602) - rollout_corr/training_log_ppl:np.float64(0.2457950550597161) - rollout_corr/kl:np.float64(0.000877036690678068) - rollout_corr/k3_kl:np.float64(0.0011785225366338636) - rollout_corr/rollout_ppl:np.float64(1.28162847366184) - rollout_corr/rollout_log_ppl:np.float64(0.24491801782278344) - rollout_corr/log_ppl_diff:np.float64(0.0008770372369326651) - rollout_corr/log_ppl_abs_diff:np.float64(0.0037955568404868245) - rollout_corr/log_ppl_diff_max:np.float64(0.023132890462875366) - rollout_corr/log_ppl_diff_min:np.float64(-0.015985578298568726) - rollout_corr/ppl_ratio:np.float64(1.00088875554502) - rollout_corr/chi2_token:np.float64(0.0029351285193115473) - rollout_corr/chi2_seq:np.float64(0.7805889812298119) - actor/pg_loss:np.float64(-0.058499566745013) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.07196394269703887) - srpo/sdpo_loss:np.float64(0.013464375952025875) - srpo/routed_loss:np.float64(-0.058499566745013) - srpo/routed_token_count:np.float64(135.12890625) - srpo/correct_sample_fraction:np.float64(0.54296875) - srpo/sdpo_sample_fraction:np.float64(0.23828125) - srpo/grpo_sample_fraction:np.float64(0.76171875) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.04485847149044275) - self_distillation/dynamic_weight_mean:np.float64(0.23828125186264515) - self_distillation/dynamic_weight_max:np.float64(1.3402243852615356) - self_distillation/empty_target_batch:np.float64(0.23046875) - actor/grad_norm:np.float64(0.5218213796615601) - perf/mfu/actor:np.float64(0.024081572137893244) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(146.07122802734375) - actor/lr:np.float64(5e-06) - training/global_step:13 - training/epoch:0 - critic/score/mean:0.54296875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.54296875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.003996473271399736 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.003996473271399736 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:135.12890625 - response_length/max:292.0 - response_length/min:
step:14 - global_seqlen/min:11630 - global_seqlen/max:16624 - global_seqlen/minmax_diff:4994 - global_seqlen/balanced_min:13263 - global_seqlen/balanced_max:13265 - global_seqlen/mean:13264.0 - actor/entropy:0.21762239933013916 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.5948744416236877 - training/rollout_probs_diff_mean:0.006083684973418713 - training/rollout_probs_diff_std:0.016663407906889915 - training/rollout_actor_probs_pearson_corr:0.9970629215240479 - self_distillation/success_group_fraction:0.78125 - self_distillation/success_sample_fraction:0.7734375 - self_distillation/correct_sample_fraction:0.57421875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.7734375 - rollout_corr/rollout_is_mean:1.0001834630966187 - rollout_corr/rollout_is_ratio_fraction_high:7.968550926307216e-05 - rollout_corr/rollout_is_ratio_fraction_low:0.00010624734568409622 - rollout_corr/rollout_is_max:6.223536968231201 - rollout_corr/rollout_is_min:0.2723909616470337 - rollout_corr/rollout_is_std:0.046900425106287 - rollout_corr/rollout_is_eff_sample_size:0.9978055340888018 - rollout_corr/rollout_is_seq_mean:1.0002174377441406 - rollout_corr/rollout_is_seq_std:0.005304012447595596 - rollout_corr/rollout_is_seq_max:1.0482916831970215 - rollout_corr/rollout_is_seq_min:0.9844281077384949 - rollout_corr/rollout_is_seq_max_deviation:0.048291683197021484 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.244126062374562) - rollout_corr/training_log_ppl:np.float64(0.2155887533299392) - rollout_corr/kl:np.float64(0.00100494465818457) - rollout_corr/k3_kl:np.float64(0.001222398752929621) - rollout_corr/rollout_ppl:np.float64(1.242862823419273) - rollout_corr/rollout_log_ppl:np.float64(0.21458380825060885) - rollout_corr/log_ppl_diff:np.float64(0.001004945079330355) - rollout_corr/log_ppl_abs_diff:np.float64(0.0035068701836280525) - rollout_corr/log_ppl_diff_max:np.float64(0.02235954999923706) - rollout_corr/log_ppl_diff_min:np.float64(-0.02297656238079071) - rollout_corr/ppl_ratio:np.float64(1.001016147667542) - rollout_corr/chi2_token:np.float64(0.003429332748055458) - rollout_corr/chi2_seq:np.float64(3.376733945449814) - actor/pg_loss:np.float64(-0.05129040131578222) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.06385141424834728) - srpo/sdpo_loss:np.float64(0.012561012932565063) - srpo/routed_loss:np.float64(-0.05129040131578222) - srpo/routed_token_count:np.float64(147.0625) - srpo/correct_sample_fraction:np.float64(0.57421875) - srpo/sdpo_sample_fraction:np.float64(0.20703125) - srpo/grpo_sample_fraction:np.float64(0.79296875) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.034523157053627074) - self_distillation/dynamic_weight_mean:np.float64(0.20703125093132257) - self_distillation/dynamic_weight_max:np.float64(1.2374225854873657) - self_distillation/empty_target_batch:np.float64(0.2265625) - actor/grad_norm:np.float64(0.49815797805786133) - perf/mfu/actor:np.float64(0.023799144706030853) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(145.8756866455078) - actor/lr:np.float64(5e-06) - training/global_step:14 - training/epoch:0 - critic/score/mean:0.57421875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.57421875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.0018493678653612733 - critic/advantages/max:0.875 - critic/advantages/min:-0.75 - critic/returns/mean:0.0018493678653612733 - critic/returns/max:0.875 - critic/returns/min:-0.75 - response_length/mean:147.0625 - response_length/max:358.0 - response_length/min:93.0 - respon
step:15 - global_seqlen/min:13569 - global_seqlen/max:16301 - global_seqlen/minmax_diff:2732 - global_seqlen/balanced_min:14630 - global_seqlen/balanced_max:14631 - global_seqlen/mean:14630.25 - actor/entropy:0.22992399334907532 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.5268460512161255 - training/rollout_probs_diff_mean:0.006227492354810238 - training/rollout_probs_diff_std:0.016257645562291145 - training/rollout_actor_probs_pearson_corr:0.997356116771698 - self_distillation/success_group_fraction:0.71875 - self_distillation/success_sample_fraction:0.71484375 - self_distillation/correct_sample_fraction:0.55859375 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.71484375 - rollout_corr/rollout_is_mean:0.9997258186340332 - rollout_corr/rollout_is_ratio_fraction_high:4.587787407217547e-05 - rollout_corr/rollout_is_ratio_fraction_low:0.0001376336149405688 - rollout_corr/rollout_is_max:2.654146909713745 - rollout_corr/rollout_is_min:0.23840628564357758 - rollout_corr/rollout_is_std:0.04835771769285202 - rollout_corr/rollout_is_eff_sample_size:0.9976656221352493 - rollout_corr/rollout_is_seq_mean:0.9997716546058655 - rollout_corr/rollout_is_seq_std:0.004038466140627861 - rollout_corr/rollout_is_seq_max:1.0231895446777344 - rollout_corr/rollout_is_seq_min:0.9911590814590454 - rollout_corr/rollout_is_seq_max_deviation:0.023189544677734375 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.2686172099784017) - rollout_corr/training_log_ppl:np.float64(0.23468218676862307) - rollout_corr/kl:np.float64(0.0014550133051969283) - rollout_corr/k3_kl:np.float64(0.0012266793937953935) - rollout_corr/rollout_ppl:np.float64(1.2666818108409643) - rollout_corr/rollout_log_ppl:np.float64(0.23322717219707556) - rollout_corr/log_ppl_diff:np.float64(0.0014550145715475082) - rollout_corr/log_ppl_abs_diff:np.float64(0.0032450900180265307) - rollout_corr/log_ppl_diff_max:np.float64(0.012959711253643036) - rollout_corr/log_ppl_diff_min:np.float64(-0.0180235356092453) - rollout_corr/ppl_ratio:np.float64(1.0014635229017586) - rollout_corr/chi2_token:np.float64(0.0020344809163361788) - rollout_corr/chi2_seq:np.float64(0.7756948107853532) - actor/pg_loss:np.float64(-0.056775170742184855) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.06494994385866448) - srpo/sdpo_loss:np.float64(0.008174773116479628) - srpo/routed_loss:np.float64(-0.056775170742184855) - srpo/routed_token_count:np.float64(170.2890625) - srpo/correct_sample_fraction:np.float64(0.55859375) - srpo/sdpo_sample_fraction:np.float64(0.16015625) - srpo/grpo_sample_fraction:np.float64(0.83984375) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.02742746843432542) - self_distillation/dynamic_weight_mean:np.float64(0.16015625) - self_distillation/dynamic_weight_max:np.float64(1.2870488166809082) - self_distillation/empty_target_batch:np.float64(0.28515625) - actor/grad_norm:np.float64(0.30202457308769226) - perf/mfu/actor:np.float64(0.02258706420405264) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(162.70365142822266) - actor/lr:np.float64(5e-06) - training/global_step:15 - training/epoch:0 - critic/score/mean:0.55859375 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.55859375 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.0013075193855911493 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.0013075193855911493 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:170.2890625 - response_length/max:342.0 - response_length/mi
step:16 - global_seqlen/min:12401 - global_seqlen/max:17479 - global_seqlen/minmax_diff:5078 - global_seqlen/balanced_min:14432 - global_seqlen/balanced_max:14435 - global_seqlen/mean:14434.0 - actor/entropy:0.2233237773180008 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.379741907119751 - training/rollout_probs_diff_mean:0.006004318129271269 - training/rollout_probs_diff_std:0.015533234924077988 - training/rollout_actor_probs_pearson_corr:0.9974988698959351 - self_distillation/success_group_fraction:0.6875 - self_distillation/success_sample_fraction:0.671875 - self_distillation/correct_sample_fraction:0.49609375 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.671875 - rollout_corr/rollout_is_mean:0.9997921586036682 - rollout_corr/rollout_is_ratio_fraction_high:4.586314389598556e-05 - rollout_corr/rollout_is_ratio_fraction_low:2.293157194799278e-05 - rollout_corr/rollout_is_max:3.0816445350646973 - rollout_corr/rollout_is_min:0.4386822283267975 - rollout_corr/rollout_is_std:0.04628831520676613 - rollout_corr/rollout_is_eff_sample_size:0.9978612012388379 - rollout_corr/rollout_is_seq_mean:0.9998055696487427 - rollout_corr/rollout_is_seq_std:0.0037333492655307055 - rollout_corr/rollout_is_seq_max:1.0110981464385986 - rollout_corr/rollout_is_seq_min:0.9854405522346497 - rollout_corr/rollout_is_seq_max_deviation:0.014559447765350342 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.2563437917269766) - rollout_corr/training_log_ppl:np.float64(0.22527961098239757) - rollout_corr/kl:np.float64(0.0012903766853931131) - rollout_corr/k3_kl:np.float64(0.0010959419946061644) - rollout_corr/rollout_ppl:np.float64(1.2547327843494713) - rollout_corr/rollout_log_ppl:np.float64(0.2239892342768144) - rollout_corr/log_ppl_diff:np.float64(0.001290376705583185) - rollout_corr/log_ppl_abs_diff:np.float64(0.003100399684626609) - rollout_corr/log_ppl_diff_max:np.float64(0.017924219369888306) - rollout_corr/log_ppl_diff_min:np.float64(-0.009630560874938965) - rollout_corr/ppl_ratio:np.float64(1.001298324437812) - rollout_corr/chi2_token:np.float64(0.0018203719519078732) - rollout_corr/chi2_seq:np.float64(0.35558420792222023) - actor/pg_loss:np.float64(-0.04338210337300552) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.05323187873000279) - srpo/sdpo_loss:np.float64(0.00984977535699727) - srpo/routed_loss:np.float64(-0.04338210337300552) - srpo/routed_token_count:np.float64(170.34375) - srpo/correct_sample_fraction:np.float64(0.49609375) - srpo/sdpo_sample_fraction:np.float64(0.19140625) - srpo/grpo_sample_fraction:np.float64(0.80859375) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.03241944115143269) - self_distillation/dynamic_weight_mean:np.float64(0.19140624976716936) - self_distillation/dynamic_weight_max:np.float64(1.2760415077209473) - self_distillation/empty_target_batch:np.float64(0.328125) - actor/grad_norm:np.float64(0.2724934220314026) - perf/mfu/actor:np.float64(0.026047289693676673) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(150.52282333374023) - actor/lr:np.float64(5e-06) - training/global_step:16 - training/epoch:0 - critic/score/mean:0.49609375 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.49609375 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.0018029948696494102 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.0018029948696494102 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:170.34375 - response_length/max:300.0 - response_length/min:113.0
step:17 - global_seqlen/min:13846 - global_seqlen/max:20356 - global_seqlen/minmax_diff:6510 - global_seqlen/balanced_min:17171 - global_seqlen/balanced_max:17179 - global_seqlen/mean:17175.5 - actor/entropy:0.24988165497779846 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.230194091796875 - training/rollout_probs_diff_mean:0.005981224589049816 - training/rollout_probs_diff_std:0.014481606893241405 - training/rollout_actor_probs_pearson_corr:0.9979730844497681 - self_distillation/success_group_fraction:0.625 - self_distillation/success_sample_fraction:0.609375 - self_distillation/correct_sample_fraction:0.44140625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.609375 - rollout_corr/rollout_is_mean:0.9998762607574463 - rollout_corr/rollout_is_ratio_fraction_high:3.8387715903809294e-05 - rollout_corr/rollout_is_ratio_fraction_low:1.9193857951904647e-05 - rollout_corr/rollout_is_max:2.46216082572937 - rollout_corr/rollout_is_min:0.49211713671684265 - rollout_corr/rollout_is_std:0.045019954442977905 - rollout_corr/rollout_is_eff_sample_size:0.9979768283435333 - rollout_corr/rollout_is_seq_mean:0.9998485445976257 - rollout_corr/rollout_is_seq_std:0.0033213337883353233 - rollout_corr/rollout_is_seq_max:1.0103394985198975 - rollout_corr/rollout_is_seq_min:0.9905241131782532 - rollout_corr/rollout_is_seq_max_deviation:0.010339498519897461 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.3067224454134703) - rollout_corr/training_log_ppl:np.float64(0.2639249793573981) - rollout_corr/kl:np.float64(0.0012190569805170526) - rollout_corr/k3_kl:np.float64(0.0010675782360749508) - rollout_corr/rollout_ppl:np.float64(1.3051045737229288) - rollout_corr/rollout_log_ppl:np.float64(0.262705921268207) - rollout_corr/log_ppl_diff:np.float64(0.001219058089191094) - rollout_corr/log_ppl_abs_diff:np.float64(0.002767152473097667) - rollout_corr/log_ppl_diff_max:np.float64(0.011501610279083252) - rollout_corr/log_ppl_diff_min:np.float64(-0.0075855553150177) - rollout_corr/ppl_ratio:np.float64(1.001225349958986) - rollout_corr/chi2_token:np.float64(0.0018664661329239607) - rollout_corr/chi2_seq:np.float64(0.32409870927222073) - actor/pg_loss:np.float64(-0.042099951147974934) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.05026186932809651) - srpo/sdpo_loss:np.float64(0.008161918180121575) - srpo/routed_loss:np.float64(-0.042099951147974934) - srpo/routed_token_count:np.float64(203.515625) - srpo/correct_sample_fraction:np.float64(0.44140625) - srpo/sdpo_sample_fraction:np.float64(0.18359375) - srpo/grpo_sample_fraction:np.float64(0.81640625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.04136811778880656) - self_distillation/dynamic_weight_mean:np.float64(0.18359374930150807) - self_distillation/dynamic_weight_max:np.float64(1.319330096244812) - self_distillation/empty_target_batch:np.float64(0.390625) - actor/grad_norm:np.float64(0.264835923910141) - perf/mfu/actor:np.float64(0.030680730668686915) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(147.25261211395264) - actor/lr:np.float64(5e-06) - training/global_step:17 - training/epoch:0 - critic/score/mean:0.44140625 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.44140625 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.002308061346411705 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.002308061346411705 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:203.515625 - response_length/max:806.0 - response_length/min:127.0
step:18 - global_seqlen/min:12834 - global_seqlen/max:17258 - global_seqlen/minmax_diff:4424 - global_seqlen/balanced_min:14834 - global_seqlen/balanced_max:14835 - global_seqlen/mean:14834.75 - actor/entropy:0.2508573532104492 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.2995908856391907 - training/rollout_probs_diff_mean:0.005982003640383482 - training/rollout_probs_diff_std:0.014694152399897575 - training/rollout_actor_probs_pearson_corr:0.9979224801063538 - self_distillation/success_group_fraction:0.78125 - self_distillation/success_sample_fraction:0.7734375 - self_distillation/correct_sample_fraction:0.53515625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.7734375 - rollout_corr/rollout_is_mean:1.0001429319381714 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.9872629642486572 - rollout_corr/rollout_is_min:0.5352635979652405 - rollout_corr/rollout_is_std:0.044811610132455826 - rollout_corr/rollout_is_eff_sample_size:0.9979964187516931 - rollout_corr/rollout_is_seq_mean:1.0001869201660156 - rollout_corr/rollout_is_seq_std:0.0035082194954156876 - rollout_corr/rollout_is_seq_max:1.0119105577468872 - rollout_corr/rollout_is_seq_min:0.9922018647193909 - rollout_corr/rollout_is_seq_max_deviation:0.011910557746887207 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.2883695508353412) - rollout_corr/training_log_ppl:np.float64(0.2501904987148009) - rollout_corr/kl:np.float64(0.0008023567795927278) - rollout_corr/k3_kl:np.float64(0.0009892383238252478) - rollout_corr/rollout_ppl:np.float64(1.287324839271605) - rollout_corr/rollout_log_ppl:np.float64(0.24938814173219725) - rollout_corr/log_ppl_diff:np.float64(0.0008023569826036692) - rollout_corr/log_ppl_abs_diff:np.float64(0.0028190017328597605) - rollout_corr/log_ppl_diff_max:np.float64(0.009036228060722351) - rollout_corr/log_ppl_diff_min:np.float64(-0.009636253118515015) - rollout_corr/ppl_ratio:np.float64(1.000808710930869) - rollout_corr/chi2_token:np.float64(0.0023844758979976177) - rollout_corr/chi2_seq:np.float64(0.6987090280745178) - actor/pg_loss:np.float64(-0.06903548053378472) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.08153472898993641) - srpo/sdpo_loss:np.float64(0.01249924845615169) - srpo/routed_loss:np.float64(-0.06903548053378472) - srpo/routed_token_count:np.float64(190.3046875) - srpo/correct_sample_fraction:np.float64(0.53515625) - srpo/sdpo_sample_fraction:np.float64(0.24609375) - srpo/grpo_sample_fraction:np.float64(0.75390625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.04811318687279709) - self_distillation/dynamic_weight_mean:np.float64(0.2460937495343387) - self_distillation/dynamic_weight_max:np.float64(1.2742080688476562) - self_distillation/empty_target_batch:np.float64(0.2265625) - actor/grad_norm:np.float64(0.34010589122772217) - perf/mfu/actor:np.float64(0.02679017243579962) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(146.73824310302734) - actor/lr:np.float64(5e-06) - training/global_step:18 - training/epoch:0 - critic/score/mean:0.53515625 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.53515625 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.0006157887983135879 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.0006157887983135879 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:190.3046875 - response_length/max:422.0 - response_length/min:117.0 - response_length/clip_ra
step:19 - global_seqlen/min:12977 - global_seqlen/max:19848 - global_seqlen/minmax_diff:6871 - global_seqlen/balanced_min:15708 - global_seqlen/balanced_max:15711 - global_seqlen/mean:15709.25 - actor/entropy:0.23554760217666626 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.32848986983299255 - training/rollout_probs_diff_mean:0.00590132363140583 - training/rollout_probs_diff_std:0.014728233218193054 - training/rollout_actor_probs_pearson_corr:0.9978373050689697 - self_distillation/success_group_fraction:0.78125 - self_distillation/success_sample_fraction:0.76953125 - self_distillation/correct_sample_fraction:0.62890625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.76953125 - rollout_corr/rollout_is_mean:1.0001251697540283 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:2.069108268187847e-05 - rollout_corr/rollout_is_max:1.965978980064392 - rollout_corr/rollout_is_min:0.47117674350738525 - rollout_corr/rollout_is_std:0.044738393276929855 - rollout_corr/rollout_is_eff_sample_size:0.998002949058664 - rollout_corr/rollout_is_seq_mean:1.0001322031021118 - rollout_corr/rollout_is_seq_std:0.0035070704761892557 - rollout_corr/rollout_is_seq_max:1.0112321376800537 - rollout_corr/rollout_is_seq_min:0.990557849407196 - rollout_corr/rollout_is_seq_max_deviation:0.011232137680053711 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.2734890608116984) - rollout_corr/training_log_ppl:np.float64(0.23948266039951704) - rollout_corr/kl:np.float64(0.0008788166518414187) - rollout_corr/k3_kl:np.float64(0.0010110637919922283) - rollout_corr/rollout_ppl:np.float64(1.2723404876887798) - rollout_corr/rollout_log_ppl:np.float64(0.23860384317231365) - rollout_corr/log_ppl_diff:np.float64(0.0008788172272033989) - rollout_corr/log_ppl_abs_diff:np.float64(0.0027745234547182918) - rollout_corr/log_ppl_diff_max:np.float64(0.01143936812877655) - rollout_corr/log_ppl_diff_min:np.float64(-0.008928149938583374) - rollout_corr/ppl_ratio:np.float64(1.0008853829931468) - rollout_corr/chi2_token:np.float64(0.0023030273150652647) - rollout_corr/chi2_seq:np.float64(0.7266433585900813) - actor/pg_loss:np.float64(-0.044986527114815544) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.0522527985740453) - srpo/sdpo_loss:np.float64(0.007266271459229756) - srpo/routed_loss:np.float64(-0.044986527114815544) - srpo/routed_token_count:np.float64(188.7890625) - srpo/correct_sample_fraction:np.float64(0.62890625) - srpo/sdpo_sample_fraction:np.float64(0.15234375) - srpo/grpo_sample_fraction:np.float64(0.84765625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.029183782055042684) - self_distillation/dynamic_weight_mean:np.float64(0.15234375186264515) - self_distillation/dynamic_weight_max:np.float64(1.3182636499404907) - self_distillation/empty_target_batch:np.float64(0.23046875) - actor/grad_norm:np.float64(0.2474699318408966) - perf/mfu/actor:np.float64(0.027955120314738714) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(146.50935745239258) - actor/lr:np.float64(5e-06) - training/global_step:19 - training/epoch:0 - critic/score/mean:0.62890625 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.62890625 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.002133767819032073 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.002133767819032073 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:188.7890625 - response_length/max:368.0 - response_length/min:108.0 -
test_gen_batch meta info: {'eos_token_id': 151645, 'pad_token_id': 151643, 'recompute_log_prob': False, 'do_sample': True, 'validate': True, 'global_steps': 20}
validation generation end
[prompt] system

Given a question and four options, please select the right answer. Respond in the following format:
<reasoning>
...
</reasoning>
<answer>
...
</answer>

For the answer, only output the letter corresponding to the correct option (A, B, C, or D), and nothing else. Do not restate the answer text. For example, if the answer is "A", just output:
<answer>
A
</answer>

user
Which of the following molecules is the only correct reactant used in the synthesis of the product "Nc1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1" ?

A: O=C1c2ccccc2C(=O)N1c1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1
B: O=C1c2ccccc2C(Cl)N1c1ccc2ccc(Cl)nc2n1
C: O=C1c2ccccc2C(=O)N1Cc1ccc(Oc2ccc(Cl)c(C(F)(F)F)c2)cc1
D: O=C1c2ccccc2C(=O)N1CC=C(Cc1ccccc1)C(F)(F)F
Please reason step by step.
assistant
<think>

</think>


[response] <reasoning>
The molecule "Nc1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1" contains a nitrogen atom bonded to a carbon chain, which is connected to a chlorinated aromatic ring and a trifluoromethyl group. The correct reactant must include these key functional groups. Option A includes a carboxylic acid group, a nitrogen, and the necessary chlorinated and fluorinated groups, making it the most structurally accurate. Option B lacks the fluorinated group and has a different arrangement of substituents. Option C includes a hydroxyl group and a different structure, which does not match the product. Option D has a different arrangement and lacks the necessary chlorinated and fluorinated groups. Therefore, Option A is the correct reactant.
</reasoning>
<answer>
A
</answer>
[ground_truth] A
[score] 1.0
[acc] 1.0
[pred] A
[incorrect_format] 1
[feedback]
local_global_step_folder: /workspace/L2T/checkpoints/datasets/sciknoweval/chemistry/qwen3gen-chemistry-SRPO_TR-Qwen-Qwen3-4B-mbs32-tr0.1-dwtrue-train32-rollout8-lr5e-6-vllm0.8/global_step_20
step:20 - global_seqlen/min:13857 - global_seqlen/max:16336 - global_seqlen/minmax_diff:2479 - global_seqlen/balanced_min:15337 - global_seqlen/balanced_max:15519 - global_seqlen/mean:15364.75 - actor/entropy:0.22814059257507324 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.42996495962142944 - training/rollout_probs_diff_mean:0.005766410380601883 - training/rollout_probs_diff_std:0.01470246259123087 - training/rollout_actor_probs_pearson_corr:0.9977887272834778 - self_distillation/success_group_fraction:0.78125 - self_distillation/success_sample_fraction:0.78125 - self_distillation/correct_sample_fraction:0.6796875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.78125 - rollout_corr/rollout_is_mean:0.9998474717140198 - rollout_corr/rollout_is_ratio_fraction_high:1.9873603378073312e-05 - rollout_corr/rollout_is_ratio_fraction_low:9.936802234733477e-05 - rollout_corr/rollout_is_max:2.4575564861297607 - rollout_corr/rollout_is_min:0.2921242415904999 - rollout_corr/rollout_is_std:0.044810280203819275 - rollout_corr/rollout_is_eff_sample_size:0.9979954688959803 - rollout_corr/rollout_is_seq_mean:0.9998536705970764 - rollout_corr/rollout_is_seq_std:0.003364948322996497 - rollout_corr/rollout_is_seq_max:1.0086809396743774 - rollout_corr/rollout_is_seq_min:0.9870060086250305 - rollout_corr/rollout_is_seq_max_deviation:0.012993991374969482 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.273806983139366) - rollout_corr/training_log_ppl:np.float64(0.23885453169350512) - rollout_corr/kl:np.float64(0.0011725538762092924) - rollout_corr/k3_kl:np.float64(0.0010262332477850578) - rollout_corr/rollout_ppl:np.float64(1.272272175643593) - rollout_corr/rollout_log_ppl:np.float64(0.23768197777098976) - rollout_corr/log_ppl_diff:np.float64(0.0011725539225153625) - rollout_corr/log_ppl_abs_diff:np.float64(0.0027502442826516926) - rollout_corr/log_ppl_diff_max:np.float64(0.01584485173225403) - rollout_corr/log_ppl_diff_min:np.float64(-0.007140606641769409) - rollout_corr/ppl_ratio:np.float64(1.001179120503366) - rollout_corr/chi2_token:np.float64(0.001748459180817008) - rollout_corr/chi2_seq:np.float64(0.37529285345226526) - actor/pg_loss:np.float64(-0.04677329720288981) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.052706683782162145) - srpo/sdpo_loss:np.float64(0.005933386579272337) - srpo/routed_loss:np.float64(-0.04677329720288981) - srpo/routed_token_count:np.float64(196.5546875) - srpo/correct_sample_fraction:np.float64(0.6796875) - srpo/sdpo_sample_fraction:np.float64(0.1015625) - srpo/grpo_sample_fraction:np.float64(0.8984375) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.021367511653807014) - self_distillation/dynamic_weight_mean:np.float64(0.10156250186264515) - self_distillation/dynamic_weight_max:np.float64(1.2766902446746826) - self_distillation/empty_target_batch:np.float64(0.21875) - actor/grad_norm:np.float64(0.208760604262352) - perf/mfu/actor:np.float64(0.02772251511657539) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(146.2978596687317) - actor/lr:np.float64(5e-06) - val-aux/sciknoweval/reward/mean@16:np.float64(0.6404761904761904) - val-aux/sciknoweval/reward/std@16:np.float64(0.1366981177485994) - val-aux/sciknoweval/reward/best@2/mean:np.float64(0.694952380952381) - val-aux/sciknoweval/reward/best@2/std:np.float64(0.10576726925529584) - val-aux/sciknoweval/reward/worst@2/mean:np.float64(0.5854857142857143) - val-aux/sciknoweval/reward/worst@2/std:np.float64(0.1255756496866224) - val-aux/sciknoweval/reward/maj@2/mean:np.float64(0.6398761904761903) - val-a
step:21 - global_seqlen/min:13054 - global_seqlen/max:17528 - global_seqlen/minmax_diff:4474 - global_seqlen/balanced_min:15132 - global_seqlen/balanced_max:15133 - global_seqlen/mean:15132.125 - actor/entropy:0.23250436782836914 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.2978816628456116 - training/rollout_probs_diff_mean:0.005548770539462566 - training/rollout_probs_diff_std:0.014246656559407711 - training/rollout_actor_probs_pearson_corr:0.9979720711708069 - self_distillation/success_group_fraction:0.65625 - self_distillation/success_sample_fraction:0.65234375 - self_distillation/correct_sample_fraction:0.45703125 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.65234375 - rollout_corr/rollout_is_mean:1.0001723766326904 - rollout_corr/rollout_is_ratio_fraction_high:2.0592657165252604e-05 - rollout_corr/rollout_is_ratio_fraction_low:2.0592657165252604e-05 - rollout_corr/rollout_is_max:3.2374002933502197 - rollout_corr/rollout_is_min:0.4690566062927246 - rollout_corr/rollout_is_std:0.043473485857248306 - rollout_corr/rollout_is_eff_sample_size:0.9981142148743221 - rollout_corr/rollout_is_seq_mean:1.0002150535583496 - rollout_corr/rollout_is_seq_std:0.0035405426751822233 - rollout_corr/rollout_is_seq_max:1.0139319896697998 - rollout_corr/rollout_is_seq_min:0.9879990220069885 - rollout_corr/rollout_is_seq_max_deviation:0.013931989669799805 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.270667006727308) - rollout_corr/training_log_ppl:np.float64(0.23694895819062367) - rollout_corr/kl:np.float64(0.00074582341378715) - rollout_corr/k3_kl:np.float64(0.0009608740459725595) - rollout_corr/rollout_ppl:np.float64(1.2696765228174627) - rollout_corr/rollout_log_ppl:np.float64(0.23620313295396045) - rollout_corr/log_ppl_diff:np.float64(0.0007458252366632223) - rollout_corr/log_ppl_abs_diff:np.float64(0.0027017995598725975) - rollout_corr/log_ppl_diff_max:np.float64(0.01426425576210022) - rollout_corr/log_ppl_diff_min:np.float64(-0.009768635034561157) - rollout_corr/ppl_ratio:np.float64(1.0007520546205342) - rollout_corr/chi2_token:np.float64(0.0024234738666564226) - rollout_corr/chi2_seq:np.float64(0.5251373571809381) - actor/pg_loss:np.float64(-0.07752192272891989) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.0874040543276351) - srpo/sdpo_loss:np.float64(0.009882131598715205) - srpo/routed_loss:np.float64(-0.07752192272891989) - srpo/routed_token_count:np.float64(189.69140625) - srpo/correct_sample_fraction:np.float64(0.45703125) - srpo/sdpo_sample_fraction:np.float64(0.19921875) - srpo/grpo_sample_fraction:np.float64(0.80078125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.03428851210628636) - self_distillation/dynamic_weight_mean:np.float64(0.19921875093132257) - self_distillation/dynamic_weight_max:np.float64(1.234755039215088) - self_distillation/empty_target_batch:np.float64(0.34765625) - actor/grad_norm:np.float64(0.2946995496749878) - perf/mfu/actor:np.float64(0.027533141555892468) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(147.7117042541504) - actor/lr:np.float64(5e-06) - training/global_step:21 - training/epoch:0 - critic/score/mean:0.45703125 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.45703125 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.003184139495715499 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.003184139495715499 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:189.69140625 - response_length/max:440.0 - response
step:22 - global_seqlen/min:12249 - global_seqlen/max:15880 - global_seqlen/minmax_diff:3631 - global_seqlen/balanced_min:13558 - global_seqlen/balanced_max:13560 - global_seqlen/mean:13559.125 - actor/entropy:0.23321504890918732 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.22432908415794373 - training/rollout_probs_diff_mean:0.005397323053330183 - training/rollout_probs_diff_std:0.013442481867969036 - training/rollout_actor_probs_pearson_corr:0.9981771111488342 - self_distillation/success_group_fraction:0.625 - self_distillation/success_sample_fraction:0.62109375 - self_distillation/correct_sample_fraction:0.45703125 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.62109375 - rollout_corr/rollout_is_mean:0.999760091304779 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:4.6458688302664086e-05 - rollout_corr/rollout_is_max:1.8703174591064453 - rollout_corr/rollout_is_min:0.4762100875377655 - rollout_corr/rollout_is_std:0.04148738831281662 - rollout_corr/rollout_is_eff_sample_size:0.9982809819240698 - rollout_corr/rollout_is_seq_mean:0.9997053742408752 - rollout_corr/rollout_is_seq_std:0.00317194452509284 - rollout_corr/rollout_is_seq_max:1.010477900505066 - rollout_corr/rollout_is_seq_min:0.9896388649940491 - rollout_corr/rollout_is_seq_max_deviation:0.010477900505065918 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.269182104151696) - rollout_corr/training_log_ppl:np.float64(0.23521971036097966) - rollout_corr/kl:np.float64(0.0011776632664028952) - rollout_corr/k3_kl:np.float64(0.0008830840325231293) - rollout_corr/rollout_ppl:np.float64(1.2675996227189898) - rollout_corr/rollout_log_ppl:np.float64(0.23404204478720203) - rollout_corr/log_ppl_diff:np.float64(0.001177665573777631) - rollout_corr/log_ppl_abs_diff:np.float64(0.002638570760609582) - rollout_corr/log_ppl_diff_max:np.float64(0.012041032314300537) - rollout_corr/log_ppl_diff_min:np.float64(-0.008604377508163452) - rollout_corr/ppl_ratio:np.float64(1.001183545216918) - rollout_corr/chi2_token:np.float64(0.0011474080383777618) - rollout_corr/chi2_seq:np.float64(0.15597240440547466) - actor/pg_loss:np.float64(-0.06394396480027353) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.07179310562787578) - srpo/sdpo_loss:np.float64(0.007849140827602241) - srpo/routed_loss:np.float64(-0.06394396480027353) - srpo/routed_token_count:np.float64(168.16015625) - srpo/correct_sample_fraction:np.float64(0.45703125) - srpo/sdpo_sample_fraction:np.float64(0.16796875) - srpo/grpo_sample_fraction:np.float64(0.83203125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.02643726542009972) - self_distillation/dynamic_weight_mean:np.float64(0.16796874930150807) - self_distillation/dynamic_weight_max:np.float64(1.2064497470855713) - self_distillation/empty_target_batch:np.float64(0.37890625) - actor/grad_norm:np.float64(0.24736884236335754) - perf/mfu/actor:np.float64(0.02495573011502593) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(147.5552520751953) - actor/lr:np.float64(5e-06) - training/global_step:22 - training/epoch:0 - critic/score/mean:0.45703125 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.45703125 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.003231782466173172 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.003231782466173172 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:168.16015625 - response_length/max:310.0 - response_length/min:104.0 - resp
step:23 - global_seqlen/min:11644 - global_seqlen/max:18796 - global_seqlen/minmax_diff:7152 - global_seqlen/balanced_min:14362 - global_seqlen/balanced_max:14364 - global_seqlen/mean:14363.125 - actor/entropy:0.24326016008853912 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.18455862998962402 - training/rollout_probs_diff_mean:0.005476845894008875 - training/rollout_probs_diff_std:0.013478978537023067 - training/rollout_actor_probs_pearson_corr:0.9982510805130005 - self_distillation/success_group_fraction:0.875 - self_distillation/success_sample_fraction:0.8671875 - self_distillation/correct_sample_fraction:0.7421875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.8671875 - rollout_corr/rollout_is_mean:1.0000938177108765 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:2.2826359781902283e-05 - rollout_corr/rollout_is_max:1.812645435333252 - rollout_corr/rollout_is_min:0.1213831678032875 - rollout_corr/rollout_is_std:0.04193537309765816 - rollout_corr/rollout_is_eff_sample_size:0.9982448680925968 - rollout_corr/rollout_is_seq_mean:1.0001449584960938 - rollout_corr/rollout_is_seq_std:0.0031712648924440145 - rollout_corr/rollout_is_seq_max:1.008840560913086 - rollout_corr/rollout_is_seq_min:0.9912358522415161 - rollout_corr/rollout_is_seq_max_deviation:0.008840560913085938 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.291250050533563) - rollout_corr/training_log_ppl:np.float64(0.25189825012057554) - rollout_corr/kl:np.float64(0.000751608556935679) - rollout_corr/k3_kl:np.float64(0.0008966185441181551) - rollout_corr/rollout_ppl:np.float64(1.290232315659523) - rollout_corr/rollout_log_ppl:np.float64(0.25114664020657074) - rollout_corr/log_ppl_diff:np.float64(0.0007516099140048027) - rollout_corr/log_ppl_abs_diff:np.float64(0.0024888557964004576) - rollout_corr/log_ppl_diff_max:np.float64(0.012800097465515137) - rollout_corr/log_ppl_diff_min:np.float64(-0.007279247045516968) - rollout_corr/ppl_ratio:np.float64(1.0007570697925985) - rollout_corr/chi2_token:np.float64(0.002067605499178171) - rollout_corr/chi2_seq:np.float64(0.3450325191952288) - actor/pg_loss:np.float64(-0.028455271989514586) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.03513297243625857) - srpo/sdpo_loss:np.float64(0.006677700446743984) - srpo/routed_loss:np.float64(-0.028455271989514586) - srpo/routed_token_count:np.float64(171.12890625) - srpo/correct_sample_fraction:np.float64(0.7421875) - srpo/sdpo_sample_fraction:np.float64(0.1328125) - srpo/grpo_sample_fraction:np.float64(0.8671875) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.022638513008132577) - self_distillation/dynamic_weight_mean:np.float64(0.1328124995343387) - self_distillation/dynamic_weight_max:np.float64(1.2379279136657715) - self_distillation/empty_target_batch:np.float64(0.1328125) - actor/grad_norm:np.float64(0.25471702218055725) - perf/mfu/actor:np.float64(0.02639344277378026) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(147.0048942565918) - actor/lr:np.float64(5e-06) - training/global_step:23 - training/epoch:0 - critic/score/mean:0.7421875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.7421875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.00017975758237298578 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.00017975758237298578 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:171.12890625 - response_length/max:339.0 - response_length/min:104.0 - respon
step:24 - global_seqlen/min:13003 - global_seqlen/max:16506 - global_seqlen/minmax_diff:3503 - global_seqlen/balanced_min:14332 - global_seqlen/balanced_max:14333 - global_seqlen/mean:14332.375 - actor/entropy:0.24030952155590057 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.3759361803531647 - training/rollout_probs_diff_mean:0.0055581750348210335 - training/rollout_probs_diff_std:0.014044037088751793 - training/rollout_actor_probs_pearson_corr:0.9980523586273193 - self_distillation/success_group_fraction:0.84375 - self_distillation/success_sample_fraction:0.83984375 - self_distillation/correct_sample_fraction:0.703125 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.83984375 - rollout_corr/rollout_is_mean:0.9999313950538635 - rollout_corr/rollout_is_ratio_fraction_high:6.995779403951019e-05 - rollout_corr/rollout_is_ratio_fraction_low:0.00013991558807902038 - rollout_corr/rollout_is_max:2.5239269733428955 - rollout_corr/rollout_is_min:0.37889376282691956 - rollout_corr/rollout_is_std:0.0425337515771389 - rollout_corr/rollout_is_eff_sample_size:0.9981937904899303 - rollout_corr/rollout_is_seq_mean:0.99990314245224 - rollout_corr/rollout_is_seq_std:0.003512876806780696 - rollout_corr/rollout_is_seq_max:1.0126826763153076 - rollout_corr/rollout_is_seq_min:0.9887622594833374 - rollout_corr/rollout_is_seq_max_deviation:0.012682676315307617 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.2754703140817583) - rollout_corr/training_log_ppl:np.float64(0.24036350613459945) - rollout_corr/kl:np.float64(0.0010113500392066044) - rollout_corr/k3_kl:np.float64(0.0009145264096446226) - rollout_corr/rollout_ppl:np.float64(1.274109295103699) - rollout_corr/rollout_log_ppl:np.float64(0.2393521537014749) - rollout_corr/log_ppl_diff:np.float64(0.0010113524331245571) - rollout_corr/log_ppl_abs_diff:np.float64(0.002851540135452524) - rollout_corr/log_ppl_diff_max:np.float64(0.013599008321762085) - rollout_corr/log_ppl_diff_min:np.float64(-0.010319262742996216) - rollout_corr/ppl_ratio:np.float64(1.0010180643294007) - rollout_corr/chi2_token:np.float64(0.0016518519259989262) - rollout_corr/chi2_seq:np.float64(0.4878803454339504) - actor/pg_loss:np.float64(-0.053963682577887084) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.06159844534704462) - srpo/sdpo_loss:np.float64(0.007634762769157533) - srpo/routed_loss:np.float64(-0.053963682577887084) - srpo/routed_token_count:np.float64(167.51171875) - srpo/correct_sample_fraction:np.float64(0.703125) - srpo/sdpo_sample_fraction:np.float64(0.140625) - srpo/grpo_sample_fraction:np.float64(0.859375) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.021844514267286286) - self_distillation/dynamic_weight_mean:np.float64(0.14062500069849193) - self_distillation/dynamic_weight_max:np.float64(1.1977055072784424) - self_distillation/empty_target_batch:np.float64(0.16015625) - actor/grad_norm:np.float64(0.2729068696498871) - perf/mfu/actor:np.float64(0.026355368480914146) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(146.63806533813477) - actor/lr:np.float64(5e-06) - training/global_step:24 - training/epoch:0 - critic/score/mean:0.703125 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.703125 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.0069199916906654835 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.0069199916906654835 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:167.51171875 - response_length/max:355.0 - response_length
step:25 - global_seqlen/min:12220 - global_seqlen/max:17389 - global_seqlen/minmax_diff:5169 - global_seqlen/balanced_min:15077 - global_seqlen/balanced_max:15078 - global_seqlen/mean:15077.5 - actor/entropy:0.22903265058994293 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.24103376269340515 - training/rollout_probs_diff_mean:0.004953456576913595 - training/rollout_probs_diff_std:0.012582902796566486 - training/rollout_actor_probs_pearson_corr:0.9983879327774048 - self_distillation/success_group_fraction:0.875 - self_distillation/success_sample_fraction:0.8671875 - self_distillation/correct_sample_fraction:0.6796875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.8671875 - rollout_corr/rollout_is_mean:1.0001150369644165 - rollout_corr/rollout_is_ratio_fraction_high:2.3193246306618676e-05 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:2.016805410385132 - rollout_corr/rollout_is_min:0.5369345545768738 - rollout_corr/rollout_is_std:0.038726892322301865 - rollout_corr/rollout_is_eff_sample_size:0.9985029492863209 - rollout_corr/rollout_is_seq_mean:1.0000851154327393 - rollout_corr/rollout_is_seq_std:0.00314034684561193 - rollout_corr/rollout_is_seq_max:1.0106863975524902 - rollout_corr/rollout_is_seq_min:0.989309549331665 - rollout_corr/rollout_is_seq_max_deviation:0.010690450668334961 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.2655659476295114) - rollout_corr/training_log_ppl:np.float64(0.23187048602267168) - rollout_corr/kl:np.float64(0.0006667741857748766) - rollout_corr/k3_kl:np.float64(0.0007518027360902124) - rollout_corr/rollout_ppl:np.float64(1.2646738542243838) - rollout_corr/rollout_log_ppl:np.float64(0.2312037106457865) - rollout_corr/log_ppl_diff:np.float64(0.0006667753768851981) - rollout_corr/log_ppl_abs_diff:np.float64(0.0025084516528295353) - rollout_corr/log_ppl_diff_max:np.float64(0.012062400579452515) - rollout_corr/log_ppl_diff_min:np.float64(-0.007304131984710693) - rollout_corr/ppl_ratio:np.float64(1.0006719822995365) - rollout_corr/chi2_token:np.float64(0.001688110874965787) - rollout_corr/chi2_seq:np.float64(0.33377147698774934) - actor/pg_loss:np.float64(-0.06762910270845168) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.07614822205505334) - srpo/sdpo_loss:np.float64(0.008519119346601656) - srpo/routed_loss:np.float64(-0.06762910270845168) - srpo/routed_token_count:np.float64(168.421875) - srpo/correct_sample_fraction:np.float64(0.6796875) - srpo/sdpo_sample_fraction:np.float64(0.1953125) - srpo/grpo_sample_fraction:np.float64(0.8046875) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.02724976747413166) - self_distillation/dynamic_weight_mean:np.float64(0.19531250349245965) - self_distillation/dynamic_weight_max:np.float64(1.209930658340454) - self_distillation/empty_target_batch:np.float64(0.1328125) - actor/grad_norm:np.float64(0.28091341257095337) - perf/mfu/actor:np.float64(0.02765784987825978) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(146.6257209777832) - actor/lr:np.float64(5e-06) - training/global_step:25 - training/epoch:0 - critic/score/mean:0.6796875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.6796875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.0017365942476317286 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.0017365942476317286 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:168.421875 - response_length/max:537.0 - response_length/min:97.0 - response_length/
step:26 - global_seqlen/min:12497 - global_seqlen/max:16694 - global_seqlen/minmax_diff:4197 - global_seqlen/balanced_min:14579 - global_seqlen/balanced_max:14580 - global_seqlen/mean:14579.75 - actor/entropy:0.23520025610923767 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.23996931314468384 - training/rollout_probs_diff_mean:0.004998512100428343 - training/rollout_probs_diff_std:0.01255847979336977 - training/rollout_actor_probs_pearson_corr:0.9984226226806641 - self_distillation/success_group_fraction:0.8125 - self_distillation/success_sample_fraction:0.80078125 - self_distillation/correct_sample_fraction:0.64453125 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.80078125 - rollout_corr/rollout_is_mean:0.9998680353164673 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:2.4535060219932348e-05 - rollout_corr/rollout_is_max:1.8261206150054932 - rollout_corr/rollout_is_min:0.48056796193122864 - rollout_corr/rollout_is_std:0.03780318796634674 - rollout_corr/rollout_is_eff_sample_size:0.9985724829178868 - rollout_corr/rollout_is_seq_mean:0.999844491481781 - rollout_corr/rollout_is_seq_std:0.0032007568515837193 - rollout_corr/rollout_is_seq_max:1.0089396238327026 - rollout_corr/rollout_is_seq_min:0.9900839328765869 - rollout_corr/rollout_is_seq_max_deviation:0.009916067123413086 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.2724189613945782) - rollout_corr/training_log_ppl:np.float64(0.23696001572534442) - rollout_corr/kl:np.float64(0.0008851287797995155) - rollout_corr/k3_kl:np.float64(0.0007296326811854215) - rollout_corr/rollout_ppl:np.float64(1.2712081945501268) - rollout_corr/rollout_log_ppl:np.float64(0.23607488459674641) - rollout_corr/log_ppl_diff:np.float64(0.0008851311285980046) - rollout_corr/log_ppl_abs_diff:np.float64(0.0025998828932642937) - rollout_corr/log_ppl_diff_max:np.float64(0.011183828115463257) - rollout_corr/log_ppl_diff_min:np.float64(-0.00798746943473816) - rollout_corr/ppl_ratio:np.float64(1.0008909134194255) - rollout_corr/chi2_token:np.float64(0.0011250425595790148) - rollout_corr/chi2_seq:np.float64(0.2334622663911432) - actor/pg_loss:np.float64(-0.051130082276358735) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.05902399751357734) - srpo/sdpo_loss:np.float64(0.007893915237218607) - srpo/routed_loss:np.float64(-0.051130082276358735) - srpo/routed_token_count:np.float64(159.2109375) - srpo/correct_sample_fraction:np.float64(0.64453125) - srpo/sdpo_sample_fraction:np.float64(0.16796875) - srpo/grpo_sample_fraction:np.float64(0.83203125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.025004166338476352) - self_distillation/dynamic_weight_mean:np.float64(0.1679687537252903) - self_distillation/dynamic_weight_max:np.float64(1.2456731796264648) - self_distillation/empty_target_batch:np.float64(0.19921875) - actor/grad_norm:np.float64(0.24266968667507172) - perf/mfu/actor:np.float64(0.026878699834065888) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(146.4528045654297) - actor/lr:np.float64(5e-06) - training/global_step:26 - training/epoch:0 - critic/score/mean:0.64453125 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.64453125 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.00037722656270489097 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.00037722656270489097 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:159.2109375 - response_length/max:384.0 - response_length/min:
step:27 - global_seqlen/min:11458 - global_seqlen/max:15162 - global_seqlen/minmax_diff:3704 - global_seqlen/balanced_min:13518 - global_seqlen/balanced_max:13549 - global_seqlen/mean:13522.375 - actor/entropy:0.2159886658191681 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.18462896347045898 - training/rollout_probs_diff_mean:0.004923113156110048 - training/rollout_probs_diff_std:0.012971784919500351 - training/rollout_actor_probs_pearson_corr:0.9982289671897888 - self_distillation/success_group_fraction:0.75 - self_distillation/success_sample_fraction:0.73828125 - self_distillation/correct_sample_fraction:0.53515625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.73828125 - rollout_corr/rollout_is_mean:1.0000406503677368 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.00011845253902720287 - rollout_corr/rollout_is_max:1.6395440101623535 - rollout_corr/rollout_is_min:0.3593428432941437 - rollout_corr/rollout_is_std:0.03901364281773567 - rollout_corr/rollout_is_eff_sample_size:0.9984802489496959 - rollout_corr/rollout_is_seq_mean:1.000063419342041 - rollout_corr/rollout_is_seq_std:0.003165917471051216 - rollout_corr/rollout_is_seq_max:1.0123790502548218 - rollout_corr/rollout_is_seq_min:0.9913737773895264 - rollout_corr/rollout_is_seq_max_deviation:0.012379050254821777 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.2515486287884414) - rollout_corr/training_log_ppl:np.float64(0.22092660737689584) - rollout_corr/kl:np.float64(0.0007296365073727884) - rollout_corr/k3_kl:np.float64(0.0007930857485121123) - rollout_corr/rollout_ppl:np.float64(1.250603314023465) - rollout_corr/rollout_log_ppl:np.float64(0.22019697143696249) - rollout_corr/log_ppl_diff:np.float64(0.0007296359399333596) - rollout_corr/log_ppl_abs_diff:np.float64(0.002573695150204003) - rollout_corr/log_ppl_diff_max:np.float64(0.00984889268875122) - rollout_corr/log_ppl_diff_min:np.float64(-0.010038644075393677) - rollout_corr/ppl_ratio:np.float64(1.0007349480874836) - rollout_corr/chi2_token:np.float64(0.0016876000445336103) - rollout_corr/chi2_seq:np.float64(0.22252861387096345) - actor/pg_loss:np.float64(-0.05002184311160818) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.060075879271607846) - srpo/sdpo_loss:np.float64(0.010054036159999669) - srpo/routed_loss:np.float64(-0.05002184311160818) - srpo/routed_token_count:np.float64(164.88671875) - srpo/correct_sample_fraction:np.float64(0.53515625) - srpo/sdpo_sample_fraction:np.float64(0.21484375) - srpo/grpo_sample_fraction:np.float64(0.78515625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.03194424846151378) - self_distillation/dynamic_weight_mean:np.float64(0.21484375069849193) - self_distillation/dynamic_weight_max:np.float64(1.2086870670318604) - self_distillation/empty_target_batch:np.float64(0.26171875) - actor/grad_norm:np.float64(0.28600582480430603) - perf/mfu/actor:np.float64(0.02493247913839777) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(146.47361373901367) - actor/lr:np.float64(5e-06) - training/global_step:27 - training/epoch:0 - critic/score/mean:0.53515625 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.53515625 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.006970931775867939 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.006970931775867939 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:164.88671875 - response_length/max:506.0 - response_length/min:93.0 -
step:28 - global_seqlen/min:12600 - global_seqlen/max:15372 - global_seqlen/minmax_diff:2772 - global_seqlen/balanced_min:13737 - global_seqlen/balanced_max:13738 - global_seqlen/mean:13737.375 - actor/entropy:0.20955954492092133 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.18672823905944824 - training/rollout_probs_diff_mean:0.004737798124551773 - training/rollout_probs_diff_std:0.01256495900452137 - training/rollout_actor_probs_pearson_corr:0.998292863368988 - self_distillation/success_group_fraction:0.84375 - self_distillation/success_sample_fraction:0.8359375 - self_distillation/correct_sample_fraction:0.6953125 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.8359375 - rollout_corr/rollout_is_mean:1.0001912117004395 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:2.4898537958506495e-05 - rollout_corr/rollout_is_max:1.6001068353652954 - rollout_corr/rollout_is_min:0.49207618832588196 - rollout_corr/rollout_is_std:0.03749287873506546 - rollout_corr/rollout_is_eff_sample_size:0.9985966139457456 - rollout_corr/rollout_is_seq_mean:1.0002046823501587 - rollout_corr/rollout_is_seq_std:0.003286403603851795 - rollout_corr/rollout_is_seq_max:1.0099278688430786 - rollout_corr/rollout_is_seq_min:0.9890894293785095 - rollout_corr/rollout_is_seq_max_deviation:0.010910570621490479 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.2381370649673045) - rollout_corr/training_log_ppl:np.float64(0.2106408952968195) - rollout_corr/kl:np.float64(0.0004954457176253868) - rollout_corr/k3_kl:np.float64(0.0007001244731839051) - rollout_corr/rollout_ppl:np.float64(1.2374732308089733) - rollout_corr/rollout_log_ppl:np.float64(0.2101454485382419) - rollout_corr/log_ppl_diff:np.float64(0.0004954467585776001) - rollout_corr/log_ppl_abs_diff:np.float64(0.002659360849065706) - rollout_corr/log_ppl_diff_max:np.float64(0.01233169436454773) - rollout_corr/log_ppl_diff_min:np.float64(-0.009259805083274841) - rollout_corr/ppl_ratio:np.float64(1.0005010685417801) - rollout_corr/chi2_token:np.float64(0.001807889435440302) - rollout_corr/chi2_seq:np.float64(0.5121276995632797) - actor/pg_loss:np.float64(-0.055125981889432296) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.06153894716408104) - srpo/sdpo_loss:np.float64(0.006412965274648741) - srpo/routed_loss:np.float64(-0.055125981889432296) - srpo/routed_token_count:np.float64(156.88671875) - srpo/correct_sample_fraction:np.float64(0.6953125) - srpo/sdpo_sample_fraction:np.float64(0.1484375) - srpo/grpo_sample_fraction:np.float64(0.8515625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.01852271431562258) - self_distillation/dynamic_weight_mean:np.float64(0.148437503259629) - self_distillation/dynamic_weight_max:np.float64(1.19734525680542) - self_distillation/empty_target_batch:np.float64(0.1640625) - actor/grad_norm:np.float64(0.2831661105155945) - perf/mfu/actor:np.float64(0.024971307326231082) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(146.5975341796875) - actor/lr:np.float64(5e-06) - training/global_step:28 - training/epoch:0 - critic/score/mean:0.6953125 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.6953125 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.004964146297425032 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.004964146297425032 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:156.88671875 - response_length/max:370.0 - response_length/min:82.0 - response_lengt
step:29 - global_seqlen/min:12220 - global_seqlen/max:16893 - global_seqlen/minmax_diff:4673 - global_seqlen/balanced_min:14770 - global_seqlen/balanced_max:14793 - global_seqlen/mean:14774.625 - actor/entropy:0.18968364596366882 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.16357311606407166 - training/rollout_probs_diff_mean:0.004385186359286308 - training/rollout_probs_diff_std:0.012322661466896534 - training/rollout_actor_probs_pearson_corr:0.9982406497001648 - self_distillation/success_group_fraction:0.75 - self_distillation/success_sample_fraction:0.7421875 - self_distillation/correct_sample_fraction:0.55078125 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.7421875 - rollout_corr/rollout_is_mean:0.9998930096626282 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.7172743082046509 - rollout_corr/rollout_is_min:0.5372289419174194 - rollout_corr/rollout_is_std:0.03543657064437866 - rollout_corr/rollout_is_eff_sample_size:0.9987455864550007 - rollout_corr/rollout_is_seq_mean:0.9999368190765381 - rollout_corr/rollout_is_seq_std:0.0030607348307967186 - rollout_corr/rollout_is_seq_max:1.0094572305679321 - rollout_corr/rollout_is_seq_min:0.9885603785514832 - rollout_corr/rollout_is_seq_max_deviation:0.011439621448516846 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.215682178735733) - rollout_corr/training_log_ppl:np.float64(0.1930056637938833) - rollout_corr/kl:np.float64(0.0007337201581449904) - rollout_corr/k3_kl:np.float64(0.0006705050529376422) - rollout_corr/rollout_ppl:np.float64(1.2147906124591827) - rollout_corr/rollout_log_ppl:np.float64(0.1922719416616019) - rollout_corr/log_ppl_diff:np.float64(0.0007337221322814003) - rollout_corr/log_ppl_abs_diff:np.float64(0.0024559692392358556) - rollout_corr/log_ppl_diff_max:np.float64(0.012552857398986816) - rollout_corr/log_ppl_diff_min:np.float64(-0.007960855960845947) - rollout_corr/ppl_ratio:np.float64(1.0007386424113065) - rollout_corr/chi2_token:np.float64(0.001203316729515791) - rollout_corr/chi2_seq:np.float64(0.254745269427076) - actor/pg_loss:np.float64(-0.06958455390849849) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.07762434854521416) - srpo/sdpo_loss:np.float64(0.00803979463671567) - srpo/routed_loss:np.float64(-0.06958455390849849) - srpo/routed_token_count:np.float64(184.95703125) - srpo/correct_sample_fraction:np.float64(0.55078125) - srpo/sdpo_sample_fraction:np.float64(0.19921875) - srpo/grpo_sample_fraction:np.float64(0.80078125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.027471024688566104) - self_distillation/dynamic_weight_mean:np.float64(0.1992187537252903) - self_distillation/dynamic_weight_max:np.float64(1.1863983869552612) - self_distillation/empty_target_batch:np.float64(0.2578125) - actor/grad_norm:np.float64(0.24951665103435516) - perf/mfu/actor:np.float64(0.02659499231400301) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(146.56739044189453) - actor/lr:np.float64(5e-06) - training/global_step:29 - training/epoch:0 - critic/score/mean:0.55078125 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.55078125 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.0025818918365985155 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.0025818918365985155 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:184.95703125 - response_length/max:734.0 - response_length/min:81.0 - response_length/clip_rat
test_gen_batch meta info: {'eos_token_id': 151645, 'pad_token_id': 151643, 'recompute_log_prob': False, 'do_sample': True, 'validate': True, 'global_steps': 30}
validation generation end
[prompt] system

Given a question and four options, please select the right answer. Respond in the following format:
<reasoning>
...
</reasoning>
<answer>
...
</answer>

For the answer, only output the letter corresponding to the correct option (A, B, C, or D), and nothing else. Do not restate the answer text. For example, if the answer is "A", just output:
<answer>
A
</answer>

user
Which of the following molecules is the only correct reactant used in the synthesis of the product "Nc1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1" ?

A: O=C1c2ccccc2C(=O)N1c1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1
B: O=C1c2ccccc2C(Cl)N1c1ccc2ccc(Cl)nc2n1
C: O=C1c2ccccc2C(=O)N1Cc1ccc(Oc2ccc(Cl)c(C(F)(F)F)c2)cc1
D: O=C1c2ccccc2C(=O)N1CC=C(Cc1ccccc1)C(F)(F)F
Please reason step by step.
assistant
<think>

</think>


[response] <reasoning>
The molecule "Nc1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1" contains a nitrile group (–CN), a chlorinated aromatic ring (–Cl), and a trifluoromethyl group (–C(F)(F)F). The correct reactant must include these functional groups and the correct structure. Option A contains a carbonyl group (C=O), a nitrile group, a chlorinated aromatic ring, and a trifluoromethyl group, matching the structure of the product. Option B contains a chlorine and a nitrile group but lacks the trifluoromethyl group. Option C contains a hydroxyl group and a chlorinated aromatic ring but lacks the nitrile group. Option D contains a carbonyl group and a chlorinated aromatic ring but lacks the nitrile and trifluoromethyl groups. Therefore, Option A is the correct reactant.
</reasoning>
<answer>
A
</answer>
[ground_truth] A
[score] 1.0
[acc] 1.0
[pred] A
[incorrect_format] 1
[feedback]
local_global_step_folder: /workspace/L2T/checkpoints/datasets/sciknoweval/chemistry/qwen3gen-chemistry-SRPO_TR-Qwen-Qwen3-4B-mbs32-tr0.1-dwtrue-train32-rollout8-lr5e-6-vllm0.8/global_step_30
step:30 - global_seqlen/min:12110 - global_seqlen/max:18283 - global_seqlen/minmax_diff:6173 - global_seqlen/balanced_min:14894 - global_seqlen/balanced_max:14968 - global_seqlen/mean:14904.25 - actor/entropy:0.17737767100334167 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.16888147592544556 - training/rollout_probs_diff_mean:0.004543249029666185 - training/rollout_probs_diff_std:0.013195773586630821 - training/rollout_actor_probs_pearson_corr:0.9978705644607544 - self_distillation/success_group_fraction:0.625 - self_distillation/success_sample_fraction:0.6171875 - self_distillation/correct_sample_fraction:0.45703125 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.6171875 - rollout_corr/rollout_is_mean:1.0000190734863281 - rollout_corr/rollout_is_ratio_fraction_high:4.99176348967012e-05 - rollout_corr/rollout_is_ratio_fraction_low:2.49588174483506e-05 - rollout_corr/rollout_is_max:2.096496820449829 - rollout_corr/rollout_is_min:0.18548955023288727 - rollout_corr/rollout_is_std:0.03955228999257088 - rollout_corr/rollout_is_eff_sample_size:0.9984380599664522 - rollout_corr/rollout_is_seq_mean:0.9999653697013855 - rollout_corr/rollout_is_seq_std:0.0033687110990285873 - rollout_corr/rollout_is_seq_max:1.0123766660690308 - rollout_corr/rollout_is_seq_min:0.990477442741394 - rollout_corr/rollout_is_seq_max_deviation:0.012376666069030762 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.1955560026690364) - rollout_corr/training_log_ppl:np.float64(0.17635832817177288) - rollout_corr/kl:np.float64(0.000810405399558789) - rollout_corr/k3_kl:np.float64(0.0007757414921485406) - rollout_corr/rollout_ppl:np.float64(1.1945632514543831) - rollout_corr/rollout_log_ppl:np.float64(0.1755479210987687) - rollout_corr/log_ppl_diff:np.float64(0.0008104070730041713) - rollout_corr/log_ppl_abs_diff:np.float64(0.0026718885637819767) - rollout_corr/log_ppl_diff_max:np.float64(0.010643362998962402) - rollout_corr/log_ppl_diff_min:np.float64(-0.00970710813999176) - rollout_corr/ppl_ratio:np.float64(1.0008163943421096) - rollout_corr/chi2_token:np.float64(0.0014906392898410559) - rollout_corr/chi2_seq:np.float64(0.30464905640110373) - actor/pg_loss:np.float64(-0.05809112397400895) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.06397552089765668) - srpo/sdpo_loss:np.float64(0.005884396923647728) - srpo/routed_loss:np.float64(-0.05809112397400895) - srpo/routed_token_count:np.float64(156.5078125) - srpo/correct_sample_fraction:np.float64(0.45703125) - srpo/sdpo_sample_fraction:np.float64(0.16796875) - srpo/grpo_sample_fraction:np.float64(0.83203125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.017346625561003748) - self_distillation/dynamic_weight_mean:np.float64(0.16796875232830644) - self_distillation/dynamic_weight_max:np.float64(1.1874160766601562) - self_distillation/empty_target_batch:np.float64(0.3828125) - actor/grad_norm:np.float64(0.25031211972236633) - perf/mfu/actor:np.float64(0.027154364409840537) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(146.5798110961914) - actor/lr:np.float64(5e-06) - val-aux/sciknoweval/reward/mean@16:np.float64(0.6607142857142857) - val-aux/sciknoweval/reward/std@16:np.float64(0.11771387065932913) - val-aux/sciknoweval/reward/best@2/mean:np.float64(0.7097142857142859) - val-aux/sciknoweval/reward/best@2/std:np.float64(0.10020945984457824) - val-aux/sciknoweval/reward/worst@2/mean:np.float64(0.6123095238095239) - val-aux/sciknoweval/reward/worst@2/std:np.float64(0.09892828956835148) - val-aux/sciknoweval/reward/maj@2/mean:np.float64(0.6611857142857
step:31 - global_seqlen/min:12313 - global_seqlen/max:16715 - global_seqlen/minmax_diff:4402 - global_seqlen/balanced_min:14483 - global_seqlen/balanced_max:14484 - global_seqlen/mean:14483.75 - actor/entropy:0.188824862241745 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.23766681551933289 - training/rollout_probs_diff_mean:0.004711161833256483 - training/rollout_probs_diff_std:0.013171381317079067 - training/rollout_actor_probs_pearson_corr:0.9979630708694458 - self_distillation/success_group_fraction:0.8125 - self_distillation/success_sample_fraction:0.80859375 - self_distillation/correct_sample_fraction:0.62109375 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.80859375 - rollout_corr/rollout_is_mean:1.0005096197128296 - rollout_corr/rollout_is_ratio_fraction_high:4.521817754721269e-05 - rollout_corr/rollout_is_ratio_fraction_low:9.043635509442538e-05 - rollout_corr/rollout_is_max:2.2306411266326904 - rollout_corr/rollout_is_min:0.32581621408462524 - rollout_corr/rollout_is_std:0.03988543525338173 - rollout_corr/rollout_is_eff_sample_size:0.9984132236126418 - rollout_corr/rollout_is_seq_mean:1.0006487369537354 - rollout_corr/rollout_is_seq_std:0.0030418585520237684 - rollout_corr/rollout_is_seq_max:1.0128170251846313 - rollout_corr/rollout_is_seq_min:0.9928036332130432 - rollout_corr/rollout_is_seq_max_deviation:0.012817025184631348 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.2101536733098328) - rollout_corr/training_log_ppl:np.float64(0.18854393177025486) - rollout_corr/kl:np.float64(0.00015732704312654278) - rollout_corr/k3_kl:np.float64(0.0008060121500648165) - rollout_corr/rollout_ppl:np.float64(1.2099390258081257) - rollout_corr/rollout_log_ppl:np.float64(0.1883866030984791) - rollout_corr/log_ppl_diff:np.float64(0.00015732867177575827) - rollout_corr/log_ppl_abs_diff:np.float64(0.0023341347405221313) - rollout_corr/log_ppl_diff_max:np.float64(0.009192712604999542) - rollout_corr/log_ppl_diff_min:np.float64(-0.011026129126548767) - rollout_corr/ppl_ratio:np.float64(1.0001619246322662) - rollout_corr/chi2_token:np.float64(0.0029538748785853386) - rollout_corr/chi2_seq:np.float64(0.4136215769685805) - actor/pg_loss:np.float64(-0.07211342807568144) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.07960189631558023) - srpo/sdpo_loss:np.float64(0.007488468239898793) - srpo/routed_loss:np.float64(-0.07211342807568144) - srpo/routed_token_count:np.float64(172.7734375) - srpo/correct_sample_fraction:np.float64(0.62109375) - srpo/sdpo_sample_fraction:np.float64(0.19140625) - srpo/grpo_sample_fraction:np.float64(0.80859375) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.023209005667013116) - self_distillation/dynamic_weight_mean:np.float64(0.19140625279396772) - self_distillation/dynamic_weight_max:np.float64(1.1894848346710205) - self_distillation/empty_target_batch:np.float64(0.19140625) - actor/grad_norm:np.float64(0.2335694432258606) - perf/mfu/actor:np.float64(0.02650972958877991) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(147.7403450012207) - actor/lr:np.float64(5e-06) - training/global_step:31 - training/epoch:0 - critic/score/mean:0.62109375 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.62109375 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.007138819899410009 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.007138819899410009 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:172.7734375 - response_length/max:477.0 - response_l
step:32 - global_seqlen/min:11587 - global_seqlen/max:16158 - global_seqlen/minmax_diff:4571 - global_seqlen/balanced_min:13771 - global_seqlen/balanced_max:13772 - global_seqlen/mean:13771.75 - actor/entropy:0.18404783308506012 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.18460848927497864 - training/rollout_probs_diff_mean:0.004470156505703926 - training/rollout_probs_diff_std:0.012472731061279774 - training/rollout_actor_probs_pearson_corr:0.9981592297554016 - self_distillation/success_group_fraction:0.75 - self_distillation/success_sample_fraction:0.74609375 - self_distillation/correct_sample_fraction:0.6328125 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.74609375 - rollout_corr/rollout_is_mean:0.9998106360435486 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:2.331763244001195e-05 - rollout_corr/rollout_is_max:1.8450249433517456 - rollout_corr/rollout_is_min:0.3978606164455414 - rollout_corr/rollout_is_std:0.037302419543266296 - rollout_corr/rollout_is_eff_sample_size:0.9986099281155342 - rollout_corr/rollout_is_seq_mean:0.9998985528945923 - rollout_corr/rollout_is_seq_std:0.003195658093318343 - rollout_corr/rollout_is_seq_max:1.0109714269638062 - rollout_corr/rollout_is_seq_min:0.9906988739967346 - rollout_corr/rollout_is_seq_max_deviation:0.010971426963806152 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.2176153464242816) - rollout_corr/training_log_ppl:np.float64(0.19434502808144316) - rollout_corr/kl:np.float64(0.0008236887766273071) - rollout_corr/k3_kl:np.float64(0.0007222598528642266) - rollout_corr/rollout_ppl:np.float64(1.2165799094364047) - rollout_corr/rollout_log_ppl:np.float64(0.19352133850043174) - rollout_corr/log_ppl_diff:np.float64(0.000823689581011422) - rollout_corr/log_ppl_abs_diff:np.float64(0.002594720179331489) - rollout_corr/log_ppl_diff_max:np.float64(0.011539280414581299) - rollout_corr/log_ppl_diff_min:np.float64(-0.010043546557426453) - rollout_corr/ppl_ratio:np.float64(1.0008292831480503) - rollout_corr/chi2_token:np.float64(0.0012312422040849924) - rollout_corr/chi2_seq:np.float64(0.18337941938079894) - actor/pg_loss:np.float64(-0.05026090465253219) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.05467792661511339) - srpo/sdpo_loss:np.float64(0.004417021962581202) - srpo/routed_loss:np.float64(-0.05026090465253219) - srpo/routed_token_count:np.float64(167.5234375) - srpo/correct_sample_fraction:np.float64(0.6328125) - srpo/sdpo_sample_fraction:np.float64(0.1171875) - srpo/grpo_sample_fraction:np.float64(0.8828125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.01452769483148586) - self_distillation/dynamic_weight_mean:np.float64(0.1171874983701855) - self_distillation/dynamic_weight_max:np.float64(1.199898600578308) - self_distillation/empty_target_batch:np.float64(0.25390625) - actor/grad_norm:np.float64(0.17714713513851166) - perf/mfu/actor:np.float64(0.02547200226504209) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(147.54729843139648) - actor/lr:np.float64(5e-06) - training/global_step:32 - training/epoch:0 - critic/score/mean:0.6328125 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.6328125 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.0004051438590977341 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.0004051438590977341 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:167.5234375 - response_length/max:417.0 - response_length/min:76.0 - response_
step:33 - global_seqlen/min:12150 - global_seqlen/max:17213 - global_seqlen/minmax_diff:5063 - global_seqlen/balanced_min:14632 - global_seqlen/balanced_max:14635 - global_seqlen/mean:14634.0 - actor/entropy:0.18772776424884796 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.23061612248420715 - training/rollout_probs_diff_mean:0.004567546304315329 - training/rollout_probs_diff_std:0.012933501973748207 - training/rollout_actor_probs_pearson_corr:0.9980109930038452 - self_distillation/success_group_fraction:0.84375 - self_distillation/success_sample_fraction:0.828125 - self_distillation/correct_sample_fraction:0.625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.828125 - rollout_corr/rollout_is_mean:0.9999642968177795 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:2.3882308596512303e-05 - rollout_corr/rollout_is_max:1.9137009382247925 - rollout_corr/rollout_is_min:0.46146896481513977 - rollout_corr/rollout_is_std:0.03866219520568848 - rollout_corr/rollout_is_eff_sample_size:0.998507346853073 - rollout_corr/rollout_is_seq_mean:1.0000168085098267 - rollout_corr/rollout_is_seq_std:0.003385877702385187 - rollout_corr/rollout_is_seq_max:1.0134313106536865 - rollout_corr/rollout_is_seq_min:0.9908832907676697 - rollout_corr/rollout_is_seq_max_deviation:0.013431310653686523 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.2073346334509552) - rollout_corr/training_log_ppl:np.float64(0.18655031187518034) - rollout_corr/kl:np.float64(0.0007271006989242323) - rollout_corr/k3_kl:np.float64(0.000743974521995483) - rollout_corr/rollout_ppl:np.float64(1.20645138528198) - rollout_corr/rollout_log_ppl:np.float64(0.18582320961286314) - rollout_corr/log_ppl_diff:np.float64(0.000727102262317203) - rollout_corr/log_ppl_abs_diff:np.float64(0.0026685910852393135) - rollout_corr/log_ppl_diff_max:np.float64(0.010880842804908752) - rollout_corr/log_ppl_diff_min:np.float64(-0.009981662034988403) - rollout_corr/ppl_ratio:np.float64(1.0007328442297876) - rollout_corr/chi2_token:np.float64(0.001532071502879262) - rollout_corr/chi2_seq:np.float64(0.4351268464233726) - actor/pg_loss:np.float64(-0.05233799253619509) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.06148528491030447) - srpo/sdpo_loss:np.float64(0.009147292374109384) - srpo/routed_loss:np.float64(-0.05233799253619509) - srpo/routed_token_count:np.float64(163.5625) - srpo/correct_sample_fraction:np.float64(0.625) - srpo/sdpo_sample_fraction:np.float64(0.21875) - srpo/grpo_sample_fraction:np.float64(0.78125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.026329860389523674) - self_distillation/dynamic_weight_mean:np.float64(0.21875000302679837) - self_distillation/dynamic_weight_max:np.float64(1.1801494359970093) - self_distillation/empty_target_batch:np.float64(0.171875) - actor/grad_norm:np.float64(0.24161788821220398) - perf/mfu/actor:np.float64(0.02685388791170372) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(147.23774909973145) - actor/lr:np.float64(5e-06) - training/global_step:33 - training/epoch:0 - critic/score/mean:0.625 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.625 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.00019105846877209842 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.00019105846877209842 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:163.5625 - response_length/max:392.0 - response_length/min:79.0 - response_length/clip_ratio:0.0 - res
step:34 - global_seqlen/min:12942 - global_seqlen/max:18895 - global_seqlen/minmax_diff:5953 - global_seqlen/balanced_min:14914 - global_seqlen/balanced_max:14954 - global_seqlen/mean:14919.625 - actor/entropy:0.1748884618282318 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.27834680676460266 - training/rollout_probs_diff_mean:0.00437568174675107 - training/rollout_probs_diff_std:0.012682687491178513 - training/rollout_actor_probs_pearson_corr:0.998024582862854 - self_distillation/success_group_fraction:0.875 - self_distillation/success_sample_fraction:0.8671875 - self_distillation/correct_sample_fraction:0.63671875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.8671875 - rollout_corr/rollout_is_mean:0.9998446106910706 - rollout_corr/rollout_is_ratio_fraction_high:2.2650569007964805e-05 - rollout_corr/rollout_is_ratio_fraction_low:2.2650569007964805e-05 - rollout_corr/rollout_is_max:2.075303316116333 - rollout_corr/rollout_is_min:0.44729122519493103 - rollout_corr/rollout_is_std:0.03772900998592377 - rollout_corr/rollout_is_eff_sample_size:0.9985781886714111 - rollout_corr/rollout_is_seq_mean:0.9997809529304504 - rollout_corr/rollout_is_seq_std:0.003234319156035781 - rollout_corr/rollout_is_seq_max:1.0120673179626465 - rollout_corr/rollout_is_seq_min:0.9911702275276184 - rollout_corr/rollout_is_seq_max_deviation:0.012067317962646484 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.1961486032232642) - rollout_corr/training_log_ppl:np.float64(0.17702986954827793) - rollout_corr/kl:np.float64(0.0009291277369527506) - rollout_corr/k3_kl:np.float64(0.0007101078383868753) - rollout_corr/rollout_ppl:np.float64(1.1949990568682551) - rollout_corr/rollout_log_ppl:np.float64(0.17610074182448443) - rollout_corr/log_ppl_diff:np.float64(0.0009291277237934992) - rollout_corr/log_ppl_abs_diff:np.float64(0.0026410768186906353) - rollout_corr/log_ppl_diff_max:np.float64(0.010610044002532959) - rollout_corr/log_ppl_diff_min:np.float64(-0.009553775191307068) - rollout_corr/ppl_ratio:np.float64(1.0009348259773105) - rollout_corr/chi2_token:np.float64(0.0009947649668902159) - rollout_corr/chi2_seq:np.float64(0.44467601692304015) - actor/pg_loss:np.float64(-0.08193617772849393) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.09220349756651558) - srpo/sdpo_loss:np.float64(0.010267319838021649) - srpo/routed_loss:np.float64(-0.08193617772849393) - srpo/routed_token_count:np.float64(172.45703125) - srpo/correct_sample_fraction:np.float64(0.63671875) - srpo/sdpo_sample_fraction:np.float64(0.23828125) - srpo/grpo_sample_fraction:np.float64(0.76171875) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.026627622435626108) - self_distillation/dynamic_weight_mean:np.float64(0.23828125419095159) - self_distillation/dynamic_weight_max:np.float64(1.175632357597351) - self_distillation/empty_target_batch:np.float64(0.1328125) - actor/grad_norm:np.float64(0.30183109641075134) - perf/mfu/actor:np.float64(0.027363600802092405) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(146.96025848388672) - actor/lr:np.float64(5e-06) - training/global_step:34 - training/epoch:0 - critic/score/mean:0.63671875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.63671875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.005144510418176651 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.005144510418176651 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:172.45703125 - response_length/max:828.0 - response
step:35 - global_seqlen/min:11424 - global_seqlen/max:15744 - global_seqlen/minmax_diff:4320 - global_seqlen/balanced_min:14013 - global_seqlen/balanced_max:14113 - global_seqlen/mean:14028.875 - actor/entropy:0.1841861754655838 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.24013252556324005 - training/rollout_probs_diff_mean:0.004258418921381235 - training/rollout_probs_diff_std:0.01207087840884924 - training/rollout_actor_probs_pearson_corr:0.9982625842094421 - self_distillation/success_group_fraction:0.71875 - self_distillation/success_sample_fraction:0.71484375 - self_distillation/correct_sample_fraction:0.58984375 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.71484375 - rollout_corr/rollout_is_mean:0.9998727440834045 - rollout_corr/rollout_is_ratio_fraction_high:4.8762649385025725e-05 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:2.01124906539917 - rollout_corr/rollout_is_min:0.5762324929237366 - rollout_corr/rollout_is_std:0.036103107035160065 - rollout_corr/rollout_is_eff_sample_size:0.9986980245594285 - rollout_corr/rollout_is_seq_mean:0.9998555779457092 - rollout_corr/rollout_is_seq_std:0.002791300881654024 - rollout_corr/rollout_is_seq_max:1.0089480876922607 - rollout_corr/rollout_is_seq_min:0.9933629035949707 - rollout_corr/rollout_is_seq_max_deviation:0.008948087692260742 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.2016498176380992) - rollout_corr/training_log_ppl:np.float64(0.1817102071508998) - rollout_corr/kl:np.float64(0.0007822071734153724) - rollout_corr/k3_kl:np.float64(0.0006378280384922164) - rollout_corr/rollout_ppl:np.float64(1.2007191772572696) - rollout_corr/rollout_log_ppl:np.float64(0.18092799821170047) - rollout_corr/log_ppl_diff:np.float64(0.0007822089391993359) - rollout_corr/log_ppl_abs_diff:np.float64(0.002295123369549401) - rollout_corr/log_ppl_diff_max:np.float64(0.007710397243499756) - rollout_corr/log_ppl_diff_min:np.float64(-0.007772907614707947) - rollout_corr/ppl_ratio:np.float64(1.000786368502304) - rollout_corr/chi2_token:np.float64(0.0010063061490654945) - rollout_corr/chi2_seq:np.float64(0.19242452923208475) - actor/pg_loss:np.float64(-0.05601486234809272) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.06103534350404516) - srpo/sdpo_loss:np.float64(0.005020481155952439) - srpo/routed_loss:np.float64(-0.05601486234809272) - srpo/routed_token_count:np.float64(160.21484375) - srpo/correct_sample_fraction:np.float64(0.58984375) - srpo/sdpo_sample_fraction:np.float64(0.12890625) - srpo/grpo_sample_fraction:np.float64(0.87109375) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.014977284445194528) - self_distillation/dynamic_weight_mean:np.float64(0.12890624976716936) - self_distillation/dynamic_weight_max:np.float64(1.2217007875442505) - self_distillation/empty_target_batch:np.float64(0.28515625) - actor/grad_norm:np.float64(0.18823640048503876) - perf/mfu/actor:np.float64(0.025889175035722266) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(146.7101593017578) - actor/lr:np.float64(5e-06) - training/global_step:35 - training/epoch:0 - critic/score/mean:0.58984375 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.58984375 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.0037882481701672077 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.0037882481701672077 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:160.21484375 - response_length/max:881.0 - response_length/min:80.
step:36 - global_seqlen/min:10994 - global_seqlen/max:17070 - global_seqlen/minmax_diff:6076 - global_seqlen/balanced_min:14160 - global_seqlen/balanced_max:14161 - global_seqlen/mean:14160.875 - actor/entropy:0.1855504959821701 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.2412572205066681 - training/rollout_probs_diff_mean:0.004480615258216858 - training/rollout_probs_diff_std:0.012676697224378586 - training/rollout_actor_probs_pearson_corr:0.998103141784668 - self_distillation/success_group_fraction:0.84375 - self_distillation/success_sample_fraction:0.83203125 - self_distillation/correct_sample_fraction:0.66015625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.83203125 - rollout_corr/rollout_is_mean:0.9998720288276672 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:4.748450737679377e-05 - rollout_corr/rollout_is_max:1.797577977180481 - rollout_corr/rollout_is_min:0.334295392036438 - rollout_corr/rollout_is_std:0.03790553659200668 - rollout_corr/rollout_is_eff_sample_size:0.9985648753479487 - rollout_corr/rollout_is_seq_mean:0.9998975992202759 - rollout_corr/rollout_is_seq_std:0.002919631777331233 - rollout_corr/rollout_is_seq_max:1.0101168155670166 - rollout_corr/rollout_is_seq_min:0.9907914400100708 - rollout_corr/rollout_is_seq_max_deviation:0.010116815567016602 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.209792582783848) - rollout_corr/training_log_ppl:np.float64(0.18843161966651678) - rollout_corr/kl:np.float64(0.0008425731536743086) - rollout_corr/k3_kl:np.float64(0.0007401798200703524) - rollout_corr/rollout_ppl:np.float64(1.2087329658679664) - rollout_corr/rollout_log_ppl:np.float64(0.18758904494461603) - rollout_corr/log_ppl_diff:np.float64(0.0008425747219007462) - rollout_corr/log_ppl_abs_diff:np.float64(0.0023987655877135694) - rollout_corr/log_ppl_diff_max:np.float64(0.011744856834411621) - rollout_corr/log_ppl_diff_min:np.float64(-0.008812174201011658) - rollout_corr/ppl_ratio:np.float64(1.000847433693707) - rollout_corr/chi2_token:np.float64(0.001261165365576744) - rollout_corr/chi2_seq:np.float64(0.1816629678942263) - actor/pg_loss:np.float64(-0.05659248860320076) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.06393627385841683) - srpo/sdpo_loss:np.float64(0.007343785255216062) - srpo/routed_loss:np.float64(-0.05659248860320076) - srpo/routed_token_count:np.float64(164.52734375) - srpo/correct_sample_fraction:np.float64(0.66015625) - srpo/sdpo_sample_fraction:np.float64(0.18359375) - srpo/grpo_sample_fraction:np.float64(0.81640625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.019395788229303434) - self_distillation/dynamic_weight_mean:np.float64(0.18359375139698386) - self_distillation/dynamic_weight_max:np.float64(1.1676045656204224) - self_distillation/empty_target_batch:np.float64(0.16796875) - actor/grad_norm:np.float64(0.22572661936283112) - perf/mfu/actor:np.float64(0.026105056892750958) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(146.71802520751953) - actor/lr:np.float64(5e-06) - training/global_step:36 - training/epoch:0 - critic/score/mean:0.66015625 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.66015625 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.0021071250084787607 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.0021071250084787607 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:164.52734375 - response_length/max:457.0 - response_length/min:100.0 -
step:37 - global_seqlen/min:12616 - global_seqlen/max:18821 - global_seqlen/minmax_diff:6205 - global_seqlen/balanced_min:15361 - global_seqlen/balanced_max:15366 - global_seqlen/mean:15363.875 - actor/entropy:0.18329748511314392 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.6190891265869141 - training/rollout_probs_diff_mean:0.004406999796628952 - training/rollout_probs_diff_std:0.01318670529872179 - training/rollout_actor_probs_pearson_corr:0.9979522228240967 - self_distillation/success_group_fraction:0.75 - self_distillation/success_sample_fraction:0.7265625 - self_distillation/correct_sample_fraction:0.5390625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.7265625 - rollout_corr/rollout_is_mean:0.9996622800827026 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.00011671607353491709 - rollout_corr/rollout_is_max:1.6661827564239502 - rollout_corr/rollout_is_min:0.34132999181747437 - rollout_corr/rollout_is_std:0.037202417850494385 - rollout_corr/rollout_is_eff_sample_size:0.9986169419727516 - rollout_corr/rollout_is_seq_mean:0.9996205568313599 - rollout_corr/rollout_is_seq_std:0.003007488092407584 - rollout_corr/rollout_is_seq_max:1.0131856203079224 - rollout_corr/rollout_is_seq_min:0.985701322555542 - rollout_corr/rollout_is_seq_max_deviation:0.014298677444458008 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.208476509898901) - rollout_corr/training_log_ppl:np.float64(0.18690017331391573) - rollout_corr/kl:np.float64(0.0010778297032736361) - rollout_corr/k3_kl:np.float64(0.0006983787412195852) - rollout_corr/rollout_ppl:np.float64(1.2071509337984025) - rollout_corr/rollout_log_ppl:np.float64(0.18582234288624022) - rollout_corr/log_ppl_diff:np.float64(0.001077830427675508) - rollout_corr/log_ppl_abs_diff:np.float64(0.0024270536814583465) - rollout_corr/log_ppl_diff_max:np.float64(0.015799403190612793) - rollout_corr/log_ppl_diff_min:np.float64(-0.01166275143623352) - rollout_corr/ppl_ratio:np.float64(1.0010831418912858) - rollout_corr/chi2_token:np.float64(0.0006131951231509447) - rollout_corr/chi2_seq:np.float64(0.21189003996551037) - actor/pg_loss:np.float64(-0.046745838335482404) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.05468930123606697) - srpo/sdpo_loss:np.float64(0.007943462900584564) - srpo/routed_loss:np.float64(-0.046745838335482404) - srpo/routed_token_count:np.float64(167.33984375) - srpo/correct_sample_fraction:np.float64(0.5390625) - srpo/sdpo_sample_fraction:np.float64(0.2109375) - srpo/grpo_sample_fraction:np.float64(0.7890625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.02321287294034846) - self_distillation/dynamic_weight_mean:np.float64(0.21093750419095159) - self_distillation/dynamic_weight_max:np.float64(1.1716129779815674) - self_distillation/empty_target_batch:np.float64(0.2734375) - actor/grad_norm:np.float64(0.20181487500667572) - perf/mfu/actor:np.float64(0.028130861893423974) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(146.7349395751953) - actor/lr:np.float64(5e-06) - training/global_step:37 - training/epoch:0 - critic/score/mean:0.5390625 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.5390625 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.002725320402532816 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.002725320402532816 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:167.33984375 - response_length/max:505.0 - response_length/min:94.0 - response_
step:38 - global_seqlen/min:11336 - global_seqlen/max:16039 - global_seqlen/minmax_diff:4703 - global_seqlen/balanced_min:13804 - global_seqlen/balanced_max:13806 - global_seqlen/mean:13805.0 - actor/entropy:0.19469979405403137 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.19663426280021667 - training/rollout_probs_diff_mean:0.004310626536607742 - training/rollout_probs_diff_std:0.011823032982647419 - training/rollout_actor_probs_pearson_corr:0.9984094500541687 - self_distillation/success_group_fraction:0.75 - self_distillation/success_sample_fraction:0.74609375 - self_distillation/correct_sample_fraction:0.6484375 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.74609375 - rollout_corr/rollout_is_mean:1.0000101327896118 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.6423577070236206 - rollout_corr/rollout_is_min:0.6304458379745483 - rollout_corr/rollout_is_std:0.035698000341653824 - rollout_corr/rollout_is_eff_sample_size:0.9987271556828731 - rollout_corr/rollout_is_seq_mean:0.9999057054519653 - rollout_corr/rollout_is_seq_std:0.0030004396103322506 - rollout_corr/rollout_is_seq_max:1.010695457458496 - rollout_corr/rollout_is_seq_min:0.9909241795539856 - rollout_corr/rollout_is_seq_max_deviation:0.010695457458496094 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.2244838746264577) - rollout_corr/training_log_ppl:np.float64(0.20037621886876877) - rollout_corr/kl:np.float64(0.0007337382971455142) - rollout_corr/k3_kl:np.float64(0.0006395210393890238) - rollout_corr/rollout_ppl:np.float64(1.2235533706843853) - rollout_corr/rollout_log_ppl:np.float64(0.19964247905591037) - rollout_corr/log_ppl_diff:np.float64(0.0007337398128584027) - rollout_corr/log_ppl_abs_diff:np.float64(0.0023461630335077643) - rollout_corr/log_ppl_diff_max:np.float64(0.009968236088752747) - rollout_corr/log_ppl_diff_min:np.float64(-0.009383782744407654) - rollout_corr/ppl_ratio:np.float64(1.0007385348435491) - rollout_corr/chi2_token:np.float64(0.0010978132486343384) - rollout_corr/chi2_seq:np.float64(0.2926100918557495) - actor/pg_loss:np.float64(-0.031498019299760927) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.036149865394691005) - srpo/sdpo_loss:np.float64(0.004651846094930079) - srpo/routed_loss:np.float64(-0.031498019299760927) - srpo/routed_token_count:np.float64(168.25) - srpo/correct_sample_fraction:np.float64(0.6484375) - srpo/sdpo_sample_fraction:np.float64(0.1015625) - srpo/grpo_sample_fraction:np.float64(0.8984375) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.015959200187353417) - self_distillation/dynamic_weight_mean:np.float64(0.10156250116415322) - self_distillation/dynamic_weight_max:np.float64(1.186326026916504) - self_distillation/empty_target_batch:np.float64(0.25390625) - actor/grad_norm:np.float64(0.16004018485546112) - perf/mfu/actor:np.float64(0.025317718893953366) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(146.77448654174805) - actor/lr:np.float64(5e-06) - training/global_step:38 - training/epoch:0 - critic/score/mean:0.6484375 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.6484375 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.0030182020273059607 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.0030182020273059607 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:168.25 - response_length/max:483.0 - response_length/min:83.0 - response_length/clip_ratio:0.
step:39 - global_seqlen/min:11460 - global_seqlen/max:17458 - global_seqlen/minmax_diff:5998 - global_seqlen/balanced_min:14821 - global_seqlen/balanced_max:14824 - global_seqlen/mean:14823.0 - actor/entropy:0.19859203696250916 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.24129104614257812 - training/rollout_probs_diff_mean:0.004370540380477905 - training/rollout_probs_diff_std:0.011863414198160172 - training/rollout_actor_probs_pearson_corr:0.9984091520309448 - self_distillation/success_group_fraction:0.8125 - self_distillation/success_sample_fraction:0.8046875 - self_distillation/correct_sample_fraction:0.64453125 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.8046875 - rollout_corr/rollout_is_mean:1.0002260208129883 - rollout_corr/rollout_is_ratio_fraction_high:2.1588946765405126e-05 - rollout_corr/rollout_is_ratio_fraction_low:4.317789353081025e-05 - rollout_corr/rollout_is_max:3.9151694774627686 - rollout_corr/rollout_is_min:0.4447897970676422 - rollout_corr/rollout_is_std:0.03595255687832832 - rollout_corr/rollout_is_eff_sample_size:0.998709676804875 - rollout_corr/rollout_is_seq_mean:1.0001412630081177 - rollout_corr/rollout_is_seq_std:0.0029865617398172617 - rollout_corr/rollout_is_seq_max:1.0113250017166138 - rollout_corr/rollout_is_seq_min:0.9910593032836914 - rollout_corr/rollout_is_seq_max_deviation:0.01132500171661377 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.2288674963638186) - rollout_corr/training_log_ppl:np.float64(0.20383806466998067) - rollout_corr/kl:np.float64(0.0005210551098713267) - rollout_corr/k3_kl:np.float64(0.0006623042312696725) - rollout_corr/rollout_ppl:np.float64(1.2281886180862784) - rollout_corr/rollout_log_ppl:np.float64(0.2033170077484101) - rollout_corr/log_ppl_diff:np.float64(0.0005210569215705618) - rollout_corr/log_ppl_abs_diff:np.float64(0.0023519034584751353) - rollout_corr/log_ppl_diff_max:np.float64(0.011496543884277344) - rollout_corr/log_ppl_diff_min:np.float64(-0.009112030267715454) - rollout_corr/ppl_ratio:np.float64(1.0005256843287498) - rollout_corr/chi2_token:np.float64(0.0016580140218138695) - rollout_corr/chi2_seq:np.float64(0.47340532299131155) - actor/pg_loss:np.float64(-0.04030750980746234) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.04741122867562808) - srpo/sdpo_loss:np.float64(0.007103718868165743) - srpo/routed_loss:np.float64(-0.04030750980746234) - srpo/routed_token_count:np.float64(180.9375) - srpo/correct_sample_fraction:np.float64(0.64453125) - srpo/sdpo_sample_fraction:np.float64(0.16796875) - srpo/grpo_sample_fraction:np.float64(0.83203125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.021628366186632775) - self_distillation/dynamic_weight_mean:np.float64(0.16796875116415322) - self_distillation/dynamic_weight_max:np.float64(1.192583441734314) - self_distillation/empty_target_batch:np.float64(0.1953125) - actor/grad_norm:np.float64(0.18529443442821503) - perf/mfu/actor:np.float64(0.02714486318336152) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(146.8396873474121) - actor/lr:np.float64(5e-06) - training/global_step:39 - training/epoch:0 - critic/score/mean:0.64453125 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.64453125 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.0024422495625913143 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.0024422495625913143 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:180.9375 - response_length/max:498.0 - response_length/m
test_gen_batch meta info: {'eos_token_id': 151645, 'pad_token_id': 151643, 'recompute_log_prob': False, 'do_sample': True, 'validate': True, 'global_steps': 40}
validation generation end
[prompt] system

Given a question and four options, please select the right answer. Respond in the following format:
<reasoning>
...
</reasoning>
<answer>
...
</answer>

For the answer, only output the letter corresponding to the correct option (A, B, C, or D), and nothing else. Do not restate the answer text. For example, if the answer is "A", just output:
<answer>
A
</answer>

user
Which of the following molecules is the only correct reactant used in the synthesis of the product "Nc1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1" ?

A: O=C1c2ccccc2C(=O)N1c1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1
B: O=C1c2ccccc2C(Cl)N1c1ccc2ccc(Cl)nc2n1
C: O=C1c2ccccc2C(=O)N1Cc1ccc(Oc2ccc(Cl)c(C(F)(F)F)c2)cc1
D: O=C1c2ccccc2C(=O)N1CC=C(Cc1ccccc1)C(F)(F)F
Please reason step by step.
assistant
<think>

</think>


[response] <reasoning>
To determine the correct reactant for the synthesis of the product "Nc1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1", we need to analyze the structure and functional groups present in the product. The product contains a nitrile group (–CN), a chlorinated benzene ring (–Cl), and a trifluorinated group (–C(F)(F)F).

Option A contains a nitrile group, a benzene ring, and a chlorinated benzene ring, but also has a carbonyl group (C=O) and a carboxylic acid group (–COOH), which are not present in the product. This makes Option A incorrect.

Option B contains a nitrile group, a chlorinated benzene ring, and a fluorinated group, but it also includes a different structure that does not match the product's exact structure. This makes Option B incorrect.

Option C contains a nitrile group, a chlorinated benzene ring, and a fluorinated group, but the structure is not identical to the product. This makes Option C incorrect.

Option D contains a nitrile group, a chlorinated benzene ring, and a fluorinated group, but the structure is not identical to the product. This makes Option D incorrect.

Option A is the only one that contains the correct functional groups and structure to synthesize the product "Nc1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1". Therefore, Option A is the correct answer.
</reasoning>
<answer>
A
</answer>
[ground_truth] A
[score] 1.0
[acc] 1.0
[pred] A
[incorrect_format] 1
[feedback]
local_global_step_folder: /workspace/L2T/checkpoints/datasets/sciknoweval/chemistry/qwen3gen-chemistry-SRPO_TR-Qwen-Qwen3-4B-mbs32-tr0.1-dwtrue-train32-rollout8-lr5e-6-vllm0.8/global_step_40
step:40 - global_seqlen/min:14271 - global_seqlen/max:19433 - global_seqlen/minmax_diff:5162 - global_seqlen/balanced_min:16332 - global_seqlen/balanced_max:16334 - global_seqlen/mean:16333.375 - actor/entropy:0.2194555550813675 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.2772992253303528 - training/rollout_probs_diff_mean:0.004598392639309168 - training/rollout_probs_diff_std:0.011828729882836342 - training/rollout_actor_probs_pearson_corr:0.9985129833221436 - self_distillation/success_group_fraction:0.84375 - self_distillation/success_sample_fraction:0.83203125 - self_distillation/correct_sample_fraction:0.59765625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.83203125 - rollout_corr/rollout_is_mean:1.0002126693725586 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:1.9755427274503745e-05 - rollout_corr/rollout_is_max:1.6968867778778076 - rollout_corr/rollout_is_min:0.47978681325912476 - rollout_corr/rollout_is_std:0.03602708503603935 - rollout_corr/rollout_is_eff_sample_size:0.9987043262502454 - rollout_corr/rollout_is_seq_mean:1.0002046823501587 - rollout_corr/rollout_is_seq_std:0.0026752494741231203 - rollout_corr/rollout_is_seq_max:1.009555459022522 - rollout_corr/rollout_is_seq_min:0.9927974343299866 - rollout_corr/rollout_is_seq_max_deviation:0.009555459022521973 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.2491243318654597) - rollout_corr/training_log_ppl:np.float64(0.21993053224286996) - rollout_corr/kl:np.float64(0.0004517044676290993) - rollout_corr/k3_kl:np.float64(0.0006562921028034907) - rollout_corr/rollout_ppl:np.float64(1.2485425383783877) - rollout_corr/rollout_log_ppl:np.float64(0.21947882653330453) - rollout_corr/log_ppl_diff:np.float64(0.0004517057095654309) - rollout_corr/log_ppl_abs_diff:np.float64(0.002097801072522998) - rollout_corr/log_ppl_diff_max:np.float64(0.008046537637710571) - rollout_corr/log_ppl_diff_min:np.float64(-0.007534638047218323) - rollout_corr/ppl_ratio:np.float64(1.0004554213955998) - rollout_corr/chi2_token:np.float64(0.0017199947033077478) - rollout_corr/chi2_seq:np.float64(0.38831929420121014) - actor/pg_loss:np.float64(-0.05615739007771481) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.0669517561327666) - srpo/sdpo_loss:np.float64(0.010794366055051796) - srpo/routed_loss:np.float64(-0.05615739007771481) - srpo/routed_token_count:np.float64(197.73046875) - srpo/correct_sample_fraction:np.float64(0.59765625) - srpo/sdpo_sample_fraction:np.float64(0.24609375) - srpo/grpo_sample_fraction:np.float64(0.75390625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.040204741453635506) - self_distillation/dynamic_weight_mean:np.float64(0.24609375186264515) - self_distillation/dynamic_weight_max:np.float64(1.2624411582946777) - self_distillation/empty_target_batch:np.float64(0.16796875) - actor/grad_norm:np.float64(0.2739606201648712) - perf/mfu/actor:np.float64(0.029520311401028263) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(146.8380889892578) - actor/lr:np.float64(5e-06) - val-aux/sciknoweval/reward/mean@16:np.float64(0.6651785714285714) - val-aux/sciknoweval/reward/std@16:np.float64(0.16291382155179027) - val-aux/sciknoweval/reward/best@2/mean:np.float64(0.7331999999999999) - val-aux/sciknoweval/reward/best@2/std:np.float64(0.14419773330539495) - val-aux/sciknoweval/reward/worst@2/mean:np.float64(0.5976142857142858) - val-aux/sciknoweval/reward/worst@2/std:np.float64(0.1318677106608367) - val-aux/sciknoweval/reward/maj@2/mean:np.float64(0.6655) - val-aux/sciknow
step:41 - global_seqlen/min:12158 - global_seqlen/max:20613 - global_seqlen/minmax_diff:8455 - global_seqlen/balanced_min:15303 - global_seqlen/balanced_max:15383 - global_seqlen/mean:15317.25 - actor/entropy:0.2202489972114563 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.16518634557724 - training/rollout_probs_diff_mean:0.004007747396826744 - training/rollout_probs_diff_std:0.010575744323432446 - training/rollout_actor_probs_pearson_corr:0.9988387227058411 - self_distillation/success_group_fraction:0.84375 - self_distillation/success_sample_fraction:0.8359375 - self_distillation/correct_sample_fraction:0.609375 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.8359375 - rollout_corr/rollout_is_mean:1.000217318534851 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.661005973815918 - rollout_corr/rollout_is_min:0.5504818558692932 - rollout_corr/rollout_is_std:0.031229011714458466 - rollout_corr/rollout_is_eff_sample_size:0.999026055921873 - rollout_corr/rollout_is_seq_mean:1.000249981880188 - rollout_corr/rollout_is_seq_std:0.0024840221740305424 - rollout_corr/rollout_is_seq_max:1.010892629623413 - rollout_corr/rollout_is_seq_min:0.9906388521194458 - rollout_corr/rollout_is_seq_max_deviation:0.010892629623413086 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.2482907208614051) - rollout_corr/training_log_ppl:np.float64(0.21946839710290078) - rollout_corr/kl:np.float64(0.00026617174690812817) - rollout_corr/k3_kl:np.float64(0.0005161261577484311) - rollout_corr/rollout_ppl:np.float64(1.2479309476912022) - rollout_corr/rollout_log_ppl:np.float64(0.2192022251256276) - rollout_corr/log_ppl_diff:np.float64(0.00026617197727318853) - rollout_corr/log_ppl_abs_diff:np.float64(0.0019030998373636976) - rollout_corr/log_ppl_diff_max:np.float64(0.010309606790542603) - rollout_corr/log_ppl_diff_min:np.float64(-0.009731829166412354) - rollout_corr/ppl_ratio:np.float64(1.0002693377900869) - rollout_corr/chi2_token:np.float64(0.0015331592876464128) - rollout_corr/chi2_seq:np.float64(0.3273937862832099) - actor/pg_loss:np.float64(-0.09923090587835759) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.10845388934831135) - srpo/sdpo_loss:np.float64(0.00922298346995376) - srpo/routed_loss:np.float64(-0.09923090587835759) - srpo/routed_token_count:np.float64(214.7578125) - srpo/correct_sample_fraction:np.float64(0.609375) - srpo/sdpo_sample_fraction:np.float64(0.234375) - srpo/grpo_sample_fraction:np.float64(0.765625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.03177466163469944) - self_distillation/dynamic_weight_mean:np.float64(0.2343750016298145) - self_distillation/dynamic_weight_max:np.float64(1.1834641695022583) - self_distillation/empty_target_batch:np.float64(0.1640625) - actor/grad_norm:np.float64(0.229465514421463) - perf/mfu/actor:np.float64(0.02776822448922716) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(148.35927200317383) - actor/lr:np.float64(5e-06) - training/global_step:41 - training/epoch:0 - critic/score/mean:0.609375 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.609375 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.008125977590680122 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.008125977590680122 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:214.7578125 - response_length/max:902.0 - response_length/min:103.0 - response_length/clip_ratio:0.0 - respons
step:42 - global_seqlen/min:11929 - global_seqlen/max:19427 - global_seqlen/minmax_diff:7498 - global_seqlen/balanced_min:15583 - global_seqlen/balanced_max:15587 - global_seqlen/mean:15586.25 - actor/entropy:0.2096705138683319 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.22949355840682983 - training/rollout_probs_diff_mean:0.0037969874683767557 - training/rollout_probs_diff_std:0.010228271596133709 - training/rollout_actor_probs_pearson_corr:0.9988732933998108 - self_distillation/success_group_fraction:0.90625 - self_distillation/success_sample_fraction:0.89453125 - self_distillation/correct_sample_fraction:0.6484375 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.89453125 - rollout_corr/rollout_is_mean:1.000077247619629 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.96262526512146 - rollout_corr/rollout_is_min:0.6048500537872314 - rollout_corr/rollout_is_std:0.03112194500863552 - rollout_corr/rollout_is_eff_sample_size:0.9990323617530771 - rollout_corr/rollout_is_seq_mean:1.00010347366333 - rollout_corr/rollout_is_seq_std:0.002266454277560115 - rollout_corr/rollout_is_seq_max:1.0074236392974854 - rollout_corr/rollout_is_seq_min:0.9918352365493774 - rollout_corr/rollout_is_seq_max_deviation:0.008164763450622559 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.244527695234865) - rollout_corr/training_log_ppl:np.float64(0.21598423254908994) - rollout_corr/kl:np.float64(0.00041906705896010976) - rollout_corr/k3_kl:np.float64(0.0005225578993872659) - rollout_corr/rollout_ppl:np.float64(1.2439827038906515) - rollout_corr/rollout_log_ppl:np.float64(0.2155651640787255) - rollout_corr/log_ppl_diff:np.float64(0.0004190684703644365) - rollout_corr/log_ppl_abs_diff:np.float64(0.001786618639016524) - rollout_corr/log_ppl_diff_max:np.float64(0.009148120880126953) - rollout_corr/log_ppl_diff_min:np.float64(-0.00681149959564209) - rollout_corr/ppl_ratio:np.float64(1.0004217009991407) - rollout_corr/chi2_token:np.float64(0.0012506491038948298) - rollout_corr/chi2_seq:np.float64(0.24967773514799774) - actor/pg_loss:np.float64(-0.08863078810463776) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.09768452640855685) - srpo/sdpo_loss:np.float64(0.00905373830391909) - srpo/routed_loss:np.float64(-0.08863078810463776) - srpo/routed_token_count:np.float64(230.4453125) - srpo/correct_sample_fraction:np.float64(0.6484375) - srpo/sdpo_sample_fraction:np.float64(0.2578125) - srpo/grpo_sample_fraction:np.float64(0.7421875) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.03816179878776893) - self_distillation/dynamic_weight_mean:np.float64(0.25781250442378223) - self_distillation/dynamic_weight_max:np.float64(1.2137748003005981) - self_distillation/empty_target_batch:np.float64(0.10546875) - actor/grad_norm:np.float64(0.21108746528625488) - perf/mfu/actor:np.float64(0.02800170638246767) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(148.04235076904297) - actor/lr:np.float64(5e-06) - training/global_step:42 - training/epoch:0 - critic/score/mean:0.6484375 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.6484375 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.0014323490904644132 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.0014323490904644132 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:230.4453125 - response_length/max:765.0 - response_length/min:115.0 - response_length/clip_ratio:
step:43 - global_seqlen/min:14175 - global_seqlen/max:19390 - global_seqlen/minmax_diff:5215 - global_seqlen/balanced_min:16410 - global_seqlen/balanced_max:16413 - global_seqlen/mean:16411.875 - actor/entropy:0.19181518256664276 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.18327054381370544 - training/rollout_probs_diff_mean:0.0038967973086982965 - training/rollout_probs_diff_std:0.010845281183719635 - training/rollout_actor_probs_pearson_corr:0.9986199140548706 - self_distillation/success_group_fraction:0.90625 - self_distillation/success_sample_fraction:0.890625 - self_distillation/correct_sample_fraction:0.68359375 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.890625 - rollout_corr/rollout_is_mean:0.999852180480957 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:1.8607073798193596e-05 - rollout_corr/rollout_is_max:1.6942881345748901 - rollout_corr/rollout_is_min:0.4437640309333801 - rollout_corr/rollout_is_std:0.03268203139305115 - rollout_corr/rollout_is_eff_sample_size:0.9989326675535104 - rollout_corr/rollout_is_seq_mean:0.9998082518577576 - rollout_corr/rollout_is_seq_std:0.0024596734438091516 - rollout_corr/rollout_is_seq_max:1.0083962678909302 - rollout_corr/rollout_is_seq_min:0.9891464114189148 - rollout_corr/rollout_is_seq_max_deviation:0.010853588581085205 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.2160081127658486) - rollout_corr/training_log_ppl:np.float64(0.1932830414007185) - rollout_corr/kl:np.float64(0.000742185355782965) - rollout_corr/k3_kl:np.float64(0.0005504285711595003) - rollout_corr/rollout_ppl:np.float64(1.2150892168283463) - rollout_corr/rollout_log_ppl:np.float64(0.19254085489956196) - rollout_corr/log_ppl_diff:np.float64(0.0007421865011565387) - rollout_corr/log_ppl_abs_diff:np.float64(0.0019740653806366026) - rollout_corr/log_ppl_diff_max:np.float64(0.011882990598678589) - rollout_corr/log_ppl_diff_min:np.float64(-0.007615581154823303) - rollout_corr/ppl_ratio:np.float64(1.0007455546874553) - rollout_corr/chi2_token:np.float64(0.000717848539352417) - rollout_corr/chi2_seq:np.float64(0.13417912367731333) - actor/pg_loss:np.float64(-0.067869349339162) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.07665850463672541) - srpo/sdpo_loss:np.float64(0.008789155297563411) - srpo/routed_loss:np.float64(-0.067869349339162) - srpo/routed_token_count:np.float64(209.93359375) - srpo/correct_sample_fraction:np.float64(0.68359375) - srpo/sdpo_sample_fraction:np.float64(0.22265625) - srpo/grpo_sample_fraction:np.float64(0.77734375) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.02835027806213475) - self_distillation/dynamic_weight_mean:np.float64(0.2226562504656613) - self_distillation/dynamic_weight_max:np.float64(1.164712905883789) - self_distillation/empty_target_batch:np.float64(0.109375) - actor/grad_norm:np.float64(0.22107210755348206) - perf/mfu/actor:np.float64(0.029784760086882273) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(147.70797729492188) - actor/lr:np.float64(5e-06) - training/global_step:43 - training/epoch:0 - critic/score/mean:0.68359375 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.68359375 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.012917960993945599 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.012917960993945599 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:209.93359375 - response_length/max:795.0 - response_length/min:100.0 - respons
step:44 - global_seqlen/min:13956 - global_seqlen/max:18114 - global_seqlen/minmax_diff:4158 - global_seqlen/balanced_min:16446 - global_seqlen/balanced_max:16708 - global_seqlen/mean:16511.5 - actor/entropy:0.17668423056602478 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.2140364646911621 - training/rollout_probs_diff_mean:0.0038560531102120876 - training/rollout_probs_diff_std:0.011371465399861336 - training/rollout_actor_probs_pearson_corr:0.9984410405158997 - self_distillation/success_group_fraction:0.875 - self_distillation/success_sample_fraction:0.8671875 - self_distillation/correct_sample_fraction:0.625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.8671875 - rollout_corr/rollout_is_mean:1.0000303983688354 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:1.9108767446596175e-05 - rollout_corr/rollout_is_max:1.5924890041351318 - rollout_corr/rollout_is_min:0.4550092816352844 - rollout_corr/rollout_is_std:0.0329960361123085 - rollout_corr/rollout_is_eff_sample_size:0.9989123266706693 - rollout_corr/rollout_is_seq_mean:0.9999781847000122 - rollout_corr/rollout_is_seq_std:0.002471084939315915 - rollout_corr/rollout_is_seq_max:1.006706953048706 - rollout_corr/rollout_is_seq_min:0.9895926713943481 - rollout_corr/rollout_is_seq_max_deviation:0.010407328605651855 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.1989943333901465) - rollout_corr/training_log_ppl:np.float64(0.17951429932145402) - rollout_corr/kl:np.float64(0.0005977936627026281) - rollout_corr/k3_kl:np.float64(0.000576007029380321) - rollout_corr/rollout_ppl:np.float64(1.1982451593503356) - rollout_corr/rollout_log_ppl:np.float64(0.1789165055815829) - rollout_corr/log_ppl_diff:np.float64(0.0005977937398711219) - rollout_corr/log_ppl_abs_diff:np.float64(0.0019055179000133649) - rollout_corr/log_ppl_diff_max:np.float64(0.014274433255195618) - rollout_corr/log_ppl_diff_min:np.float64(-0.00620366632938385) - rollout_corr/ppl_ratio:np.float64(1.0006012115627527) - rollout_corr/chi2_token:np.float64(0.001096524065360427) - rollout_corr/chi2_seq:np.float64(0.13899740250781178) - actor/pg_loss:np.float64(-0.08647566499166714) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.09567046555457637) - srpo/sdpo_loss:np.float64(0.009194800562909222) - srpo/routed_loss:np.float64(-0.08647566499166714) - srpo/routed_token_count:np.float64(204.421875) - srpo/correct_sample_fraction:np.float64(0.625) - srpo/sdpo_sample_fraction:np.float64(0.25) - srpo/grpo_sample_fraction:np.float64(0.75) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.02999964888294926) - self_distillation/dynamic_weight_mean:np.float64(0.2499999995343387) - self_distillation/dynamic_weight_max:np.float64(1.1909689903259277) - self_distillation/empty_target_batch:np.float64(0.1328125) - actor/grad_norm:np.float64(0.24352417886257172) - perf/mfu/actor:np.float64(0.03006782358281605) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(147.65663146972656) - actor/lr:np.float64(5e-06) - training/global_step:44 - training/epoch:0 - critic/score/mean:0.625 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.625 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.000260356959188357 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.000260356959188357 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:204.421875 - response_length/max:1210.0 - response_length/min:111.0 - response_length/clip_ratio:0.0 - response
step:45 - global_seqlen/min:14670 - global_seqlen/max:22081 - global_seqlen/minmax_diff:7411 - global_seqlen/balanced_min:16909 - global_seqlen/balanced_max:16912 - global_seqlen/mean:16910.375 - actor/entropy:0.16450713574886322 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.5744317770004272 - training/rollout_probs_diff_mean:0.00368192489258945 - training/rollout_probs_diff_std:0.011572764255106449 - training/rollout_actor_probs_pearson_corr:0.9982879757881165 - self_distillation/success_group_fraction:0.90625 - self_distillation/success_sample_fraction:0.90234375 - self_distillation/correct_sample_fraction:0.625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.90234375 - rollout_corr/rollout_is_mean:1.000152349472046 - rollout_corr/rollout_is_ratio_fraction_high:1.8273851310368627e-05 - rollout_corr/rollout_is_ratio_fraction_low:3.6547702620737255e-05 - rollout_corr/rollout_is_max:5.564130783081055 - rollout_corr/rollout_is_min:0.3406289219856262 - rollout_corr/rollout_is_std:0.03279309347271919 - rollout_corr/rollout_is_eff_sample_size:0.9989260061205386 - rollout_corr/rollout_is_seq_mean:1.000104308128357 - rollout_corr/rollout_is_seq_std:0.002796849003061652 - rollout_corr/rollout_is_seq_max:1.0177242755889893 - rollout_corr/rollout_is_seq_min:0.9921460747718811 - rollout_corr/rollout_is_seq_max_deviation:0.017724275588989258 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.1804170426912606) - rollout_corr/training_log_ppl:np.float64(0.16382929268002044) - rollout_corr/kl:np.float64(0.0005147547473072223) - rollout_corr/k3_kl:np.float64(0.0006190431875694458) - rollout_corr/rollout_ppl:np.float64(1.1797793707810342) - rollout_corr/rollout_log_ppl:np.float64(0.16331453679595143) - rollout_corr/log_ppl_diff:np.float64(0.0005147558840690181) - rollout_corr/log_ppl_abs_diff:np.float64(0.0020618598646251485) - rollout_corr/log_ppl_diff_max:np.float64(0.010426923632621765) - rollout_corr/log_ppl_diff_min:np.float64(-0.007745638489723206) - rollout_corr/ppl_ratio:np.float64(1.000518445391208) - rollout_corr/chi2_token:np.float64(0.0016647973097860813) - rollout_corr/chi2_seq:np.float64(0.354720268631354) - actor/pg_loss:np.float64(-0.10309704847895773) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.11229040732723661) - srpo/sdpo_loss:np.float64(0.009193358848278876) - srpo/routed_loss:np.float64(-0.10309704847895773) - srpo/routed_token_count:np.float64(213.76171875) - srpo/correct_sample_fraction:np.float64(0.625) - srpo/sdpo_sample_fraction:np.float64(0.28125) - srpo/grpo_sample_fraction:np.float64(0.71875) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.028985414523958752) - self_distillation/dynamic_weight_mean:np.float64(0.2812500046566129) - self_distillation/dynamic_weight_max:np.float64(1.2057530879974365) - self_distillation/empty_target_batch:np.float64(0.09765625) - actor/grad_norm:np.float64(0.24234017729759216) - perf/mfu/actor:np.float64(0.030782650787326137) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(147.4744415283203) - actor/lr:np.float64(5e-06) - training/global_step:45 - training/epoch:0 - critic/score/mean:0.625 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.625 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.0030608701054006815 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.0030608701054006815 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:213.76171875 - response_length/max:888.0 - response_length/min:98.0 - respon
step:46 - global_seqlen/min:13765 - global_seqlen/max:26190 - global_seqlen/minmax_diff:12425 - global_seqlen/balanced_min:15780 - global_seqlen/balanced_max:22951 - global_seqlen/mean:16717.875 - actor/entropy:0.12779982388019562 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.17326676845550537 - training/rollout_probs_diff_mean:0.0030381628312170506 - training/rollout_probs_diff_std:0.010591267608106136 - training/rollout_actor_probs_pearson_corr:0.9982393980026245 - self_distillation/success_group_fraction:0.875 - self_distillation/success_sample_fraction:0.8671875 - self_distillation/correct_sample_fraction:0.60546875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.8671875 - rollout_corr/rollout_is_mean:0.9997753500938416 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:3.159807238262147e-05 - rollout_corr/rollout_is_max:1.6370503902435303 - rollout_corr/rollout_is_min:0.3131791949272156 - rollout_corr/rollout_is_std:0.029856111854314804 - rollout_corr/rollout_is_eff_sample_size:0.9991089899118506 - rollout_corr/rollout_is_seq_mean:0.9997105002403259 - rollout_corr/rollout_is_seq_std:0.002642316510900855 - rollout_corr/rollout_is_seq_max:1.0087188482284546 - rollout_corr/rollout_is_seq_min:0.9920142292976379 - rollout_corr/rollout_is_seq_max_deviation:0.00871884822845459 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.1529178912751377) - rollout_corr/training_log_ppl:np.float64(0.14044021783774951) - rollout_corr/kl:np.float64(0.0008405998159641825) - rollout_corr/k3_kl:np.float64(0.0005510257111787098) - rollout_corr/rollout_ppl:np.float64(1.151909634936601) - rollout_corr/rollout_log_ppl:np.float64(0.1395996167921112) - rollout_corr/log_ppl_diff:np.float64(0.0008406010456383228) - rollout_corr/log_ppl_abs_diff:np.float64(0.0021316910861060023) - rollout_corr/log_ppl_diff_max:np.float64(0.009405359625816345) - rollout_corr/log_ppl_diff_min:np.float64(-0.007796436548233032) - rollout_corr/ppl_ratio:np.float64(1.0008446942083538) - rollout_corr/chi2_token:np.float64(0.0004959369543939829) - rollout_corr/chi2_seq:np.float64(0.24749498208984733) - actor/pg_loss:np.float64(-0.09199389960849658) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.10003131043049507) - srpo/sdpo_loss:np.float64(0.008037410821998492) - srpo/routed_loss:np.float64(-0.09199389960849658) - srpo/routed_token_count:np.float64(247.24609375) - srpo/correct_sample_fraction:np.float64(0.60546875) - srpo/sdpo_sample_fraction:np.float64(0.26953125) - srpo/grpo_sample_fraction:np.float64(0.73046875) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.02618265039700418) - self_distillation/dynamic_weight_mean:np.float64(0.2695312553551048) - self_distillation/dynamic_weight_max:np.float64(1.1995611190795898) - self_distillation/empty_target_batch:np.float64(0.1328125) - actor/grad_norm:np.float64(0.21591348946094513) - perf/mfu/actor:np.float64(0.02867385464933913) - perf/max_memory_allocated_gb:np.float64(138.04474258422852) - perf/max_memory_reserved_gb:np.float64(142.51953125) - perf/cpu_memory_used_gb:np.float64(147.59630966186523) - actor/lr:np.float64(5e-06) - training/global_step:46 - training/epoch:0 - critic/score/mean:0.60546875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.60546875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.0020834978204220533 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.0020834978204220533 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:247.24609375 - response_length/max:8192.0 - response_length/min:110.
step:47 - global_seqlen/min:11619 - global_seqlen/max:22164 - global_seqlen/minmax_diff:10545 - global_seqlen/balanced_min:15192 - global_seqlen/balanced_max:22331 - global_seqlen/mean:16096.75 - actor/entropy:0.10864654928445816 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.19765973091125488 - training/rollout_probs_diff_mean:0.0030689530540257692 - training/rollout_probs_diff_std:0.011083385907113552 - training/rollout_actor_probs_pearson_corr:0.997769296169281 - self_distillation/success_group_fraction:0.96875 - self_distillation/success_sample_fraction:0.953125 - self_distillation/correct_sample_fraction:0.671875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.953125 - rollout_corr/rollout_is_mean:0.9999327659606934 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.9515085220336914 - rollout_corr/rollout_is_min:0.5965996384620667 - rollout_corr/rollout_is_std:0.030689943581819534 - rollout_corr/rollout_is_eff_sample_size:0.9990588947074439 - rollout_corr/rollout_is_seq_mean:0.9998971819877625 - rollout_corr/rollout_is_seq_std:0.003020802279934287 - rollout_corr/rollout_is_seq_max:1.0125021934509277 - rollout_corr/rollout_is_seq_min:0.9906147718429565 - rollout_corr/rollout_is_seq_max_deviation:0.012502193450927734 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.1375383478589356) - rollout_corr/training_log_ppl:np.float64(0.1273589103766426) - rollout_corr/kl:np.float64(0.0006866522540653897) - rollout_corr/k3_kl:np.float64(0.0005837977853744292) - rollout_corr/rollout_ppl:np.float64(1.1367522794753313) - rollout_corr/rollout_log_ppl:np.float64(0.12667225811492244) - rollout_corr/log_ppl_diff:np.float64(0.0006866522617201554) - rollout_corr/log_ppl_abs_diff:np.float64(0.002286595337864128) - rollout_corr/log_ppl_diff_max:np.float64(0.010518759489059448) - rollout_corr/log_ppl_diff_min:np.float64(-0.010414078831672668) - rollout_corr/ppl_ratio:np.float64(1.0006913184188306) - rollout_corr/chi2_token:np.float64(0.0009732998441904783) - rollout_corr/chi2_seq:np.float64(0.3054108992218971) - actor/pg_loss:np.float64(-0.07908561830936378) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.0888555436686147) - srpo/sdpo_loss:np.float64(0.00976992535925092) - srpo/routed_loss:np.float64(-0.07908561830936378) - srpo/routed_token_count:np.float64(209.7421875) - srpo/correct_sample_fraction:np.float64(0.671875) - srpo/sdpo_sample_fraction:np.float64(0.296875) - srpo/grpo_sample_fraction:np.float64(0.703125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.023933729971020057) - self_distillation/dynamic_weight_mean:np.float64(0.29687500139698386) - self_distillation/dynamic_weight_max:np.float64(1.1622016429901123) - self_distillation/empty_target_batch:np.float64(0.046875) - actor/grad_norm:np.float64(0.29469189047813416) - perf/mfu/actor:np.float64(0.027900713800772043) - perf/max_memory_allocated_gb:np.float64(138.08052349090576) - perf/max_memory_reserved_gb:np.float64(142.578125) - perf/cpu_memory_used_gb:np.float64(147.4286346435547) - actor/lr:np.float64(5e-06) - training/global_step:47 - training/epoch:0 - critic/score/mean:0.671875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.671875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.02422989532351494 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.02422989532351494 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:209.7421875 - response_length/max:8192.0 - response_length/min:91.0 - response_length/clip_ratio:0.00390625
step:48 - global_seqlen/min:11150 - global_seqlen/max:19079 - global_seqlen/minmax_diff:7929 - global_seqlen/balanced_min:14230 - global_seqlen/balanced_max:14238 - global_seqlen/mean:14234.625 - actor/entropy:0.12266239523887634 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.18411844968795776 - training/rollout_probs_diff_mean:0.0033930158242583275 - training/rollout_probs_diff_std:0.011652214452624321 - training/rollout_actor_probs_pearson_corr:0.9977874159812927 - self_distillation/success_group_fraction:0.84375 - self_distillation/success_sample_fraction:0.83984375 - self_distillation/correct_sample_fraction:0.671875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.83984375 - rollout_corr/rollout_is_mean:1.0001568794250488 - rollout_corr/rollout_is_ratio_fraction_high:2.1808824385516346e-05 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:2.140164852142334 - rollout_corr/rollout_is_min:0.5283122062683105 - rollout_corr/rollout_is_std:0.03286571800708771 - rollout_corr/rollout_is_eff_sample_size:0.9989212480085242 - rollout_corr/rollout_is_seq_mean:1.0002424716949463 - rollout_corr/rollout_is_seq_std:0.0028206203132867813 - rollout_corr/rollout_is_seq_max:1.0104187726974487 - rollout_corr/rollout_is_seq_min:0.9920458793640137 - rollout_corr/rollout_is_seq_max_deviation:0.01041877269744873 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.136358075775206) - rollout_corr/training_log_ppl:np.float64(0.1262950351301697) - rollout_corr/kl:np.float64(0.000362454386269917) - rollout_corr/k3_kl:np.float64(0.0006049648889785431) - rollout_corr/rollout_ppl:np.float64(1.1359192323870957) - rollout_corr/rollout_log_ppl:np.float64(0.12593258010747377) - rollout_corr/log_ppl_diff:np.float64(0.00036245502269594) - rollout_corr/log_ppl_abs_diff:np.float64(0.002079490579490084) - rollout_corr/log_ppl_diff_max:np.float64(0.009729623794555664) - rollout_corr/log_ppl_diff_min:np.float64(-0.0076146721839904785) - rollout_corr/ppl_ratio:np.float64(1.0003665119875222) - rollout_corr/chi2_token:np.float64(0.0017175127286463976) - rollout_corr/chi2_seq:np.float64(0.26308279670774937) - actor/pg_loss:np.float64(-0.061830449867557036) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.0674303068080917) - srpo/sdpo_loss:np.float64(0.005599856940534664) - srpo/routed_loss:np.float64(-0.061830449867557036) - srpo/routed_token_count:np.float64(179.11328125) - srpo/correct_sample_fraction:np.float64(0.671875) - srpo/sdpo_sample_fraction:np.float64(0.171875) - srpo/grpo_sample_fraction:np.float64(0.828125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.019880753348843427) - self_distillation/dynamic_weight_mean:np.float64(0.17187500069849193) - self_distillation/dynamic_weight_max:np.float64(1.1997087001800537) - self_distillation/empty_target_batch:np.float64(0.16015625) - actor/grad_norm:np.float64(0.23831798136234283) - perf/mfu/actor:np.float64(0.025999626484324488) - perf/max_memory_allocated_gb:np.float64(138.08052349090576) - perf/max_memory_reserved_gb:np.float64(142.578125) - perf/cpu_memory_used_gb:np.float64(147.3718719482422) - actor/lr:np.float64(5e-06) - training/global_step:48 - training/epoch:0 - critic/score/mean:0.671875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.671875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.01171133853495121 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.01171133853495121 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:179.11328125 - response_length/max:910.0 - response_length/min:74.0 - response_len
step:49 - global_seqlen/min:13459 - global_seqlen/max:15169 - global_seqlen/minmax_diff:1710 - global_seqlen/balanced_min:14131 - global_seqlen/balanced_max:14170 - global_seqlen/mean:14150.25 - actor/entropy:0.11217978596687317 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.23092332482337952 - training/rollout_probs_diff_mean:0.0036812860053032637 - training/rollout_probs_diff_std:0.012833652086555958 - training/rollout_actor_probs_pearson_corr:0.9971839189529419 - self_distillation/success_group_fraction:0.9375 - self_distillation/success_sample_fraction:0.9375 - self_distillation/correct_sample_fraction:0.765625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.9375 - rollout_corr/rollout_is_mean:0.9999831914901733 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:2.39085729845101e-05 - rollout_corr/rollout_is_max:1.9132364988327026 - rollout_corr/rollout_is_min:0.4921049475669861 - rollout_corr/rollout_is_std:0.03650535270571709 - rollout_corr/rollout_is_eff_sample_size:0.9986690140274969 - rollout_corr/rollout_is_seq_mean:0.9999447464942932 - rollout_corr/rollout_is_seq_std:0.003183850087225437 - rollout_corr/rollout_is_seq_max:1.012787103652954 - rollout_corr/rollout_is_seq_min:0.9903203845024109 - rollout_corr/rollout_is_seq_max_deviation:0.012787103652954102 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.1283771470189095) - rollout_corr/training_log_ppl:np.float64(0.11975863272527931) - rollout_corr/kl:np.float64(0.0007557832193860747) - rollout_corr/k3_kl:np.float64(0.0007005357908269616) - rollout_corr/rollout_ppl:np.float64(1.1275166096165776) - rollout_corr/rollout_log_ppl:np.float64(0.11900284935109084) - rollout_corr/log_ppl_diff:np.float64(0.0007557833741884679) - rollout_corr/log_ppl_abs_diff:np.float64(0.00249333304236643) - rollout_corr/log_ppl_diff_max:np.float64(0.011982902884483337) - rollout_corr/log_ppl_diff_min:np.float64(-0.01001518964767456) - rollout_corr/ppl_ratio:np.float64(1.0007611133623868) - rollout_corr/chi2_token:np.float64(0.0013036627788096666) - rollout_corr/chi2_seq:np.float64(0.28480950836092234) - actor/pg_loss:np.float64(-0.08000544076639926) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.08588153129676357) - srpo/sdpo_loss:np.float64(0.0058760905303643085) - srpo/routed_loss:np.float64(-0.08000544076639926) - srpo/routed_token_count:np.float64(163.3828125) - srpo/correct_sample_fraction:np.float64(0.765625) - srpo/sdpo_sample_fraction:np.float64(0.171875) - srpo/grpo_sample_fraction:np.float64(0.828125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.014765384265047032) - self_distillation/dynamic_weight_mean:np.float64(0.171875003259629) - self_distillation/dynamic_weight_max:np.float64(1.1998701095581055) - self_distillation/empty_target_batch:np.float64(0.0625) - actor/grad_norm:np.float64(0.24140627682209015) - perf/mfu/actor:np.float64(0.026126625085568168) - perf/max_memory_allocated_gb:np.float64(138.08052349090576) - perf/max_memory_reserved_gb:np.float64(142.578125) - perf/cpu_memory_used_gb:np.float64(147.42864990234375) - actor/lr:np.float64(5e-06) - training/global_step:49 - training/epoch:0 - critic/score/mean:0.765625 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.765625 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.004115263465791941 - critic/advantages/max:0.75 - critic/advantages/min:-0.875 - critic/returns/mean:-0.004115263465791941 - critic/returns/max:0.75 - critic/returns/min:-0.875 - response_length/mean:163.3828125 - response_length/max:591.0 - response_length/min:94.0 - response_length/clip_ratio:0.
test_gen_batch meta info: {'eos_token_id': 151645, 'pad_token_id': 151643, 'recompute_log_prob': False, 'do_sample': True, 'validate': True, 'global_steps': 50}
validation generation end
[prompt] system

Given a question and four options, please select the right answer. Respond in the following format:
<reasoning>
...
</reasoning>
<answer>
...
</answer>

For the answer, only output the letter corresponding to the correct option (A, B, C, or D), and nothing else. Do not restate the answer text. For example, if the answer is "A", just output:
<answer>
A
</answer>

user
Which of the following molecules is the only correct reactant used in the synthesis of the product "Nc1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1" ?

A: O=C1c2ccccc2C(=O)N1c1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1
B: O=C1c2ccccc2C(Cl)N1c1ccc2ccc(Cl)nc2n1
C: O=C1c2ccccc2C(=O)N1Cc1ccc(Oc2ccc(Cl)c(C(F)(F)F)c2)cc1
D: O=C1c2ccccc2C(=O)N1CC=C(Cc1ccccc1)C(F)(F)F
Please reason step by step.
assistant
<think>

</think>


[response] <reasoning>
The molecule "Nc1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1" contains a nitrile group, a chlorinated benzene ring, and a trifluorinated group. Option A includes a nitrile group, a chlorinated benzene ring, and a trifluorinated group, matching the product structure. Option B contains a nitrile group but lacks the trifluorinated group. Option C contains a nitrile group and a chlorinated benzene ring but lacks the trifluorinated group. Option D contains a nitrile group and a chlorinated benzene ring but lacks the trifluorinated group. Therefore, Option A is the correct answer.
</reasoning>
<answer>
A
</answer>
[ground_truth] A
[score] 1.0
[acc] 1.0
[pred] A
[incorrect_format] 1
[feedback]
local_global_step_folder: /workspace/L2T/checkpoints/datasets/sciknoweval/chemistry/qwen3gen-chemistry-SRPO_TR-Qwen-Qwen3-4B-mbs32-tr0.1-dwtrue-train32-rollout8-lr5e-6-vllm0.8/global_step_50
step:50 - global_seqlen/min:11417 - global_seqlen/max:26908 - global_seqlen/minmax_diff:15491 - global_seqlen/balanced_min:13867 - global_seqlen/balanced_max:20987 - global_seqlen/mean:14790.25 - actor/entropy:0.08891979604959488 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.23573046922683716 - training/rollout_probs_diff_mean:0.002700401470065117 - training/rollout_probs_diff_std:0.010878867469727993 - training/rollout_actor_probs_pearson_corr:0.9974605441093445 - self_distillation/success_group_fraction:0.9375 - self_distillation/success_sample_fraction:0.9296875 - self_distillation/correct_sample_fraction:0.62109375 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.9296875 - rollout_corr/rollout_is_mean:1.0000568628311157 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.96596097946167 - rollout_corr/rollout_is_min:0.5504809617996216 - rollout_corr/rollout_is_std:0.03079851157963276 - rollout_corr/rollout_is_eff_sample_size:0.9990523505743122 - rollout_corr/rollout_is_seq_mean:1.0000368356704712 - rollout_corr/rollout_is_seq_std:0.0028278713580220938 - rollout_corr/rollout_is_seq_max:1.006735920906067 - rollout_corr/rollout_is_seq_min:0.9882084131240845 - rollout_corr/rollout_is_seq_max_deviation:0.011791586875915527 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.1134736202657223) - rollout_corr/training_log_ppl:np.float64(0.1061642324202694) - rollout_corr/kl:np.float64(0.0005399192483643844) - rollout_corr/k3_kl:np.float64(0.0005767990856999639) - rollout_corr/rollout_ppl:np.float64(1.1128529668785632) - rollout_corr/rollout_log_ppl:np.float64(0.10562431207654299) - rollout_corr/log_ppl_diff:np.float64(0.0005399203437264077) - rollout_corr/log_ppl_abs_diff:np.float64(0.002112998925440479) - rollout_corr/log_ppl_diff_max:np.float64(0.014197483658790588) - rollout_corr/log_ppl_diff_min:np.float64(-0.005802683532238007) - rollout_corr/ppl_ratio:np.float64(1.0005440823733807) - rollout_corr/chi2_token:np.float64(0.0012443838641047478) - rollout_corr/chi2_seq:np.float64(0.14378715748898685) - actor/pg_loss:np.float64(-0.10046151399478731) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.1108631854003761) - srpo/sdpo_loss:np.float64(0.010401671405588786) - srpo/routed_loss:np.float64(-0.10046151399478731) - srpo/routed_token_count:np.float64(182.4453125) - srpo/correct_sample_fraction:np.float64(0.62109375) - srpo/sdpo_sample_fraction:np.float64(0.31640625) - srpo/grpo_sample_fraction:np.float64(0.68359375) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.025200146048291572) - self_distillation/dynamic_weight_mean:np.float64(0.3164062553551048) - self_distillation/dynamic_weight_max:np.float64(1.176309585571289) - self_distillation/empty_target_batch:np.float64(0.0703125) - actor/grad_norm:np.float64(0.326182097196579) - perf/mfu/actor:np.float64(0.025374781318022628) - perf/max_memory_allocated_gb:np.float64(138.08052349090576) - perf/max_memory_reserved_gb:np.float64(142.578125) - perf/cpu_memory_used_gb:np.float64(147.55796813964844) - actor/lr:np.float64(5e-06) - val-aux/sciknoweval/reward/mean@16:np.float64(0.687797619047619) - val-aux/sciknoweval/reward/std@16:np.float64(0.13954957514411095) - val-aux/sciknoweval/reward/best@2/mean:np.float64(0.7454904761904761) - val-aux/sciknoweval/reward/best@2/std:np.float64(0.11421358194176294) - val-aux/sciknoweval/reward/worst@2/mean:np.float64(0.6318571428571429) - val-aux/sciknoweval/reward/worst@2/std:np.float64(0.12110664816807734) - val-aux/sciknoweval/reward/maj@2/mean:np.float64(0.6884619047619047) - val-aux/sciknoweval/reward/maj@2/
step:51 - global_seqlen/min:13230 - global_seqlen/max:16684 - global_seqlen/minmax_diff:3454 - global_seqlen/balanced_min:14428 - global_seqlen/balanced_max:14578 - global_seqlen/mean:14450.0 - actor/entropy:0.10240519791841507 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.7039057612419128 - training/rollout_probs_diff_mean:0.0033974742982536554 - training/rollout_probs_diff_std:0.013343753293156624 - training/rollout_actor_probs_pearson_corr:0.9966081380844116 - self_distillation/success_group_fraction:0.90625 - self_distillation/success_sample_fraction:0.89453125 - self_distillation/correct_sample_fraction:0.59765625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.89453125 - rollout_corr/rollout_is_mean:1.0002756118774414 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:8.428860746789724e-05 - rollout_corr/rollout_is_max:1.6483187675476074 - rollout_corr/rollout_is_min:0.17377379536628723 - rollout_corr/rollout_is_std:0.034528423100709915 - rollout_corr/rollout_is_eff_sample_size:0.9988099211261683 - rollout_corr/rollout_is_seq_mean:1.000292420387268 - rollout_corr/rollout_is_seq_std:0.0031633833423256874 - rollout_corr/rollout_is_seq_max:1.0148882865905762 - rollout_corr/rollout_is_seq_min:0.9923903346061707 - rollout_corr/rollout_is_seq_max_deviation:0.014888286590576172 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.1078986530192196) - rollout_corr/training_log_ppl:np.float64(0.1014542787852406) - rollout_corr/kl:np.float64(0.00035091452643598586) - rollout_corr/k3_kl:np.float64(0.0006433185241263573) - rollout_corr/rollout_ppl:np.float64(1.107514247763902) - rollout_corr/rollout_log_ppl:np.float64(0.10110336293655564) - rollout_corr/log_ppl_diff:np.float64(0.00035091584868496284) - rollout_corr/log_ppl_abs_diff:np.float64(0.0022871532346471213) - rollout_corr/log_ppl_diff_max:np.float64(0.009798847138881683) - rollout_corr/log_ppl_diff_min:np.float64(-0.012791797518730164) - rollout_corr/ppl_ratio:np.float64(1.0003560336772352) - rollout_corr/chi2_token:np.float64(0.00182668911293149) - rollout_corr/chi2_seq:np.float64(0.3348260715138167) - actor/pg_loss:np.float64(-0.07512532324835774) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.0854524812893942) - srpo/sdpo_loss:np.float64(0.010327158041036455) - srpo/routed_loss:np.float64(-0.07512532324835774) - srpo/routed_token_count:np.float64(139.03125) - srpo/correct_sample_fraction:np.float64(0.59765625) - srpo/sdpo_sample_fraction:np.float64(0.30859375) - srpo/grpo_sample_fraction:np.float64(0.69140625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.02251810873070781) - self_distillation/dynamic_weight_mean:np.float64(0.3085937558207661) - self_distillation/dynamic_weight_max:np.float64(1.1560735702514648) - self_distillation/empty_target_batch:np.float64(0.10546875) - actor/grad_norm:np.float64(0.4182743430137634) - perf/mfu/actor:np.float64(0.026420113145447142) - perf/max_memory_allocated_gb:np.float64(138.08052349090576) - perf/max_memory_reserved_gb:np.float64(142.578125) - perf/cpu_memory_used_gb:np.float64(148.79671096801758) - actor/lr:np.float64(5e-06) - training/global_step:51 - training/epoch:0 - critic/score/mean:0.59765625 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.59765625 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.0024724656250327826 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.0024724656250327826 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:139.03125 - response_length/max:843.0 - response_length/min:82.0 - respo
step:52 - global_seqlen/min:11444 - global_seqlen/max:15111 - global_seqlen/minmax_diff:3667 - global_seqlen/balanced_min:13429 - global_seqlen/balanced_max:13432 - global_seqlen/mean:13431.0 - actor/entropy:0.10282018035650253 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.3223661184310913 - training/rollout_probs_diff_mean:0.00337245874106884 - training/rollout_probs_diff_std:0.012320202775299549 - training/rollout_actor_probs_pearson_corr:0.9971264004707336 - self_distillation/success_group_fraction:0.75 - self_distillation/success_sample_fraction:0.734375 - self_distillation/correct_sample_fraction:0.52734375 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.734375 - rollout_corr/rollout_is_mean:1.0001152753829956 - rollout_corr/rollout_is_ratio_fraction_high:2.491032319085207e-05 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:2.122667074203491 - rollout_corr/rollout_is_min:0.5241169333457947 - rollout_corr/rollout_is_std:0.03456982970237732 - rollout_corr/rollout_is_eff_sample_size:0.9988064722921542 - rollout_corr/rollout_is_seq_mean:0.9999793767929077 - rollout_corr/rollout_is_seq_std:0.003148139687255025 - rollout_corr/rollout_is_seq_max:1.0097883939743042 - rollout_corr/rollout_is_seq_min:0.9875168204307556 - rollout_corr/rollout_is_seq_max_deviation:0.012483179569244385 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.1099474504590034) - rollout_corr/training_log_ppl:np.float64(0.10346398585534189) - rollout_corr/kl:np.float64(0.0006092330206435292) - rollout_corr/k3_kl:np.float64(0.000588596026929622) - rollout_corr/rollout_ppl:np.float64(1.1092609246261418) - rollout_corr/rollout_log_ppl:np.float64(0.10285475269483868) - rollout_corr/log_ppl_diff:np.float64(0.0006092331605032086) - rollout_corr/log_ppl_abs_diff:np.float64(0.002434113062918186) - rollout_corr/log_ppl_diff_max:np.float64(0.014789745211601257) - rollout_corr/log_ppl_diff_min:np.float64(-0.008856430649757385) - rollout_corr/ppl_ratio:np.float64(1.0006145264487714) - rollout_corr/chi2_token:np.float64(0.0011430110316723585) - rollout_corr/chi2_seq:np.float64(0.3422911234665662) - actor/pg_loss:np.float64(-0.06127674552044482) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.06984395592007786) - srpo/sdpo_loss:np.float64(0.00856721039963304) - srpo/routed_loss:np.float64(-0.06127674552044482) - srpo/routed_token_count:np.float64(156.8125) - srpo/correct_sample_fraction:np.float64(0.52734375) - srpo/sdpo_sample_fraction:np.float64(0.22265625) - srpo/grpo_sample_fraction:np.float64(0.77734375) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.01803568200011796) - self_distillation/dynamic_weight_mean:np.float64(0.22265625023283064) - self_distillation/dynamic_weight_max:np.float64(1.1267200708389282) - self_distillation/empty_target_batch:np.float64(0.265625) - actor/grad_norm:np.float64(0.30687326192855835) - perf/mfu/actor:np.float64(0.024519697714407623) - perf/max_memory_allocated_gb:np.float64(138.08052349090576) - perf/max_memory_reserved_gb:np.float64(142.578125) - perf/cpu_memory_used_gb:np.float64(148.35751724243164) - actor/lr:np.float64(5e-06) - training/global_step:52 - training/epoch:0 - critic/score/mean:0.52734375 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.52734375 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.002148515312001109 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.002148515312001109 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:156.8125 - response_length/max:539.0 - response_length/min:83.0 - response_length/clip_ra
step:53 - global_seqlen/min:12459 - global_seqlen/max:14434 - global_seqlen/minmax_diff:1975 - global_seqlen/balanced_min:13474 - global_seqlen/balanced_max:13475 - global_seqlen/mean:13474.25 - actor/entropy:0.1023220345377922 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.2821923494338989 - training/rollout_probs_diff_mean:0.003511751303449273 - training/rollout_probs_diff_std:0.013205167837440968 - training/rollout_actor_probs_pearson_corr:0.9966805577278137 - self_distillation/success_group_fraction:0.90625 - self_distillation/success_sample_fraction:0.89453125 - self_distillation/correct_sample_fraction:0.6484375 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.89453125 - rollout_corr/rollout_is_mean:0.9998244643211365 - rollout_corr/rollout_is_ratio_fraction_high:9.337068331660703e-05 - rollout_corr/rollout_is_ratio_fraction_low:9.337068331660703e-05 - rollout_corr/rollout_is_max:2.7140324115753174 - rollout_corr/rollout_is_min:0.37084463238716125 - rollout_corr/rollout_is_std:0.03603949025273323 - rollout_corr/rollout_is_eff_sample_size:0.9987023049445277 - rollout_corr/rollout_is_seq_mean:0.9999195337295532 - rollout_corr/rollout_is_seq_std:0.0034574370365589857 - rollout_corr/rollout_is_seq_max:1.0200055837631226 - rollout_corr/rollout_is_seq_min:0.9909443259239197 - rollout_corr/rollout_is_seq_max_deviation:0.02000558376312256 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.1052239537239075) - rollout_corr/training_log_ppl:np.float64(0.0989070165596786) - rollout_corr/kl:np.float64(0.0007641341020843129) - rollout_corr/k3_kl:np.float64(0.0006837092658109611) - rollout_corr/rollout_ppl:np.float64(1.1043544383719563) - rollout_corr/rollout_log_ppl:np.float64(0.09814288152119843) - rollout_corr/log_ppl_diff:np.float64(0.0007641350384801626) - rollout_corr/log_ppl_abs_diff:np.float64(0.0026271699171047658) - rollout_corr/log_ppl_diff_max:np.float64(0.010813355445861816) - rollout_corr/log_ppl_diff_min:np.float64(-0.01482805609703064) - rollout_corr/ppl_ratio:np.float64(1.0007699772249907) - rollout_corr/chi2_token:np.float64(0.0012716853525489569) - rollout_corr/chi2_seq:np.float64(0.1352834585122764) - actor/pg_loss:np.float64(-0.07643665528303245) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.08392429069499485) - srpo/sdpo_loss:np.float64(0.007487635411962401) - srpo/routed_loss:np.float64(-0.07643665528303245) - srpo/routed_token_count:np.float64(125.5078125) - srpo/correct_sample_fraction:np.float64(0.6484375) - srpo/sdpo_sample_fraction:np.float64(0.2578125) - srpo/grpo_sample_fraction:np.float64(0.7421875) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.015349582872659084) - self_distillation/dynamic_weight_mean:np.float64(0.257812503259629) - self_distillation/dynamic_weight_max:np.float64(1.1501188278198242) - self_distillation/empty_target_batch:np.float64(0.10546875) - actor/grad_norm:np.float64(0.293261855840683) - perf/mfu/actor:np.float64(0.024637082708683183) - perf/max_memory_allocated_gb:np.float64(138.08052349090576) - perf/max_memory_reserved_gb:np.float64(142.578125) - perf/cpu_memory_used_gb:np.float64(147.97124481201172) - actor/lr:np.float64(5e-06) - training/global_step:53 - training/epoch:0 - critic/score/mean:0.6484375 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.6484375 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.0005796762998215854 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.0005796762998215854 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:125.5078125 - response_length/max:304.0 - response_length/min:
step:54 - global_seqlen/min:12321 - global_seqlen/max:20079 - global_seqlen/minmax_diff:7758 - global_seqlen/balanced_min:13477 - global_seqlen/balanced_max:20966 - global_seqlen/mean:14414.125 - actor/entropy:0.08529525250196457 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.43980076909065247 - training/rollout_probs_diff_mean:0.002337590791285038 - training/rollout_probs_diff_std:0.010104293003678322 - training/rollout_actor_probs_pearson_corr:0.9978532195091248 - self_distillation/success_group_fraction:0.8125 - self_distillation/success_sample_fraction:0.8125 - self_distillation/correct_sample_fraction:0.6015625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.8125 - rollout_corr/rollout_is_mean:0.9998311400413513 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:6.522731564473361e-05 - rollout_corr/rollout_is_max:1.6179492473602295 - rollout_corr/rollout_is_min:0.4546538293361664 - rollout_corr/rollout_is_std:0.027545727789402008 - rollout_corr/rollout_is_eff_sample_size:0.9992415701059265 - rollout_corr/rollout_is_seq_mean:0.999721884727478 - rollout_corr/rollout_is_seq_std:0.0030736972112208605 - rollout_corr/rollout_is_seq_max:1.0096096992492676 - rollout_corr/rollout_is_seq_min:0.9879119396209717 - rollout_corr/rollout_is_seq_max_deviation:0.01208806037902832 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.1109966062940657) - rollout_corr/training_log_ppl:np.float64(0.1039525318210508) - rollout_corr/kl:np.float64(0.0007696261139256144) - rollout_corr/k3_kl:np.float64(0.0004914467284908852) - rollout_corr/rollout_ppl:np.float64(1.11010615946725) - rollout_corr/rollout_log_ppl:np.float64(0.10318290432223876) - rollout_corr/log_ppl_diff:np.float64(0.0007696274988120422) - rollout_corr/log_ppl_abs_diff:np.float64(0.0023289592645596713) - rollout_corr/log_ppl_diff_max:np.float64(0.016722828149795532) - rollout_corr/log_ppl_diff_min:np.float64(-0.00848647952079773) - rollout_corr/ppl_ratio:np.float64(1.0007752883248031) - rollout_corr/chi2_token:np.float64(0.0003928346559405327) - rollout_corr/chi2_seq:np.float64(0.0809183947276324) - actor/pg_loss:np.float64(-0.09340899299422745) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.09959553126827814) - srpo/sdpo_loss:np.float64(0.00618653827405069) - srpo/routed_loss:np.float64(-0.09340899299422745) - srpo/routed_token_count:np.float64(179.66015625) - srpo/correct_sample_fraction:np.float64(0.6015625) - srpo/sdpo_sample_fraction:np.float64(0.2109375) - srpo/grpo_sample_fraction:np.float64(0.7890625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.016895458040380618) - self_distillation/dynamic_weight_mean:np.float64(0.2109375016298145) - self_distillation/dynamic_weight_max:np.float64(1.206674575805664) - self_distillation/empty_target_batch:np.float64(0.1875) - actor/grad_norm:np.float64(0.26667365431785583) - perf/mfu/actor:np.float64(0.025099777709608195) - perf/max_memory_allocated_gb:np.float64(138.08052349090576) - perf/max_memory_reserved_gb:np.float64(142.578125) - perf/cpu_memory_used_gb:np.float64(147.69944763183594) - actor/lr:np.float64(5e-06) - training/global_step:54 - training/epoch:0 - critic/score/mean:0.6015625 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.6015625 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.008558367379009724 - critic/advantages/max:0.75 - critic/advantages/min:-0.875 - critic/returns/mean:-0.008558367379009724 - critic/returns/max:0.75 - critic/returns/min:-0.875 - response_length/mean:179.66015625 - response_length/max:8192.0 - response_length/min:71.0 - response_length/clip_r
step:55 - global_seqlen/min:11531 - global_seqlen/max:15579 - global_seqlen/minmax_diff:4048 - global_seqlen/balanced_min:13418 - global_seqlen/balanced_max:13421 - global_seqlen/mean:13419.875 - actor/entropy:0.11822784692049026 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.22012001276016235 - training/rollout_probs_diff_mean:0.003732734825462103 - training/rollout_probs_diff_std:0.012745111249387264 - training/rollout_actor_probs_pearson_corr:0.9973152279853821 - self_distillation/success_group_fraction:0.875 - self_distillation/success_sample_fraction:0.87109375 - self_distillation/correct_sample_fraction:0.62890625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.87109375 - rollout_corr/rollout_is_mean:0.9998881816864014 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.7928845882415771 - rollout_corr/rollout_is_min:0.5321117043495178 - rollout_corr/rollout_is_std:0.03587953746318817 - rollout_corr/rollout_is_eff_sample_size:0.9987140761927417 - rollout_corr/rollout_is_seq_mean:0.9998309016227722 - rollout_corr/rollout_is_seq_std:0.003251153975725174 - rollout_corr/rollout_is_seq_max:1.0143595933914185 - rollout_corr/rollout_is_seq_min:0.9887118935585022 - rollout_corr/rollout_is_seq_max_deviation:0.014359593391418457 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.1266704378649592) - rollout_corr/training_log_ppl:np.float64(0.11739457109069917) - rollout_corr/kl:np.float64(0.0008162727580049989) - rollout_corr/k3_kl:np.float64(0.0006471967483014929) - rollout_corr/rollout_ppl:np.float64(1.1257292004302144) - rollout_corr/rollout_log_ppl:np.float64(0.11657829743489856) - rollout_corr/log_ppl_diff:np.float64(0.0008162736558006145) - rollout_corr/log_ppl_abs_diff:np.float64(0.0025716157324495725) - rollout_corr/log_ppl_diff_max:np.float64(0.013870865106582642) - rollout_corr/log_ppl_diff_min:np.float64(-0.011896699666976929) - rollout_corr/ppl_ratio:np.float64(1.0008221475873142) - rollout_corr/chi2_token:np.float64(0.000954492250457406) - rollout_corr/chi2_seq:np.float64(0.1067100299987942) - actor/pg_loss:np.float64(-0.08238078133945237) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.09027739521116018) - srpo/sdpo_loss:np.float64(0.007896613871707814) - srpo/routed_loss:np.float64(-0.08238078133945237) - srpo/routed_token_count:np.float64(114.87109375) - srpo/correct_sample_fraction:np.float64(0.62890625) - srpo/sdpo_sample_fraction:np.float64(0.24609375) - srpo/grpo_sample_fraction:np.float64(0.75390625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.01549677197829169) - self_distillation/dynamic_weight_mean:np.float64(0.24609375093132257) - self_distillation/dynamic_weight_max:np.float64(1.1688239574432373) - self_distillation/empty_target_batch:np.float64(0.12890625) - actor/grad_norm:np.float64(0.3957875669002533) - perf/mfu/actor:np.float64(0.024916350796112156) - perf/max_memory_allocated_gb:np.float64(138.08052349090576) - perf/max_memory_reserved_gb:np.float64(142.578125) - perf/cpu_memory_used_gb:np.float64(147.52129364013672) - actor/lr:np.float64(5e-06) - training/global_step:55 - training/epoch:0 - critic/score/mean:0.62890625 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.62890625 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.0043484545312821865 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.0043484545312821865 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:114.87109375 - response_length/max:474.0 - response_length/min:70.0 - response_length/
step:56 - global_seqlen/min:10703 - global_seqlen/max:13976 - global_seqlen/minmax_diff:3273 - global_seqlen/balanced_min:12283 - global_seqlen/balanced_max:12284 - global_seqlen/mean:12283.25 - actor/entropy:0.11803721636533737 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.351860910654068 - training/rollout_probs_diff_mean:0.0034179918002337217 - training/rollout_probs_diff_std:0.0126704266294837 - training/rollout_actor_probs_pearson_corr:0.9973037838935852 - self_distillation/success_group_fraction:0.84375 - self_distillation/success_sample_fraction:0.82421875 - self_distillation/correct_sample_fraction:0.5546875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.82421875 - rollout_corr/rollout_is_mean:1.0001063346862793 - rollout_corr/rollout_is_ratio_fraction_high:3.290989116067067e-05 - rollout_corr/rollout_is_ratio_fraction_low:6.581978232134134e-05 - rollout_corr/rollout_is_max:4.691324234008789 - rollout_corr/rollout_is_min:0.43346747756004333 - rollout_corr/rollout_is_std:0.0341884046792984 - rollout_corr/rollout_is_eff_sample_size:0.9988325175244418 - rollout_corr/rollout_is_seq_mean:1.000170111656189 - rollout_corr/rollout_is_seq_std:0.00435398006811738 - rollout_corr/rollout_is_seq_max:1.0455422401428223 - rollout_corr/rollout_is_seq_min:0.9903058409690857 - rollout_corr/rollout_is_seq_max_deviation:0.045542240142822266 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.127321653533727) - rollout_corr/training_log_ppl:np.float64(0.1185595296919928) - rollout_corr/kl:np.float64(0.0005260626988743411) - rollout_corr/k3_kl:np.float64(0.0006961628373538531) - rollout_corr/rollout_ppl:np.float64(1.1267333384603262) - rollout_corr/rollout_log_ppl:np.float64(0.11803346643864643) - rollout_corr/log_ppl_diff:np.float64(0.0005260632533463649) - rollout_corr/log_ppl_abs_diff:np.float64(0.0025540911374264397) - rollout_corr/log_ppl_diff_max:np.float64(0.01078978180885315) - rollout_corr/log_ppl_diff_min:np.float64(-0.018069714307785034) - rollout_corr/ppl_ratio:np.float64(1.0005322266370058) - rollout_corr/chi2_token:np.float64(0.0021520135924220085) - rollout_corr/chi2_seq:np.float64(0.20377207617275417) - actor/pg_loss:np.float64(-0.07369380674026615) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.08298566722078249) - srpo/sdpo_loss:np.float64(0.009291860480516334) - srpo/routed_loss:np.float64(-0.07369380674026615) - srpo/routed_token_count:np.float64(118.6953125) - srpo/correct_sample_fraction:np.float64(0.5546875) - srpo/sdpo_sample_fraction:np.float64(0.2890625) - srpo/grpo_sample_fraction:np.float64(0.7109375) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.02187979042423649) - self_distillation/dynamic_weight_mean:np.float64(0.289062503259629) - self_distillation/dynamic_weight_max:np.float64(1.1496392488479614) - self_distillation/empty_target_batch:np.float64(0.17578125) - actor/grad_norm:np.float64(0.2850371301174164) - perf/mfu/actor:np.float64(0.022561538426933148) - perf/max_memory_allocated_gb:np.float64(138.08052349090576) - perf/max_memory_reserved_gb:np.float64(142.578125) - perf/cpu_memory_used_gb:np.float64(147.57619857788086) - actor/lr:np.float64(5e-06) - training/global_step:56 - training/epoch:0 - critic/score/mean:0.5546875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.5546875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.002657473785802722 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.002657473785802722 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:118.6953125 - response_length/max:391.0 - response_length/min:63.0 -
step:57 - global_seqlen/min:11315 - global_seqlen/max:15498 - global_seqlen/minmax_diff:4183 - global_seqlen/balanced_min:12766 - global_seqlen/balanced_max:12767 - global_seqlen/mean:12766.75 - actor/entropy:0.12109623849391937 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.24023127555847168 - training/rollout_probs_diff_mean:0.0031913956627249718 - training/rollout_probs_diff_std:0.011290859431028366 - training/rollout_actor_probs_pearson_corr:0.9979162216186523 - self_distillation/success_group_fraction:0.9375 - self_distillation/success_sample_fraction:0.93359375 - self_distillation/correct_sample_fraction:0.671875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.93359375 - rollout_corr/rollout_is_mean:1.0000444650650024 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.9333796501159668 - rollout_corr/rollout_is_min:0.6036461591720581 - rollout_corr/rollout_is_std:0.032565101981163025 - rollout_corr/rollout_is_eff_sample_size:0.9989407565558476 - rollout_corr/rollout_is_seq_mean:1.00005042552948 - rollout_corr/rollout_is_seq_std:0.003176225582137704 - rollout_corr/rollout_is_seq_max:1.013447880744934 - rollout_corr/rollout_is_seq_min:0.9925165772438049 - rollout_corr/rollout_is_seq_max_deviation:0.013447880744934082 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.1318314280360937) - rollout_corr/training_log_ppl:np.float64(0.12202853620692622) - rollout_corr/kl:np.float64(0.00045980290201441676) - rollout_corr/k3_kl:np.float64(0.0005102248091120032) - rollout_corr/rollout_ppl:np.float64(1.1312983836978674) - rollout_corr/rollout_log_ppl:np.float64(0.1215687331132358) - rollout_corr/log_ppl_diff:np.float64(0.00045980309369042516) - rollout_corr/log_ppl_abs_diff:np.float64(0.0024228324909927323) - rollout_corr/log_ppl_diff_max:np.float64(0.008105307817459106) - rollout_corr/log_ppl_diff_min:np.float64(-0.011362016201019287) - rollout_corr/ppl_ratio:np.float64(1.0004647497553378) - rollout_corr/chi2_token:np.float64(0.0011328961700201035) - rollout_corr/chi2_seq:np.float64(0.20052963076159358) - actor/pg_loss:np.float64(-0.10529280213631864) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.11513252885197289) - srpo/sdpo_loss:np.float64(0.009839726715654251) - srpo/routed_loss:np.float64(-0.10529280213631864) - srpo/routed_token_count:np.float64(112.1796875) - srpo/correct_sample_fraction:np.float64(0.671875) - srpo/sdpo_sample_fraction:np.float64(0.265625) - srpo/grpo_sample_fraction:np.float64(0.734375) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.021176218283159187) - self_distillation/dynamic_weight_mean:np.float64(0.26562500302679837) - self_distillation/dynamic_weight_max:np.float64(1.1410025358200073) - self_distillation/empty_target_batch:np.float64(0.06640625) - actor/grad_norm:np.float64(0.32055988907814026) - perf/mfu/actor:np.float64(0.0237718100842549) - perf/max_memory_allocated_gb:np.float64(138.08052349090576) - perf/max_memory_reserved_gb:np.float64(142.578125) - perf/cpu_memory_used_gb:np.float64(147.61974716186523) - actor/lr:np.float64(5e-06) - training/global_step:57 - training/epoch:0 - critic/score/mean:0.671875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.671875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.00047008844558149576 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.00047008844558149576 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:112.1796875 - response_length/max:210.0 - response_length/min:69.0 - response_length/clip_ratio
step:58 - global_seqlen/min:11448 - global_seqlen/max:14862 - global_seqlen/minmax_diff:3414 - global_seqlen/balanced_min:12755 - global_seqlen/balanced_max:12756 - global_seqlen/mean:12755.25 - actor/entropy:0.1159072145819664 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.12873855233192444 - training/rollout_probs_diff_mean:0.0029940232634544373 - training/rollout_probs_diff_std:0.010456147603690624 - training/rollout_actor_probs_pearson_corr:0.9981803894042969 - self_distillation/success_group_fraction:0.9375 - self_distillation/success_sample_fraction:0.9296875 - self_distillation/correct_sample_fraction:0.71875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.9296875 - rollout_corr/rollout_is_mean:1.0000183582305908 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.5303785800933838 - rollout_corr/rollout_is_min:0.608140230178833 - rollout_corr/rollout_is_std:0.03008977510035038 - rollout_corr/rollout_is_eff_sample_size:0.9990954244436443 - rollout_corr/rollout_is_seq_mean:1.000069260597229 - rollout_corr/rollout_is_seq_std:0.0029559244867414236 - rollout_corr/rollout_is_seq_max:1.0103813409805298 - rollout_corr/rollout_is_seq_min:0.9891666769981384 - rollout_corr/rollout_is_seq_max_deviation:0.010833323001861572 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.129822202026844) - rollout_corr/training_log_ppl:np.float64(0.12048393143777503) - rollout_corr/kl:np.float64(0.0004010177552196126) - rollout_corr/k3_kl:np.float64(0.0004702733591841479) - rollout_corr/rollout_ppl:np.float64(1.129367459565401) - rollout_corr/rollout_log_ppl:np.float64(0.12008291299571283) - rollout_corr/log_ppl_diff:np.float64(0.0004010184420621954) - rollout_corr/log_ppl_abs_diff:np.float64(0.0022565912586287595) - rollout_corr/log_ppl_diff_max:np.float64(0.011820018291473389) - rollout_corr/log_ppl_diff_min:np.float64(-0.009444102644920349) - rollout_corr/ppl_ratio:np.float64(1.0004054717719555) - rollout_corr/chi2_token:np.float64(0.0010778356809169054) - rollout_corr/chi2_seq:np.float64(0.11221968871541321) - actor/pg_loss:np.float64(-0.0866720872654696) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.09377511739148758) - srpo/sdpo_loss:np.float64(0.007103030126017984) - srpo/routed_loss:np.float64(-0.0866720872654696) - srpo/routed_token_count:np.float64(121.7578125) - srpo/correct_sample_fraction:np.float64(0.71875) - srpo/sdpo_sample_fraction:np.float64(0.21875) - srpo/grpo_sample_fraction:np.float64(0.78125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.014643846483522793) - self_distillation/dynamic_weight_mean:np.float64(0.21875000116415322) - self_distillation/dynamic_weight_max:np.float64(1.1369699239730835) - self_distillation/empty_target_batch:np.float64(0.0703125) - actor/grad_norm:np.float64(0.25582170486450195) - perf/mfu/actor:np.float64(0.023389584060081847) - perf/max_memory_allocated_gb:np.float64(138.08052349090576) - perf/max_memory_reserved_gb:np.float64(142.578125) - perf/cpu_memory_used_gb:np.float64(147.6235809326172) - actor/lr:np.float64(5e-06) - training/global_step:58 - training/epoch:0 - critic/score/mean:0.71875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.71875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.006392364390194416 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.006392364390194416 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:121.7578125 - response_length/max:376.0 - response_length/min:57.0 - response_length/clip_ratio:0.0 - response_l
step:59 - global_seqlen/min:11249 - global_seqlen/max:15303 - global_seqlen/minmax_diff:4054 - global_seqlen/balanced_min:12696 - global_seqlen/balanced_max:12697 - global_seqlen/mean:12696.625 - actor/entropy:0.12942281365394592 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.15452197194099426 - training/rollout_probs_diff_mean:0.0028892590198665857 - training/rollout_probs_diff_std:0.009877903386950493 - training/rollout_actor_probs_pearson_corr:0.9984943866729736 - self_distillation/success_group_fraction:0.90625 - self_distillation/success_sample_fraction:0.89453125 - self_distillation/correct_sample_fraction:0.57421875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.89453125 - rollout_corr/rollout_is_mean:0.9999637007713318 - rollout_corr/rollout_is_ratio_fraction_high:3.0468298064079136e-05 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:2.1129724979400635 - rollout_corr/rollout_is_min:0.5899034142494202 - rollout_corr/rollout_is_std:0.028523733839392662 - rollout_corr/rollout_is_eff_sample_size:0.999186819980144 - rollout_corr/rollout_is_seq_mean:0.9998252391815186 - rollout_corr/rollout_is_seq_std:0.0027629637625068426 - rollout_corr/rollout_is_seq_max:1.0072944164276123 - rollout_corr/rollout_is_seq_min:0.9922807812690735 - rollout_corr/rollout_is_seq_max_deviation:0.007719218730926514 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.139904618728906) - rollout_corr/training_log_ppl:np.float64(0.1293062344120699) - rollout_corr/kl:np.float64(0.0005876044120043389) - rollout_corr/k3_kl:np.float64(0.00041284070293912123) - rollout_corr/rollout_ppl:np.float64(1.139193370938301) - rollout_corr/rollout_log_ppl:np.float64(0.12871862991596572) - rollout_corr/log_ppl_diff:np.float64(0.0005876044961041771) - rollout_corr/log_ppl_abs_diff:np.float64(0.0022358322821673937) - rollout_corr/log_ppl_diff_max:np.float64(0.008923619985580444) - rollout_corr/log_ppl_diff_min:np.float64(-0.006008081138134003) - rollout_corr/ppl_ratio:np.float64(1.0005916766822338) - rollout_corr/chi2_token:np.float64(0.0004835049621760845) - rollout_corr/chi2_seq:np.float64(0.10821353434585035) - actor/pg_loss:np.float64(-0.09862393164075911) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.10979278237209655) - srpo/sdpo_loss:np.float64(0.011168850731337443) - srpo/routed_loss:np.float64(-0.09862393164075911) - srpo/routed_token_count:np.float64(128.20703125) - srpo/correct_sample_fraction:np.float64(0.57421875) - srpo/sdpo_sample_fraction:np.float64(0.33203125) - srpo/grpo_sample_fraction:np.float64(0.66796875) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.027053151092331973) - self_distillation/dynamic_weight_mean:np.float64(0.3320312562864274) - self_distillation/dynamic_weight_max:np.float64(1.1448570489883423) - self_distillation/empty_target_batch:np.float64(0.10546875) - actor/grad_norm:np.float64(0.25658467411994934) - perf/mfu/actor:np.float64(0.02328989911572101) - perf/max_memory_allocated_gb:np.float64(138.08052349090576) - perf/max_memory_reserved_gb:np.float64(142.578125) - perf/cpu_memory_used_gb:np.float64(147.52571487426758) - actor/lr:np.float64(5e-06) - training/global_step:59 - training/epoch:0 - critic/score/mean:0.57421875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.57421875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.0021670577116310596 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.0021670577116310596 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:128.20703125 - response_length/max:442.0 - response_length/min:
test_gen_batch meta info: {'eos_token_id': 151645, 'pad_token_id': 151643, 'recompute_log_prob': False, 'do_sample': True, 'validate': True, 'global_steps': 60}
validation generation end
[prompt] system

Given a question and four options, please select the right answer. Respond in the following format:
<reasoning>
...
</reasoning>
<answer>
...
</answer>

For the answer, only output the letter corresponding to the correct option (A, B, C, or D), and nothing else. Do not restate the answer text. For example, if the answer is "A", just output:
<answer>
A
</answer>

user
Which of the following molecules is the only correct reactant used in the synthesis of the product "Nc1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1" ?

A: O=C1c2ccccc2C(=O)N1c1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1
B: O=C1c2ccccc2C(Cl)N1c1ccc2ccc(Cl)nc2n1
C: O=C1c2ccccc2C(=O)N1Cc1ccc(Oc2ccc(Cl)c(C(F)(F)F)c2)cc1
D: O=C1c2ccccc2C(=O)N1CC=C(Cc1ccccc1)C(F)(F)F
Please reason step by step.
assistant
<think>

</think>


[response] <reasoning>
The product "Nc1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1" contains a nitrile group (n1), a chlorinated benzene ring (Cl), and a trifluorinated group (C(F)(F)F). Option A contains a nitrile group, a chlorinated benzene ring, and a trifluorinated group, making it the correct answer. Option B contains a nitrile group and a chlorinated benzene ring, but lacks the trifluorinated group. Option C contains a nitrile group and a chlorinated benzene ring, but lacks the trifluorinated group. Option D contains a nitrile group and a chlorinated benzene ring, but lacks the trifluorinated group. Therefore, the correct answer is A.
</reasoning>
<answer>
A
</answer>
[ground_truth] A
[score] 1.0
[acc] 1.0
[pred] A
[incorrect_format] 1
[feedback]
local_global_step_folder: /workspace/L2T/checkpoints/datasets/sciknoweval/chemistry/qwen3gen-chemistry-SRPO_TR-Qwen-Qwen3-4B-mbs32-tr0.1-dwtrue-train32-rollout8-lr5e-6-vllm0.8/global_step_60
step:60 - global_seqlen/min:9988 - global_seqlen/max:13328 - global_seqlen/minmax_diff:3340 - global_seqlen/balanced_min:12147 - global_seqlen/balanced_max:12149 - global_seqlen/mean:12148.625 - actor/entropy:0.1317128837108612 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.15475644171237946 - training/rollout_probs_diff_mean:0.003006047336384654 - training/rollout_probs_diff_std:0.010005713440477848 - training/rollout_actor_probs_pearson_corr:0.998454749584198 - self_distillation/success_group_fraction:0.90625 - self_distillation/success_sample_fraction:0.90234375 - self_distillation/correct_sample_fraction:0.62890625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.90234375 - rollout_corr/rollout_is_mean:0.9999188184738159 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.7512764930725098 - rollout_corr/rollout_is_min:0.58980792760849 - rollout_corr/rollout_is_std:0.029004482552409172 - rollout_corr/rollout_is_eff_sample_size:0.9991592091491956 - rollout_corr/rollout_is_seq_mean:0.9998779296875 - rollout_corr/rollout_is_seq_std:0.0028278687968850136 - rollout_corr/rollout_is_seq_max:1.0075337886810303 - rollout_corr/rollout_is_seq_min:0.9894755482673645 - rollout_corr/rollout_is_seq_max_deviation:0.010524451732635498 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.1453012214042246) - rollout_corr/training_log_ppl:np.float64(0.13366957208199892) - rollout_corr/kl:np.float64(0.0005429412622541818) - rollout_corr/k3_kl:np.float64(0.00042084956529819806) - rollout_corr/rollout_ppl:np.float64(1.1446624575182796) - rollout_corr/rollout_log_ppl:np.float64(0.13312662898533745) - rollout_corr/log_ppl_diff:np.float64(0.0005429430966614746) - rollout_corr/log_ppl_abs_diff:np.float64(0.002194549459090922) - rollout_corr/log_ppl_diff_max:np.float64(0.012230396270751953) - rollout_corr/log_ppl_diff_min:np.float64(-0.006915971636772156) - rollout_corr/ppl_ratio:np.float64(1.0005470074247569) - rollout_corr/chi2_token:np.float64(0.0006203684024512768) - rollout_corr/chi2_seq:np.float64(0.06786972074769437) - actor/pg_loss:np.float64(-0.1263862499035895) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.13622776870033704) - srpo/sdpo_loss:np.float64(0.00984151879674755) - srpo/routed_loss:np.float64(-0.1263862499035895) - srpo/routed_token_count:np.float64(113.70703125) - srpo/correct_sample_fraction:np.float64(0.62890625) - srpo/sdpo_sample_fraction:np.float64(0.27734375) - srpo/grpo_sample_fraction:np.float64(0.72265625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.021201108055720397) - self_distillation/dynamic_weight_mean:np.float64(0.27734375512227416) - self_distillation/dynamic_weight_max:np.float64(1.131517767906189) - self_distillation/empty_target_batch:np.float64(0.09765625) - actor/grad_norm:np.float64(0.25041285157203674) - perf/mfu/actor:np.float64(0.022455052904282336) - perf/max_memory_allocated_gb:np.float64(138.08052349090576) - perf/max_memory_reserved_gb:np.float64(142.578125) - perf/cpu_memory_used_gb:np.float64(148.4881706237793) - actor/lr:np.float64(5e-06) - val-aux/sciknoweval/reward/mean@16:np.float64(0.6910714285714286) - val-aux/sciknoweval/reward/std@16:np.float64(0.21357583696111587) - val-aux/sciknoweval/reward/best@2/mean:np.float64(0.7799428571428572) - val-aux/sciknoweval/reward/best@2/std:np.float64(0.17938948996874435) - val-aux/sciknoweval/reward/worst@2/mean:np.float64(0.6014142857142857) - val-aux/sciknoweval/reward/worst@2/std:np.float64(0.18200643506483513) - val-aux/sciknoweval/reward/maj@2/mean:np.float64(0.6901333333333333) - val-aux/sciknoweval/reward/ma
step:61 - global_seqlen/min:10195 - global_seqlen/max:14926 - global_seqlen/minmax_diff:4731 - global_seqlen/balanced_min:12858 - global_seqlen/balanced_max:12860 - global_seqlen/mean:12858.875 - actor/entropy:0.13106049597263336 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.17945119738578796 - training/rollout_probs_diff_mean:0.0027389235328882933 - training/rollout_probs_diff_std:0.00912692304700613 - training/rollout_actor_probs_pearson_corr:0.9987083077430725 - self_distillation/success_group_fraction:0.96875 - self_distillation/success_sample_fraction:0.9453125 - self_distillation/correct_sample_fraction:0.60546875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.9453125 - rollout_corr/rollout_is_mean:0.9999050498008728 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.4709558486938477 - rollout_corr/rollout_is_min:0.5900130271911621 - rollout_corr/rollout_is_std:0.02607623115181923 - rollout_corr/rollout_is_eff_sample_size:0.99932025416252 - rollout_corr/rollout_is_seq_mean:0.999906063079834 - rollout_corr/rollout_is_seq_std:0.002446659840643406 - rollout_corr/rollout_is_seq_max:1.0085999965667725 - rollout_corr/rollout_is_seq_min:0.9912810325622559 - rollout_corr/rollout_is_seq_max_deviation:0.00871896743774414 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.1467761313542724) - rollout_corr/training_log_ppl:np.float64(0.13494996858935338) - rollout_corr/kl:np.float64(0.0004585390187514804) - rollout_corr/k3_kl:np.float64(0.0003646000079626788) - rollout_corr/rollout_ppl:np.float64(1.1462372890673578) - rollout_corr/rollout_log_ppl:np.float64(0.134491428732872) - rollout_corr/log_ppl_diff:np.float64(0.00045853985648136586) - rollout_corr/log_ppl_abs_diff:np.float64(0.0018609620892675593) - rollout_corr/log_ppl_diff_max:np.float64(0.00957643985748291) - rollout_corr/log_ppl_diff_min:np.float64(-0.00790812075138092) - rollout_corr/ppl_ratio:np.float64(1.0004616782534868) - rollout_corr/chi2_token:np.float64(0.0005349400453269482) - rollout_corr/chi2_seq:np.float64(0.0634721436072141) - actor/pg_loss:np.float64(-0.11780246235503) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.13032975635724142) - srpo/sdpo_loss:np.float64(0.01252729400221142) - srpo/routed_loss:np.float64(-0.11780246235503) - srpo/routed_token_count:np.float64(139.80859375) - srpo/correct_sample_fraction:np.float64(0.60546875) - srpo/sdpo_sample_fraction:np.float64(0.36328125) - srpo/grpo_sample_fraction:np.float64(0.63671875) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.030685402077324397) - self_distillation/dynamic_weight_mean:np.float64(0.36328125349245965) - self_distillation/dynamic_weight_max:np.float64(1.1727339029312134) - self_distillation/empty_target_batch:np.float64(0.0546875) - actor/grad_norm:np.float64(0.24394525587558746) - perf/mfu/actor:np.float64(0.02373782884780256) - perf/max_memory_allocated_gb:np.float64(138.08052349090576) - perf/max_memory_reserved_gb:np.float64(142.578125) - perf/cpu_memory_used_gb:np.float64(149.89813613891602) - actor/lr:np.float64(5e-06) - training/global_step:61 - training/epoch:1 - critic/score/mean:0.60546875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.60546875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.014937414787709713 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.014937414787709713 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:139.80859375 - response_length/max:470.0 - response_length/min:58.0 - response_length/clip_ratio:0.0
step:62 - global_seqlen/min:11328 - global_seqlen/max:16143 - global_seqlen/minmax_diff:4815 - global_seqlen/balanced_min:13393 - global_seqlen/balanced_max:13394 - global_seqlen/mean:13393.5 - actor/entropy:0.13022996485233307 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.12254580855369568 - training/rollout_probs_diff_mean:0.0027916892431676388 - training/rollout_probs_diff_std:0.009310689754784107 - training/rollout_actor_probs_pearson_corr:0.998661994934082 - self_distillation/success_group_fraction:0.90625 - self_distillation/success_sample_fraction:0.90234375 - self_distillation/correct_sample_fraction:0.60546875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.90234375 - rollout_corr/rollout_is_mean:0.999879002571106 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.4539344310760498 - rollout_corr/rollout_is_min:0.690205991268158 - rollout_corr/rollout_is_std:0.026178888976573944 - rollout_corr/rollout_is_eff_sample_size:0.9993150161096298 - rollout_corr/rollout_is_seq_mean:0.9999661445617676 - rollout_corr/rollout_is_seq_std:0.00277000037021935 - rollout_corr/rollout_is_seq_max:1.0085846185684204 - rollout_corr/rollout_is_seq_min:0.9901068210601807 - rollout_corr/rollout_is_seq_max_deviation:0.009893178939819336 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.141571689862758) - rollout_corr/training_log_ppl:np.float64(0.12977703064098023) - rollout_corr/kl:np.float64(0.00038943755499332156) - rollout_corr/k3_kl:np.float64(0.0003556188013700279) - rollout_corr/rollout_ppl:np.float64(1.1410978641360998) - rollout_corr/rollout_log_ppl:np.float64(0.12938759160897462) - rollout_corr/log_ppl_diff:np.float64(0.0003894390320056118) - rollout_corr/log_ppl_abs_diff:np.float64(0.002063585510768462) - rollout_corr/log_ppl_diff_max:np.float64(0.01108473539352417) - rollout_corr/log_ppl_diff_min:np.float64(-0.00767633318901062) - rollout_corr/ppl_ratio:np.float64(1.0003933960106224) - rollout_corr/chi2_token:np.float64(0.0006457723211497068) - rollout_corr/chi2_seq:np.float64(0.049512520898133516) - actor/pg_loss:np.float64(-0.1007696825581661) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.11082126054679975) - srpo/sdpo_loss:np.float64(0.010051577988633653) - srpo/routed_loss:np.float64(-0.1007696825581661) - srpo/routed_token_count:np.float64(119.296875) - srpo/correct_sample_fraction:np.float64(0.60546875) - srpo/sdpo_sample_fraction:np.float64(0.30078125) - srpo/grpo_sample_fraction:np.float64(0.69921875) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.02261198448104551) - self_distillation/dynamic_weight_mean:np.float64(0.30078125186264515) - self_distillation/dynamic_weight_max:np.float64(1.1469060182571411) - self_distillation/empty_target_batch:np.float64(0.09765625) - actor/grad_norm:np.float64(0.2539532780647278) - perf/mfu/actor:np.float64(0.0245023458438828) - perf/max_memory_allocated_gb:np.float64(138.08052349090576) - perf/max_memory_reserved_gb:np.float64(142.578125) - perf/cpu_memory_used_gb:np.float64(149.47189474105835) - actor/lr:np.float64(5e-06) - training/global_step:62 - training/epoch:1 - critic/score/mean:0.60546875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.60546875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.00509986886754632 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.00509986886754632 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:119.296875 - response_length/max:371.0 - response_length/min:57.0 - response_length/clip_ratio:0.0
step:63 - global_seqlen/min:10443 - global_seqlen/max:14352 - global_seqlen/minmax_diff:3909 - global_seqlen/balanced_min:12835 - global_seqlen/balanced_max:12837 - global_seqlen/mean:12836.25 - actor/entropy:0.13413648307323456 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.3074401021003723 - training/rollout_probs_diff_mean:0.002950686728581786 - training/rollout_probs_diff_std:0.00987362489104271 - training/rollout_actor_probs_pearson_corr:0.9985501766204834 - self_distillation/success_group_fraction:1.0 - self_distillation/success_sample_fraction:0.97265625 - self_distillation/correct_sample_fraction:0.53125 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.97265625 - rollout_corr/rollout_is_mean:0.9998690485954285 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.420251727104187 - rollout_corr/rollout_is_min:0.5253848433494568 - rollout_corr/rollout_is_std:0.02808147855103016 - rollout_corr/rollout_is_eff_sample_size:0.9992118138584731 - rollout_corr/rollout_is_seq_mean:1.0000137090682983 - rollout_corr/rollout_is_seq_std:0.002938607707619667 - rollout_corr/rollout_is_seq_max:1.0102858543395996 - rollout_corr/rollout_is_seq_min:0.9912541508674622 - rollout_corr/rollout_is_seq_max_deviation:0.01028585433959961 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.1449209083802998) - rollout_corr/training_log_ppl:np.float64(0.13339029447524808) - rollout_corr/kl:np.float64(0.00040184970444734347) - rollout_corr/k3_kl:np.float64(0.0004155050316860809) - rollout_corr/rollout_ppl:np.float64(1.1444549350999296) - rollout_corr/rollout_log_ppl:np.float64(0.1329884431543178) - rollout_corr/log_ppl_diff:np.float64(0.00040185132093029097) - rollout_corr/log_ppl_abs_diff:np.float64(0.002304590649146121) - rollout_corr/log_ppl_diff_max:np.float64(0.00954032689332962) - rollout_corr/log_ppl_diff_min:np.float64(-0.009323403239250183) - rollout_corr/ppl_ratio:np.float64(1.0004062447696924) - rollout_corr/chi2_token:np.float64(0.0008591224905103445) - rollout_corr/chi2_seq:np.float64(0.08687273063696921) - actor/pg_loss:np.float64(-0.12586119863408385) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.1425246056751348) - srpo/sdpo_loss:np.float64(0.01666340704105096) - srpo/routed_loss:np.float64(-0.12586119863408385) - srpo/routed_token_count:np.float64(114.9140625) - srpo/correct_sample_fraction:np.float64(0.53125) - srpo/sdpo_sample_fraction:np.float64(0.46875) - srpo/grpo_sample_fraction:np.float64(0.53125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.03694716752761451) - self_distillation/dynamic_weight_mean:np.float64(0.46875000069849193) - self_distillation/dynamic_weight_max:np.float64(1.190338373184204) - self_distillation/empty_target_batch:np.float64(0.02734375) - actor/grad_norm:np.float64(0.35834264755249023) - perf/mfu/actor:np.float64(0.023268703156476785) - perf/max_memory_allocated_gb:np.float64(138.08052349090576) - perf/max_memory_reserved_gb:np.float64(142.578125) - perf/cpu_memory_used_gb:np.float64(148.87540435791016) - actor/lr:np.float64(5e-06) - training/global_step:63 - training/epoch:1 - critic/score/mean:0.53125 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.53125 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.01020633615553379 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.01020633615553379 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:114.9140625 - response_length/max:420.0 - response_length/min:66.0 - response_length/clip_ratio:0.0 - response_length_n
step:64 - global_seqlen/min:10407 - global_seqlen/max:15782 - global_seqlen/minmax_diff:5375 - global_seqlen/balanced_min:13089 - global_seqlen/balanced_max:13091 - global_seqlen/mean:13090.5 - actor/entropy:0.1236448660492897 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.21511512994766235 - training/rollout_probs_diff_mean:0.002911769086495042 - training/rollout_probs_diff_std:0.010177098214626312 - training/rollout_actor_probs_pearson_corr:0.9983313083648682 - self_distillation/success_group_fraction:1.0 - self_distillation/success_sample_fraction:0.99609375 - self_distillation/correct_sample_fraction:0.76171875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.99609375 - rollout_corr/rollout_is_mean:0.9999525547027588 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.758646845817566 - rollout_corr/rollout_is_min:0.5874456763267517 - rollout_corr/rollout_is_std:0.028885044157505035 - rollout_corr/rollout_is_eff_sample_size:0.9991662307244454 - rollout_corr/rollout_is_seq_mean:0.9999455213546753 - rollout_corr/rollout_is_seq_std:0.003133178222924471 - rollout_corr/rollout_is_seq_max:1.007641315460205 - rollout_corr/rollout_is_seq_min:0.9864470362663269 - rollout_corr/rollout_is_seq_max_deviation:0.013552963733673096 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.1381661905907094) - rollout_corr/training_log_ppl:np.float64(0.12756818887282861) - rollout_corr/kl:np.float64(0.00048041942238796764) - rollout_corr/k3_kl:np.float64(0.0004259191480393554) - rollout_corr/rollout_ppl:np.float64(1.1375750806182623) - rollout_corr/rollout_log_ppl:np.float64(0.1270877684946754) - rollout_corr/log_ppl_diff:np.float64(0.00048042037815321237) - rollout_corr/log_ppl_abs_diff:np.float64(0.002432261113426648) - rollout_corr/log_ppl_diff_max:np.float64(0.016223162412643433) - rollout_corr/log_ppl_diff_min:np.float64(-0.006857722997665405) - rollout_corr/ppl_ratio:np.float64(1.000485630473122) - rollout_corr/chi2_token:np.float64(0.0007361655589193106) - rollout_corr/chi2_seq:np.float64(0.10944761778227985) - actor/pg_loss:np.float64(-0.10885798497474752) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.1166642296302598) - srpo/sdpo_loss:np.float64(0.007806244655512273) - srpo/routed_loss:np.float64(-0.10885798497474752) - srpo/routed_token_count:np.float64(116.328125) - srpo/correct_sample_fraction:np.float64(0.76171875) - srpo/sdpo_sample_fraction:np.float64(0.23828125) - srpo/grpo_sample_fraction:np.float64(0.76171875) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.019262472474110837) - self_distillation/dynamic_weight_mean:np.float64(0.23828125093132257) - self_distillation/dynamic_weight_max:np.float64(1.1325981616973877) - self_distillation/empty_target_batch:np.float64(0.00390625) - actor/grad_norm:np.float64(0.24941876530647278) - perf/mfu/actor:np.float64(0.024155413114807442) - perf/max_memory_allocated_gb:np.float64(138.08052349090576) - perf/max_memory_reserved_gb:np.float64(142.578125) - perf/cpu_memory_used_gb:np.float64(148.61318588256836) - actor/lr:np.float64(5e-06) - training/global_step:64 - training/epoch:1 - critic/score/mean:0.76171875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.76171875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.0025268637109547853 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.0025268637109547853 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:116.328125 - response_length/max:317.0 - response_length/min:52.0 - response_length/clip_rat
step:65 - global_seqlen/min:11444 - global_seqlen/max:16518 - global_seqlen/minmax_diff:5074 - global_seqlen/balanced_min:13126 - global_seqlen/balanced_max:13216 - global_seqlen/mean:13137.25 - actor/entropy:0.11462864279747009 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.18287068605422974 - training/rollout_probs_diff_mean:0.002903456799685955 - training/rollout_probs_diff_std:0.010166282765567303 - training/rollout_actor_probs_pearson_corr:0.9981832504272461 - self_distillation/success_group_fraction:0.9375 - self_distillation/success_sample_fraction:0.9296875 - self_distillation/correct_sample_fraction:0.7265625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.9296875 - rollout_corr/rollout_is_mean:1.0000417232513428 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:3.077112341998145e-05 - rollout_corr/rollout_is_max:1.5711655616760254 - rollout_corr/rollout_is_min:0.45874813199043274 - rollout_corr/rollout_is_std:0.02812177874147892 - rollout_corr/rollout_is_eff_sample_size:0.9992097904979987 - rollout_corr/rollout_is_seq_mean:1.0001335144042969 - rollout_corr/rollout_is_seq_std:0.0026986112352460623 - rollout_corr/rollout_is_seq_max:1.008333444595337 - rollout_corr/rollout_is_seq_min:0.9897950291633606 - rollout_corr/rollout_is_seq_max_deviation:0.010204970836639404 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.1242433553561568) - rollout_corr/training_log_ppl:np.float64(0.11543759734922787) - rollout_corr/kl:np.float64(0.0002704004113898151) - rollout_corr/k3_kl:np.float64(0.0004038849562508773) - rollout_corr/rollout_ppl:np.float64(1.1239047637209296) - rollout_corr/rollout_log_ppl:np.float64(0.11516719611245207) - rollout_corr/log_ppl_diff:np.float64(0.00027040123677579686) - rollout_corr/log_ppl_abs_diff:np.float64(0.0020363678704597987) - rollout_corr/log_ppl_diff_max:np.float64(0.012892752885818481) - rollout_corr/log_ppl_diff_min:np.float64(-0.007729820907115936) - rollout_corr/ppl_ratio:np.float64(1.0002742123324424) - rollout_corr/chi2_token:np.float64(0.0010756622068583965) - rollout_corr/chi2_seq:np.float64(0.11025207559578121) - actor/pg_loss:np.float64(-0.0861596555550932) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.09279226802755147) - srpo/sdpo_loss:np.float64(0.006632612472458277) - srpo/routed_loss:np.float64(-0.0861596555550932) - srpo/routed_token_count:np.float64(126.9453125) - srpo/correct_sample_fraction:np.float64(0.7265625) - srpo/sdpo_sample_fraction:np.float64(0.2109375) - srpo/grpo_sample_fraction:np.float64(0.7890625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.015819127086359686) - self_distillation/dynamic_weight_mean:np.float64(0.2109375020954758) - self_distillation/dynamic_weight_max:np.float64(1.1518150568008423) - self_distillation/empty_target_batch:np.float64(0.0703125) - actor/grad_norm:np.float64(0.20970243215560913) - perf/mfu/actor:np.float64(0.024158373167751872) - perf/max_memory_allocated_gb:np.float64(138.08052349090576) - perf/max_memory_reserved_gb:np.float64(142.578125) - perf/cpu_memory_used_gb:np.float64(148.56084442138672) - actor/lr:np.float64(5e-06) - training/global_step:65 - training/epoch:1 - critic/score/mean:0.7265625 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.7265625 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.0055580344051122665 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.0055580344051122665 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:126.9453125 - response_length/max:537.0 - response_length/min:51.0 - response
step:66 - global_seqlen/min:11871 - global_seqlen/max:14579 - global_seqlen/minmax_diff:2708 - global_seqlen/balanced_min:13254 - global_seqlen/balanced_max:13256 - global_seqlen/mean:13255.0 - actor/entropy:0.10008011758327484 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.16210246086120605 - training/rollout_probs_diff_mean:0.0025754745583981276 - training/rollout_probs_diff_std:0.009848104789853096 - training/rollout_actor_probs_pearson_corr:0.9980593323707581 - self_distillation/success_group_fraction:0.875 - self_distillation/success_sample_fraction:0.859375 - self_distillation/correct_sample_fraction:0.5703125 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.859375 - rollout_corr/rollout_is_mean:1.000096082687378 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:3.15099568979349e-05 - rollout_corr/rollout_is_max:1.8257331848144531 - rollout_corr/rollout_is_min:0.3049869239330292 - rollout_corr/rollout_is_std:0.028632190078496933 - rollout_corr/rollout_is_eff_sample_size:0.9991811072692304 - rollout_corr/rollout_is_seq_mean:1.0001782178878784 - rollout_corr/rollout_is_seq_std:0.0030022801365703344 - rollout_corr/rollout_is_seq_max:1.0119925737380981 - rollout_corr/rollout_is_seq_min:0.9901673197746277 - rollout_corr/rollout_is_seq_max_deviation:0.011992573738098145 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.1067535444162786) - rollout_corr/training_log_ppl:np.float64(0.10014134936500341) - rollout_corr/kl:np.float64(0.00024277800371663716) - rollout_corr/k3_kl:np.float64(0.00042114363462975035) - rollout_corr/rollout_ppl:np.float64(1.1064914432354271) - rollout_corr/rollout_log_ppl:np.float64(0.09989857126856805) - rollout_corr/log_ppl_diff:np.float64(0.00024277809643535875) - rollout_corr/log_ppl_abs_diff:np.float64(0.002234711853816407) - rollout_corr/log_ppl_diff_max:np.float64(0.011538147926330566) - rollout_corr/log_ppl_diff_min:np.float64(-0.009736910462379456) - rollout_corr/ppl_ratio:np.float64(1.0002472244668752) - rollout_corr/chi2_token:np.float64(0.0012093971017748117) - rollout_corr/chi2_seq:np.float64(0.2719298067968339) - actor/pg_loss:np.float64(-0.0917358262304333) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.10163980207289569) - srpo/sdpo_loss:np.float64(0.009903975842462387) - srpo/routed_loss:np.float64(-0.0917358262304333) - srpo/routed_token_count:np.float64(123.96875) - srpo/correct_sample_fraction:np.float64(0.5703125) - srpo/sdpo_sample_fraction:np.float64(0.3046875) - srpo/grpo_sample_fraction:np.float64(0.6953125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.01917144775052293) - self_distillation/dynamic_weight_mean:np.float64(0.30468750558793545) - self_distillation/dynamic_weight_max:np.float64(1.1113983392715454) - self_distillation/empty_target_batch:np.float64(0.140625) - actor/grad_norm:np.float64(0.28858262300491333) - perf/mfu/actor:np.float64(0.024393681244515707) - perf/max_memory_allocated_gb:np.float64(138.08052349090576) - perf/max_memory_reserved_gb:np.float64(142.578125) - perf/cpu_memory_used_gb:np.float64(148.60595703125) - actor/lr:np.float64(5e-06) - training/global_step:66 - training/epoch:1 - critic/score/mean:0.5703125 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.5703125 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.0020481471437960863 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.0020481471437960863 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:123.96875 - response_length/max:529.0 - response_length/min:58.0 - response_length/clip
step:67 - global_seqlen/min:10167 - global_seqlen/max:15899 - global_seqlen/minmax_diff:5732 - global_seqlen/balanced_min:13225 - global_seqlen/balanced_max:13225 - global_seqlen/mean:13225.0 - actor/entropy:0.0952252745628357 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.13839662075042725 - training/rollout_probs_diff_mean:0.002533883787691593 - training/rollout_probs_diff_std:0.009842932224273682 - training/rollout_actor_probs_pearson_corr:0.9980432391166687 - self_distillation/success_group_fraction:0.875 - self_distillation/success_sample_fraction:0.86328125 - self_distillation/correct_sample_fraction:0.62890625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.86328125 - rollout_corr/rollout_is_mean:0.9999334216117859 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.5982176065444946 - rollout_corr/rollout_is_min:0.6212241649627686 - rollout_corr/rollout_is_std:0.027478566393256187 - rollout_corr/rollout_is_eff_sample_size:0.9992452600041167 - rollout_corr/rollout_is_seq_mean:0.9999302625656128 - rollout_corr/rollout_is_seq_std:0.0026975483633577824 - rollout_corr/rollout_is_seq_max:1.0099576711654663 - rollout_corr/rollout_is_seq_min:0.9928929805755615 - rollout_corr/rollout_is_seq_max_deviation:0.009957671165466309 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.1045672693289816) - rollout_corr/training_log_ppl:np.float64(0.09825653926236555) - rollout_corr/kl:np.float64(0.00045608972865096575) - rollout_corr/k3_kl:np.float64(0.0003862984270455172) - rollout_corr/rollout_ppl:np.float64(1.1040719766169786) - rollout_corr/rollout_log_ppl:np.float64(0.09780044853687286) - rollout_corr/log_ppl_diff:np.float64(0.00045609072549268603) - rollout_corr/log_ppl_abs_diff:np.float64(0.0020357573666842654) - rollout_corr/log_ppl_diff_max:np.float64(0.008515119552612305) - rollout_corr/log_ppl_diff_min:np.float64(-0.008621647953987122) - rollout_corr/ppl_ratio:np.float64(1.0004596968647093) - rollout_corr/chi2_token:np.float64(0.0006398719269782305) - rollout_corr/chi2_seq:np.float64(0.10121880285441875) - actor/pg_loss:np.float64(-0.071889858769282) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.07958094446803443) - srpo/sdpo_loss:np.float64(0.007691085698752431) - srpo/routed_loss:np.float64(-0.071889858769282) - srpo/routed_token_count:np.float64(121.4375) - srpo/correct_sample_fraction:np.float64(0.62890625) - srpo/sdpo_sample_fraction:np.float64(0.24609375) - srpo/grpo_sample_fraction:np.float64(0.75390625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.011631388404566678) - self_distillation/dynamic_weight_mean:np.float64(0.24609375069849193) - self_distillation/dynamic_weight_max:np.float64(1.1106873750686646) - self_distillation/empty_target_batch:np.float64(0.13671875) - actor/grad_norm:np.float64(0.27441921830177307) - perf/mfu/actor:np.float64(0.02436524027615205) - perf/max_memory_allocated_gb:np.float64(138.08052349090576) - perf/max_memory_reserved_gb:np.float64(142.578125) - perf/cpu_memory_used_gb:np.float64(148.54054260253906) - actor/lr:np.float64(5e-06) - training/global_step:67 - training/epoch:1 - critic/score/mean:0.62890625 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.62890625 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.006803268101066351 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.006803268101066351 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:121.4375 - response_length/max:309.0 - response_length/min:61.0 - response_length/clip_ratio:0
step:68 - global_seqlen/min:11203 - global_seqlen/max:16353 - global_seqlen/minmax_diff:5150 - global_seqlen/balanced_min:13737 - global_seqlen/balanced_max:13738 - global_seqlen/mean:13737.75 - actor/entropy:0.10888689756393433 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.16473042964935303 - training/rollout_probs_diff_mean:0.0029038761276751757 - training/rollout_probs_diff_std:0.010477473959326744 - training/rollout_actor_probs_pearson_corr:0.9979749321937561 - self_distillation/success_group_fraction:0.90625 - self_distillation/success_sample_fraction:0.8984375 - self_distillation/correct_sample_fraction:0.546875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.8984375 - rollout_corr/rollout_is_mean:0.9999992251396179 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.6370769739151 - rollout_corr/rollout_is_min:0.607120931148529 - rollout_corr/rollout_is_std:0.029325343668460846 - rollout_corr/rollout_is_eff_sample_size:0.9991406441036216 - rollout_corr/rollout_is_seq_mean:0.9999682903289795 - rollout_corr/rollout_is_seq_std:0.0029736359138041735 - rollout_corr/rollout_is_seq_max:1.0099501609802246 - rollout_corr/rollout_is_seq_min:0.9841387867927551 - rollout_corr/rollout_is_seq_max_deviation:0.015861213207244873 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.110229044687003) - rollout_corr/training_log_ppl:np.float64(0.1033436311117839) - rollout_corr/kl:np.float64(0.0004586394768377744) - rollout_corr/k3_kl:np.float64(0.000426953503327443) - rollout_corr/rollout_ppl:np.float64(1.1096893716603518) - rollout_corr/rollout_log_ppl:np.float64(0.10288499109447002) - rollout_corr/log_ppl_diff:np.float64(0.0004586400173138827) - rollout_corr/log_ppl_abs_diff:np.float64(0.002226220298325643) - rollout_corr/log_ppl_diff_max:np.float64(0.019079208374023438) - rollout_corr/log_ppl_diff_min:np.float64(-0.0089883953332901) - rollout_corr/ppl_ratio:np.float64(1.0004633385688066) - rollout_corr/chi2_token:np.float64(0.0007876004092395306) - rollout_corr/chi2_seq:np.float64(0.10754743334837258) - actor/pg_loss:np.float64(-0.12952155137827503) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.1406843404402025) - srpo/sdpo_loss:np.float64(0.01116278906192747) - srpo/routed_loss:np.float64(-0.12952155137827503) - srpo/routed_token_count:np.float64(114.0859375) - srpo/correct_sample_fraction:np.float64(0.546875) - srpo/sdpo_sample_fraction:np.float64(0.359375) - srpo/grpo_sample_fraction:np.float64(0.640625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.02415310819799288) - self_distillation/dynamic_weight_mean:np.float64(0.35937500768341124) - self_distillation/dynamic_weight_max:np.float64(1.1500111818313599) - self_distillation/empty_target_batch:np.float64(0.1015625) - actor/grad_norm:np.float64(0.4039847254753113) - perf/mfu/actor:np.float64(0.025638799493840042) - perf/max_memory_allocated_gb:np.float64(138.08052349090576) - perf/max_memory_reserved_gb:np.float64(142.578125) - perf/cpu_memory_used_gb:np.float64(148.73432540893555) - actor/lr:np.float64(5e-06) - training/global_step:68 - training/epoch:1 - critic/score/mean:0.546875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.546875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.0005093131330795586 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.0005093131330795586 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:114.0859375 - response_length/max:343.0 - response_length/min:56.0 - response_length/clip_ratio:0.0 - response
step:69 - global_seqlen/min:12337 - global_seqlen/max:17212 - global_seqlen/minmax_diff:4875 - global_seqlen/balanced_min:14636 - global_seqlen/balanced_max:14637 - global_seqlen/mean:14636.125 - actor/entropy:0.09894971549510956 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.20145291090011597 - training/rollout_probs_diff_mean:0.0027205839287489653 - training/rollout_probs_diff_std:0.010346534661948681 - training/rollout_actor_probs_pearson_corr:0.9979661703109741 - self_distillation/success_group_fraction:0.84375 - self_distillation/success_sample_fraction:0.83203125 - self_distillation/correct_sample_fraction:0.58203125 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.83203125 - rollout_corr/rollout_is_mean:0.9999226331710815 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.8836244344711304 - rollout_corr/rollout_is_min:0.5814875364303589 - rollout_corr/rollout_is_std:0.02946525439620018 - rollout_corr/rollout_is_eff_sample_size:0.9991324328612136 - rollout_corr/rollout_is_seq_mean:0.9998986721038818 - rollout_corr/rollout_is_seq_std:0.002614550990983844 - rollout_corr/rollout_is_seq_max:1.0072904825210571 - rollout_corr/rollout_is_seq_min:0.9896863102912903 - rollout_corr/rollout_is_seq_max_deviation:0.010313689708709717 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.1083614937961102) - rollout_corr/training_log_ppl:np.float64(0.10132583577433252) - rollout_corr/kl:np.float64(0.0005330158397036655) - rollout_corr/k3_kl:np.float64(0.00043168375866409203) - rollout_corr/rollout_ppl:np.float64(1.1077509811148047) - rollout_corr/rollout_log_ppl:np.float64(0.1007928194339911) - rollout_corr/log_ppl_diff:np.float64(0.000533016340341419) - rollout_corr/log_ppl_abs_diff:np.float64(0.0019907622772734612) - rollout_corr/log_ppl_diff_max:np.float64(0.011344373226165771) - rollout_corr/log_ppl_diff_min:np.float64(-0.006300196051597595) - rollout_corr/ppl_ratio:np.float64(1.0005367156118155) - rollout_corr/chi2_token:np.float64(0.0006620162166655064) - rollout_corr/chi2_seq:np.float64(0.1840856010094285) - actor/pg_loss:np.float64(-0.08912376813532319) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.09717828460270539) - srpo/sdpo_loss:np.float64(0.0080545164673822) - srpo/routed_loss:np.float64(-0.08912376813532319) - srpo/routed_token_count:np.float64(147.56640625) - srpo/correct_sample_fraction:np.float64(0.58203125) - srpo/sdpo_sample_fraction:np.float64(0.26171875) - srpo/grpo_sample_fraction:np.float64(0.73828125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.023064203735884803) - self_distillation/dynamic_weight_mean:np.float64(0.26171875349245965) - self_distillation/dynamic_weight_max:np.float64(1.2205404043197632) - self_distillation/empty_target_batch:np.float64(0.16796875) - actor/grad_norm:np.float64(0.2577652931213379) - perf/mfu/actor:np.float64(0.02716333612402653) - perf/max_memory_allocated_gb:np.float64(138.08052349090576) - perf/max_memory_reserved_gb:np.float64(142.578125) - perf/cpu_memory_used_gb:np.float64(148.62730407714844) - actor/lr:np.float64(5e-06) - training/global_step:69 - training/epoch:1 - critic/score/mean:0.58203125 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.58203125 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.015128253027796745 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.015128253027796745 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:147.56640625 - response_length/max:566.0 - response_length/min:72.0 - response_length/cli
test_gen_batch meta info: {'eos_token_id': 151645, 'pad_token_id': 151643, 'recompute_log_prob': False, 'do_sample': True, 'validate': True, 'global_steps': 70}
validation generation end
[prompt] system

Given a question and four options, please select the right answer. Respond in the following format:
<reasoning>
...
</reasoning>
<answer>
...
</answer>

For the answer, only output the letter corresponding to the correct option (A, B, C, or D), and nothing else. Do not restate the answer text. For example, if the answer is "A", just output:
<answer>
A
</answer>

user
Which of the following molecules is the only correct reactant used in the synthesis of the product "Nc1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1" ?

A: O=C1c2ccccc2C(=O)N1c1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1
B: O=C1c2ccccc2C(Cl)N1c1ccc2ccc(Cl)nc2n1
C: O=C1c2ccccc2C(=O)N1Cc1ccc(Oc2ccc(Cl)c(C(F)(F)F)c2)cc1
D: O=C1c2ccccc2C(=O)N1CC=C(Cc1ccccc1)C(F)(F)F
Please reason step by step.
assistant
<think>

</think>


[response] <reasoning>
The product "Nc1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1" contains a nitrile group (n1), a chlorinated benzene ring (Cc2ccc(Cl)cc2), and a trifluorinated group (C(F)(F)F). The correct reactant must include these functional groups. Option A includes a nitrile group and a chlorinated benzene ring but not the trifluorinated group. Option B includes a nitrile group and a chlorinated benzene ring but not the trifluorinated group. Option C includes a nitrile group, a chlorinated benzene ring, and a trifluorinated group. Option D includes a nitrile group and a chlorinated benzene ring but not the trifluorinated group. Therefore, the correct answer is C.
</reasoning>
<answer>
C
</answer>
[ground_truth] A
[score] 0.0
[acc] 0.0
[pred] C
[incorrect_format] 1
[feedback]
local_global_step_folder: /workspace/L2T/checkpoints/datasets/sciknoweval/chemistry/qwen3gen-chemistry-SRPO_TR-Qwen-Qwen3-4B-mbs32-tr0.1-dwtrue-train32-rollout8-lr5e-6-vllm0.8/global_step_70
step:70 - global_seqlen/min:10908 - global_seqlen/max:16791 - global_seqlen/minmax_diff:5883 - global_seqlen/balanced_min:14062 - global_seqlen/balanced_max:14539 - global_seqlen/mean:14171.875 - actor/entropy:0.10606306046247482 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.4863188862800598 - training/rollout_probs_diff_mean:0.0030448962934315205 - training/rollout_probs_diff_std:0.011526479385793209 - training/rollout_actor_probs_pearson_corr:0.9976092576980591 - self_distillation/success_group_fraction:0.9375 - self_distillation/success_sample_fraction:0.93359375 - self_distillation/correct_sample_fraction:0.765625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.93359375 - rollout_corr/rollout_is_mean:0.999775230884552 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:2.7004401999874972e-05 - rollout_corr/rollout_is_max:1.9516319036483765 - rollout_corr/rollout_is_min:0.47376108169555664 - rollout_corr/rollout_is_std:0.03182549029588699 - rollout_corr/rollout_is_eff_sample_size:0.9989877467337565 - rollout_corr/rollout_is_seq_mean:0.999903678894043 - rollout_corr/rollout_is_seq_std:0.0029711732640862465 - rollout_corr/rollout_is_seq_max:1.0121804475784302 - rollout_corr/rollout_is_seq_min:0.9928668737411499 - rollout_corr/rollout_is_seq_max_deviation:0.012180447578430176 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.116599639877677) - rollout_corr/training_log_ppl:np.float64(0.10911134732305072) - rollout_corr/kl:np.float64(0.0006255446174305845) - rollout_corr/k3_kl:np.float64(0.0005292075004099672) - rollout_corr/rollout_ppl:np.float64(1.1159032019786537) - rollout_corr/rollout_log_ppl:np.float64(0.10848580173478695) - rollout_corr/log_ppl_diff:np.float64(0.0006255455882637762) - rollout_corr/log_ppl_abs_diff:np.float64(0.0023849829958635382) - rollout_corr/log_ppl_diff_max:np.float64(0.008478067815303802) - rollout_corr/log_ppl_diff_min:np.float64(-0.009955182671546936) - rollout_corr/ppl_ratio:np.float64(1.0006301102694124) - rollout_corr/chi2_token:np.float64(0.0008786383550614119) - rollout_corr/chi2_seq:np.float64(0.06351560982875526) - actor/pg_loss:np.float64(-0.0758638327170047) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.0810276246338617) - srpo/sdpo_loss:np.float64(0.005163791916857008) - srpo/routed_loss:np.float64(-0.0758638327170047) - srpo/routed_token_count:np.float64(144.65234375) - srpo/correct_sample_fraction:np.float64(0.765625) - srpo/sdpo_sample_fraction:np.float64(0.171875) - srpo/grpo_sample_fraction:np.float64(0.828125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.011721472545559664) - self_distillation/dynamic_weight_mean:np.float64(0.17187500116415322) - self_distillation/dynamic_weight_max:np.float64(1.1358423233032227) - self_distillation/empty_target_batch:np.float64(0.06640625) - actor/grad_norm:np.float64(0.25328776240348816) - perf/mfu/actor:np.float64(0.02601138158544588) - perf/max_memory_allocated_gb:np.float64(138.08052349090576) - perf/max_memory_reserved_gb:np.float64(142.578125) - perf/cpu_memory_used_gb:np.float64(148.59056091308594) - actor/lr:np.float64(5e-06) - val-aux/sciknoweval/reward/mean@16:np.float64(0.7154761904761905) - val-aux/sciknoweval/reward/std@16:np.float64(0.1622013464536736) - val-aux/sciknoweval/reward/best@2/mean:np.float64(0.7824809523809524) - val-aux/sciknoweval/reward/best@2/std:np.float64(0.1356601996671117) - val-aux/sciknoweval/reward/worst@2/mean:np.float64(0.6490380952380952) - val-aux/sciknoweval/reward/worst@2/std:np.float64(0.1390958995024516) - val-aux/sciknoweval/reward/maj@2/mean:np.float64(0.7149095238095238) - val-aux/sciknowe
step:71 - global_seqlen/min:12689 - global_seqlen/max:21034 - global_seqlen/minmax_diff:8345 - global_seqlen/balanced_min:15056 - global_seqlen/balanced_max:22455 - global_seqlen/mean:15985.375 - actor/entropy:0.07308760285377502 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.5083576440811157 - training/rollout_probs_diff_mean:0.0021449406631290913 - training/rollout_probs_diff_std:0.009686225093901157 - training/rollout_actor_probs_pearson_corr:0.9976727962493896 - self_distillation/success_group_fraction:0.96875 - self_distillation/success_sample_fraction:0.96484375 - self_distillation/correct_sample_fraction:0.6640625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.96484375 - rollout_corr/rollout_is_mean:1.0001015663146973 - rollout_corr/rollout_is_ratio_fraction_high:1.9811788661172614e-05 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:3.2827019691467285 - rollout_corr/rollout_is_min:0.5354881286621094 - rollout_corr/rollout_is_std:0.02704566903412342 - rollout_corr/rollout_is_eff_sample_size:0.999269304524189 - rollout_corr/rollout_is_seq_mean:1.0001920461654663 - rollout_corr/rollout_is_seq_std:0.0024355349596589804 - rollout_corr/rollout_is_seq_max:1.0134079456329346 - rollout_corr/rollout_is_seq_min:0.9937267303466797 - rollout_corr/rollout_is_seq_max_deviation:0.01340794563293457 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.0927376779727638) - rollout_corr/training_log_ppl:np.float64(0.08764094920752541) - rollout_corr/kl:np.float64(0.00024133898584732805) - rollout_corr/k3_kl:np.float64(0.0004333367967565005) - rollout_corr/rollout_ppl:np.float64(1.092457460705191) - rollout_corr/rollout_log_ppl:np.float64(0.08739960952198089) - rollout_corr/log_ppl_diff:np.float64(0.00024133968554451712) - rollout_corr/log_ppl_abs_diff:np.float64(0.00176785381472655) - rollout_corr/log_ppl_diff_max:np.float64(0.007296815514564514) - rollout_corr/log_ppl_diff_min:np.float64(-0.006337381899356842) - rollout_corr/ppl_ratio:np.float64(1.0002440565731376) - rollout_corr/chi2_token:np.float64(0.001331231789663434) - rollout_corr/chi2_seq:np.float64(0.13815684313885868) - actor/pg_loss:np.float64(-0.1219440627169206) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.12995886566932313) - srpo/sdpo_loss:np.float64(0.008014802952402533) - srpo/routed_loss:np.float64(-0.1219440627169206) - srpo/routed_token_count:np.float64(197.16796875) - srpo/correct_sample_fraction:np.float64(0.6640625) - srpo/sdpo_sample_fraction:np.float64(0.3046875) - srpo/grpo_sample_fraction:np.float64(0.6953125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.02064178472534195) - self_distillation/dynamic_weight_mean:np.float64(0.30468750069849193) - self_distillation/dynamic_weight_max:np.float64(1.1582491397857666) - self_distillation/empty_target_batch:np.float64(0.03515625) - actor/grad_norm:np.float64(0.29051506519317627) - perf/mfu/actor:np.float64(0.02748896150032404) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(150.24188995361328) - actor/lr:np.float64(5e-06) - training/global_step:71 - training/epoch:1 - critic/score/mean:0.6640625 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.6640625 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.08392520993947983 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.08392520993947983 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:197.16796875 - response_length/max:8192.0 - response_length/min:82.0 - response_
step:72 - global_seqlen/min:11233 - global_seqlen/max:15759 - global_seqlen/minmax_diff:4526 - global_seqlen/balanced_min:13431 - global_seqlen/balanced_max:13432 - global_seqlen/mean:13431.75 - actor/entropy:0.09807024896144867 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.1510450839996338 - training/rollout_probs_diff_mean:0.002841531066223979 - training/rollout_probs_diff_std:0.010730388574302197 - training/rollout_actor_probs_pearson_corr:0.9977133870124817 - self_distillation/success_group_fraction:0.9375 - self_distillation/success_sample_fraction:0.92578125 - self_distillation/correct_sample_fraction:0.75 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.92578125 - rollout_corr/rollout_is_mean:1.0000724792480469 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.6415709257125854 - rollout_corr/rollout_is_min:0.5509337782859802 - rollout_corr/rollout_is_std:0.030073923990130424 - rollout_corr/rollout_is_eff_sample_size:0.9990963763943072 - rollout_corr/rollout_is_seq_mean:1.0000476837158203 - rollout_corr/rollout_is_seq_std:0.002389567904174328 - rollout_corr/rollout_is_seq_max:1.0075503587722778 - rollout_corr/rollout_is_seq_min:0.9915701150894165 - rollout_corr/rollout_is_seq_max_deviation:0.008429884910583496 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.0989604387432337) - rollout_corr/training_log_ppl:np.float64(0.09297857214914984) - rollout_corr/kl:np.float64(0.0003651222882159466) - rollout_corr/k3_kl:np.float64(0.0004127980487863281) - rollout_corr/rollout_ppl:np.float64(1.0985401715151966) - rollout_corr/rollout_log_ppl:np.float64(0.09261344942751748) - rollout_corr/log_ppl_diff:np.float64(0.0003651227216323605) - rollout_corr/log_ppl_abs_diff:np.float64(0.0018616751058289083) - rollout_corr/log_ppl_diff_max:np.float64(0.008964434266090393) - rollout_corr/log_ppl_diff_min:np.float64(-0.006634891033172607) - rollout_corr/ppl_ratio:np.float64(1.0003680654335767) - rollout_corr/chi2_token:np.float64(0.0009237260092049837) - rollout_corr/chi2_seq:np.float64(0.1522837276570499) - actor/pg_loss:np.float64(-0.06985866550166975) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.0741994038107805) - srpo/sdpo_loss:np.float64(0.004340738309110748) - srpo/routed_loss:np.float64(-0.06985866550166975) - srpo/routed_token_count:np.float64(132.8984375) - srpo/correct_sample_fraction:np.float64(0.75) - srpo/sdpo_sample_fraction:np.float64(0.1875) - srpo/grpo_sample_fraction:np.float64(0.8125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.00975653810064614) - self_distillation/dynamic_weight_mean:np.float64(0.1875000037252903) - self_distillation/dynamic_weight_max:np.float64(1.1360833644866943) - self_distillation/empty_target_batch:np.float64(0.07421875) - actor/grad_norm:np.float64(0.23741291463375092) - perf/mfu/actor:np.float64(0.024835022509184478) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(149.92729568481445) - actor/lr:np.float64(5e-06) - training/global_step:72 - training/epoch:1 - critic/score/mean:0.75 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.75 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.001800305675715208 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.001800305675715208 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:132.8984375 - response_length/max:247.0 - response_length/min:51.0 - response_length/clip_ratio:0.0 - response_length_non_a
step:73 - global_seqlen/min:11977 - global_seqlen/max:15634 - global_seqlen/minmax_diff:3657 - global_seqlen/balanced_min:13503 - global_seqlen/balanced_max:13503 - global_seqlen/mean:13503.0 - actor/entropy:0.09675823897123337 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.2151225209236145 - training/rollout_probs_diff_mean:0.002858596621081233 - training/rollout_probs_diff_std:0.011015517637133598 - training/rollout_actor_probs_pearson_corr:0.9975820183753967 - self_distillation/success_group_fraction:0.875 - self_distillation/success_sample_fraction:0.86328125 - self_distillation/correct_sample_fraction:0.61328125 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.86328125 - rollout_corr/rollout_is_mean:1.0000311136245728 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.6466635465621948 - rollout_corr/rollout_is_min:0.5384849905967712 - rollout_corr/rollout_is_std:0.030300989747047424 - rollout_corr/rollout_is_eff_sample_size:0.9990828112686182 - rollout_corr/rollout_is_seq_mean:0.9999957084655762 - rollout_corr/rollout_is_seq_std:0.002753880573436618 - rollout_corr/rollout_is_seq_max:1.012641191482544 - rollout_corr/rollout_is_seq_min:0.9927515387535095 - rollout_corr/rollout_is_seq_max_deviation:0.012641191482543945 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.0984599189832807) - rollout_corr/training_log_ppl:np.float64(0.09297698400041554) - rollout_corr/kl:np.float64(0.00045679825487887626) - rollout_corr/k3_kl:np.float64(0.0004525876552104041) - rollout_corr/rollout_ppl:np.float64(1.0979560846462846) - rollout_corr/rollout_log_ppl:np.float64(0.09252018518964178) - rollout_corr/log_ppl_diff:np.float64(0.00045679881077376194) - rollout_corr/log_ppl_abs_diff:np.float64(0.0020338109934527893) - rollout_corr/log_ppl_diff_max:np.float64(0.008543252944946289) - rollout_corr/log_ppl_diff_min:np.float64(-0.009834721684455872) - rollout_corr/ppl_ratio:np.float64(1.000460479175672) - rollout_corr/chi2_token:np.float64(0.0009012501686811447) - rollout_corr/chi2_seq:np.float64(0.19242209708318114) - actor/pg_loss:np.float64(-0.08166880559110723) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.09036722671589814) - srpo/sdpo_loss:np.float64(0.008698421124790912) - srpo/routed_loss:np.float64(-0.08166880559110723) - srpo/routed_token_count:np.float64(141.75) - srpo/correct_sample_fraction:np.float64(0.61328125) - srpo/sdpo_sample_fraction:np.float64(0.26171875) - srpo/grpo_sample_fraction:np.float64(0.73828125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.016054354805035587) - self_distillation/dynamic_weight_mean:np.float64(0.26171875349245965) - self_distillation/dynamic_weight_max:np.float64(1.1557217836380005) - self_distillation/empty_target_batch:np.float64(0.13671875) - actor/grad_norm:np.float64(0.32828834652900696) - perf/mfu/actor:np.float64(0.02475179626635672) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(149.48853302001953) - actor/lr:np.float64(5e-06) - training/global_step:73 - training/epoch:1 - critic/score/mean:0.61328125 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.61328125 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.0015811012126505375 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.0015811012126505375 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:141.75 - response_length/max:331.0 - response_length/min:73.0 - response_length/clip_ratio:0.
step:74 - global_seqlen/min:10602 - global_seqlen/max:15569 - global_seqlen/minmax_diff:4967 - global_seqlen/balanced_min:13007 - global_seqlen/balanced_max:13008 - global_seqlen/mean:13007.125 - actor/entropy:0.0866396427154541 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.1687774658203125 - training/rollout_probs_diff_mean:0.002447765553370118 - training/rollout_probs_diff_std:0.009990706108510494 - training/rollout_actor_probs_pearson_corr:0.9977685809135437 - self_distillation/success_group_fraction:0.90625 - self_distillation/success_sample_fraction:0.90234375 - self_distillation/correct_sample_fraction:0.71875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.90234375 - rollout_corr/rollout_is_mean:0.9998632073402405 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:5.8711288147605956e-05 - rollout_corr/rollout_is_max:1.7075881958007812 - rollout_corr/rollout_is_min:0.40211260318756104 - rollout_corr/rollout_is_std:0.028100576251745224 - rollout_corr/rollout_is_eff_sample_size:0.999210623645437 - rollout_corr/rollout_is_seq_mean:0.9999074339866638 - rollout_corr/rollout_is_seq_std:0.0023523475974798203 - rollout_corr/rollout_is_seq_max:1.0083256959915161 - rollout_corr/rollout_is_seq_min:0.9929505586624146 - rollout_corr/rollout_is_seq_max_deviation:0.008325695991516113 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.086568511556834) - rollout_corr/training_log_ppl:np.float64(0.08173826814891072) - rollout_corr/kl:np.float64(0.00046768059648627514) - rollout_corr/k3_kl:np.float64(0.0003751251478586459) - rollout_corr/rollout_ppl:np.float64(1.0860533793456852) - rollout_corr/rollout_log_ppl:np.float64(0.08127058656464214) - rollout_corr/log_ppl_diff:np.float64(0.00046768158426857553) - rollout_corr/log_ppl_abs_diff:np.float64(0.0017630548427405301) - rollout_corr/log_ppl_diff_max:np.float64(0.007911652326583862) - rollout_corr/log_ppl_diff_min:np.float64(-0.007574200630187988) - rollout_corr/ppl_ratio:np.float64(1.0004705912433565) - rollout_corr/chi2_token:np.float64(0.0005672622937709093) - rollout_corr/chi2_seq:np.float64(0.06977778836153448) - actor/pg_loss:np.float64(-0.06654009117482929) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.07129649812122807) - srpo/sdpo_loss:np.float64(0.0047564069463987835) - srpo/routed_loss:np.float64(-0.06654009117482929) - srpo/routed_token_count:np.float64(133.06640625) - srpo/correct_sample_fraction:np.float64(0.71875) - srpo/sdpo_sample_fraction:np.float64(0.1875) - srpo/grpo_sample_fraction:np.float64(0.8125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.008584212572941397) - self_distillation/dynamic_weight_mean:np.float64(0.18750000279396772) - self_distillation/dynamic_weight_max:np.float64(1.124984860420227) - self_distillation/empty_target_batch:np.float64(0.09765625) - actor/grad_norm:np.float64(0.26836180686950684) - perf/mfu/actor:np.float64(0.023962406010035173) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(148.9883270263672) - actor/lr:np.float64(5e-06) - training/global_step:74 - training/epoch:1 - critic/score/mean:0.71875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.71875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.0012402759166434407 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.0012402759166434407 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:133.06640625 - response_length/max:391.0 - response_length/min:71.0 - response_leng
step:75 - global_seqlen/min:11650 - global_seqlen/max:15937 - global_seqlen/minmax_diff:4287 - global_seqlen/balanced_min:14277 - global_seqlen/balanced_max:14279 - global_seqlen/mean:14278.0 - actor/entropy:0.08708544075489044 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.2817443907260895 - training/rollout_probs_diff_mean:0.002716779476031661 - training/rollout_probs_diff_std:0.011301529593765736 - training/rollout_actor_probs_pearson_corr:0.9972029328346252 - self_distillation/success_group_fraction:0.96875 - self_distillation/success_sample_fraction:0.95703125 - self_distillation/correct_sample_fraction:0.80078125 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.95703125 - rollout_corr/rollout_is_mean:1.0002191066741943 - rollout_corr/rollout_is_ratio_fraction_high:2.942561150121037e-05 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:2.4571142196655273 - rollout_corr/rollout_is_min:0.6166058778762817 - rollout_corr/rollout_is_std:0.030218258500099182 - rollout_corr/rollout_is_eff_sample_size:0.9990880468875148 - rollout_corr/rollout_is_seq_mean:1.0001577138900757 - rollout_corr/rollout_is_seq_std:0.002703398000448942 - rollout_corr/rollout_is_seq_max:1.0081384181976318 - rollout_corr/rollout_is_seq_min:0.9878208637237549 - rollout_corr/rollout_is_seq_max_deviation:0.012179136276245117 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.0887567941099405) - rollout_corr/training_log_ppl:np.float64(0.0837563306431548) - rollout_corr/kl:np.float64(0.0002619972401661924) - rollout_corr/k3_kl:np.float64(0.00041975817622487074) - rollout_corr/rollout_ppl:np.float64(1.088481810875237) - rollout_corr/rollout_log_ppl:np.float64(0.08349433254625183) - rollout_corr/log_ppl_diff:np.float64(0.00026199809690297116) - rollout_corr/log_ppl_abs_diff:np.float64(0.0019779989543167176) - rollout_corr/log_ppl_diff_max:np.float64(0.013522475957870483) - rollout_corr/log_ppl_diff_min:np.float64(-0.00743432343006134) - rollout_corr/ppl_ratio:np.float64(1.0002656185533851) - rollout_corr/chi2_token:np.float64(0.001179354265332222) - rollout_corr/chi2_seq:np.float64(0.18645216152071953) - actor/pg_loss:np.float64(-0.05082129563948001) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.05323973932536319) - srpo/sdpo_loss:np.float64(0.002418443685883176) - srpo/routed_loss:np.float64(-0.05082129563948001) - srpo/routed_token_count:np.float64(132.75) - srpo/correct_sample_fraction:np.float64(0.80078125) - srpo/sdpo_sample_fraction:np.float64(0.16796875) - srpo/grpo_sample_fraction:np.float64(0.83203125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.00474218082410971) - self_distillation/dynamic_weight_mean:np.float64(0.16796875069849193) - self_distillation/dynamic_weight_max:np.float64(1.111772894859314) - self_distillation/empty_target_batch:np.float64(0.04296875) - actor/grad_norm:np.float64(0.2969547212123871) - perf/mfu/actor:np.float64(0.026491979640340054) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(148.88829803466797) - actor/lr:np.float64(5e-06) - training/global_step:75 - training/epoch:1 - critic/score/mean:0.80078125 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.80078125 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.003939353860914707 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.003939353860914707 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:132.75 - response_length/max:307.0 - response_length/min:61.0 - response_length
step:76 - global_seqlen/min:12833 - global_seqlen/max:20184 - global_seqlen/minmax_diff:7351 - global_seqlen/balanced_min:13532 - global_seqlen/balanced_max:21146 - global_seqlen/mean:14483.75 - actor/entropy:0.07019611448049545 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.34798648953437805 - training/rollout_probs_diff_mean:0.002141572767868638 - training/rollout_probs_diff_std:0.009791345335543156 - training/rollout_actor_probs_pearson_corr:0.9975031018257141 - self_distillation/success_group_fraction:0.84375 - self_distillation/success_sample_fraction:0.83984375 - self_distillation/correct_sample_fraction:0.66015625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.83984375 - rollout_corr/rollout_is_mean:1.0000206232070923 - rollout_corr/rollout_is_ratio_fraction_high:2.3400571080856025e-05 - rollout_corr/rollout_is_ratio_fraction_low:2.3400571080856025e-05 - rollout_corr/rollout_is_max:2.3260293006896973 - rollout_corr/rollout_is_min:0.3429262936115265 - rollout_corr/rollout_is_std:0.02613787353038788 - rollout_corr/rollout_is_eff_sample_size:0.9993170398962825 - rollout_corr/rollout_is_seq_mean:1.0000452995300293 - rollout_corr/rollout_is_seq_std:0.0028731445781886578 - rollout_corr/rollout_is_seq_max:1.013321042060852 - rollout_corr/rollout_is_seq_min:0.9886126518249512 - rollout_corr/rollout_is_seq_max_deviation:0.01332104206085205 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.0861750459298491) - rollout_corr/training_log_ppl:np.float64(0.08166990291829279) - rollout_corr/kl:np.float64(0.000417589516704453) - rollout_corr/k3_kl:np.float64(0.000462848503586466) - rollout_corr/rollout_ppl:np.float64(1.0856995154172182) - rollout_corr/rollout_log_ppl:np.float64(0.08125231268422795) - rollout_corr/log_ppl_diff:np.float64(0.0004175902340648463) - rollout_corr/log_ppl_abs_diff:np.float64(0.0020487928832153557) - rollout_corr/log_ppl_diff_max:np.float64(0.01700911670923233) - rollout_corr/log_ppl_diff_min:np.float64(-0.01198972761631012) - rollout_corr/ppl_ratio:np.float64(1.0004223354626447) - rollout_corr/chi2_token:np.float64(0.0010073489975184202) - rollout_corr/chi2_seq:np.float64(0.10592406312935054) - actor/pg_loss:np.float64(-0.06713852132270404) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.072730683372356) - srpo/sdpo_loss:np.float64(0.005592162049651961) - srpo/routed_loss:np.float64(-0.06713852132270404) - srpo/routed_token_count:np.float64(166.9296875) - srpo/correct_sample_fraction:np.float64(0.66015625) - srpo/sdpo_sample_fraction:np.float64(0.18359375) - srpo/grpo_sample_fraction:np.float64(0.81640625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.012420885851497587) - self_distillation/dynamic_weight_mean:np.float64(0.18359374930150807) - self_distillation/dynamic_weight_max:np.float64(1.158028244972229) - self_distillation/empty_target_batch:np.float64(0.16015625) - actor/grad_norm:np.float64(0.24376951158046722) - perf/mfu/actor:np.float64(0.025213702327357507) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(148.96228408813477) - actor/lr:np.float64(5e-06) - training/global_step:76 - training/epoch:1 - critic/score/mean:0.66015625 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.66015625 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.0035393363796174526 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.0035393363796174526 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:166.9296875 - response_length/max:8192.0 - response_le
step:77 - global_seqlen/min:12116 - global_seqlen/max:16057 - global_seqlen/minmax_diff:3941 - global_seqlen/balanced_min:13742 - global_seqlen/balanced_max:13743 - global_seqlen/mean:13742.75 - actor/entropy:0.07516742497682571 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.20888036489486694 - training/rollout_probs_diff_mean:0.002255114493891597 - training/rollout_probs_diff_std:0.009714490734040737 - training/rollout_actor_probs_pearson_corr:0.997674822807312 - self_distillation/success_group_fraction:0.90625 - self_distillation/success_sample_fraction:0.90234375 - self_distillation/correct_sample_fraction:0.703125 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.90234375 - rollout_corr/rollout_is_mean:1.0000028610229492 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.9710240364074707 - rollout_corr/rollout_is_min:0.5474115610122681 - rollout_corr/rollout_is_std:0.02890361100435257 - rollout_corr/rollout_is_eff_sample_size:0.999165278640662 - rollout_corr/rollout_is_seq_mean:1.0000054836273193 - rollout_corr/rollout_is_seq_std:0.0026133586652576923 - rollout_corr/rollout_is_seq_max:1.0114777088165283 - rollout_corr/rollout_is_seq_min:0.9882959127426147 - rollout_corr/rollout_is_seq_max_deviation:0.011704087257385254 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.074499599635601) - rollout_corr/training_log_ppl:np.float64(0.07062509996649169) - rollout_corr/kl:np.float64(0.00037311074069540373) - rollout_corr/k3_kl:np.float64(0.00037850546428192054) - rollout_corr/rollout_ppl:np.float64(1.074076794553548) - rollout_corr/rollout_log_ppl:np.float64(0.07025198871087923) - rollout_corr/log_ppl_diff:np.float64(0.00037311125561245717) - rollout_corr/log_ppl_abs_diff:np.float64(0.0017588526643521618) - rollout_corr/log_ppl_diff_max:np.float64(0.013298943638801575) - rollout_corr/log_ppl_diff_min:np.float64(-0.009798392653465271) - rollout_corr/ppl_ratio:np.float64(1.0003766296431422) - rollout_corr/chi2_token:np.float64(0.0007797726429998875) - rollout_corr/chi2_seq:np.float64(0.07150697847828269) - actor/pg_loss:np.float64(-0.06556268776535035) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.06836187961744145) - srpo/sdpo_loss:np.float64(0.0027991918520910986) - srpo/routed_loss:np.float64(-0.06556268776535035) - srpo/routed_token_count:np.float64(112.3671875) - srpo/correct_sample_fraction:np.float64(0.703125) - srpo/sdpo_sample_fraction:np.float64(0.203125) - srpo/grpo_sample_fraction:np.float64(0.796875) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.0045019245814614806) - self_distillation/dynamic_weight_mean:np.float64(0.20312500279396772) - self_distillation/dynamic_weight_max:np.float64(1.1098324060440063) - self_distillation/empty_target_batch:np.float64(0.09765625) - actor/grad_norm:np.float64(0.20624420046806335) - perf/mfu/actor:np.float64(0.02541080075560593) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(148.92205810546875) - actor/lr:np.float64(5e-06) - training/global_step:77 - training/epoch:1 - critic/score/mean:0.703125 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.703125 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.0015035110991448164 - critic/advantages/max:0.875 - critic/advantages/min:-0.75 - critic/returns/mean:0.0015035110991448164 - critic/returns/max:0.875 - critic/returns/min:-0.75 - response_length/mean:112.3671875 - response_length/max:251.0 - response_length/min:50.0 - response_length/clip_ratio:0.0
step:78 - global_seqlen/min:11170 - global_seqlen/max:14294 - global_seqlen/minmax_diff:3124 - global_seqlen/balanced_min:13225 - global_seqlen/balanced_max:13268 - global_seqlen/mean:13230.875 - actor/entropy:0.08128337562084198 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.16414302587509155 - training/rollout_probs_diff_mean:0.0022823296021670103 - training/rollout_probs_diff_std:0.009731779806315899 - training/rollout_actor_probs_pearson_corr:0.9978265762329102 - self_distillation/success_group_fraction:0.96875 - self_distillation/success_sample_fraction:0.9609375 - self_distillation/correct_sample_fraction:0.73828125 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.9609375 - rollout_corr/rollout_is_mean:0.9998404383659363 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:5.834475814481266e-05 - rollout_corr/rollout_is_max:1.5814416408538818 - rollout_corr/rollout_is_min:0.35136324167251587 - rollout_corr/rollout_is_std:0.027001556009054184 - rollout_corr/rollout_is_eff_sample_size:0.9992714471613361 - rollout_corr/rollout_is_seq_mean:0.9998856782913208 - rollout_corr/rollout_is_seq_std:0.0026968109887093306 - rollout_corr/rollout_is_seq_max:1.0091408491134644 - rollout_corr/rollout_is_seq_min:0.9904381036758423 - rollout_corr/rollout_is_seq_max_deviation:0.009561896324157715 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.0809407522901893) - rollout_corr/training_log_ppl:np.float64(0.07696737742116966) - rollout_corr/kl:np.float64(0.000483757337320867) - rollout_corr/k3_kl:np.float64(0.00036933288344465076) - rollout_corr/rollout_ppl:np.float64(1.0803994224406779) - rollout_corr/rollout_log_ppl:np.float64(0.0764836190319329) - rollout_corr/log_ppl_diff:np.float64(0.00048375838923675474) - rollout_corr/log_ppl_abs_diff:np.float64(0.0020478447349887574) - rollout_corr/log_ppl_diff_max:np.float64(0.013117805123329163) - rollout_corr/log_ppl_diff_min:np.float64(-0.008270904421806335) - rollout_corr/ppl_ratio:np.float64(1.0004878230392933) - rollout_corr/chi2_token:np.float64(0.000485497759655118) - rollout_corr/chi2_seq:np.float64(0.08525530016049743) - actor/pg_loss:np.float64(-0.07585170669335639) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.08154949767049402) - srpo/sdpo_loss:np.float64(0.005697790977137629) - srpo/routed_loss:np.float64(-0.07585170669335639) - srpo/routed_token_count:np.float64(133.90234375) - srpo/correct_sample_fraction:np.float64(0.73828125) - srpo/sdpo_sample_fraction:np.float64(0.23046875) - srpo/grpo_sample_fraction:np.float64(0.76953125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.011170415348033202) - self_distillation/dynamic_weight_mean:np.float64(0.23046875279396772) - self_distillation/dynamic_weight_max:np.float64(1.1034929752349854) - self_distillation/empty_target_batch:np.float64(0.0390625) - actor/grad_norm:np.float64(0.26540350914001465) - perf/mfu/actor:np.float64(0.02449378023378878) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(148.9018669128418) - actor/lr:np.float64(5e-06) - training/global_step:78 - training/epoch:1 - critic/score/mean:0.73828125 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.73828125 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.003318358212709427 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.003318358212709427 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:133.90234375 - response_length/max:565.0 - response_length/min:50.0
step:79 - global_seqlen/min:10004 - global_seqlen/max:15111 - global_seqlen/minmax_diff:5107 - global_seqlen/balanced_min:12233 - global_seqlen/balanced_max:12234 - global_seqlen/mean:12233.25 - actor/entropy:0.0764228031039238 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.1796872615814209 - training/rollout_probs_diff_mean:0.0022655539214611053 - training/rollout_probs_diff_std:0.009849187918007374 - training/rollout_actor_probs_pearson_corr:0.9975538849830627 - self_distillation/success_group_fraction:0.875 - self_distillation/success_sample_fraction:0.87109375 - self_distillation/correct_sample_fraction:0.66015625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.87109375 - rollout_corr/rollout_is_mean:0.9999358057975769 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:6.597611354663968e-05 - rollout_corr/rollout_is_max:1.5868538618087769 - rollout_corr/rollout_is_min:0.37340596318244934 - rollout_corr/rollout_is_std:0.026884306222200394 - rollout_corr/rollout_is_eff_sample_size:0.9992776370535978 - rollout_corr/rollout_is_seq_mean:0.9999191164970398 - rollout_corr/rollout_is_seq_std:0.002227405784651637 - rollout_corr/rollout_is_seq_max:1.0106374025344849 - rollout_corr/rollout_is_seq_min:0.9920806884765625 - rollout_corr/rollout_is_seq_max_deviation:0.010637402534484863 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.0727560594677925) - rollout_corr/training_log_ppl:np.float64(0.0687835197950335) - rollout_corr/kl:np.float64(0.000438673020267899) - rollout_corr/k3_kl:np.float64(0.00035780576829951005) - rollout_corr/rollout_ppl:np.float64(1.0722568081691861) - rollout_corr/rollout_log_ppl:np.float64(0.06834484680439346) - rollout_corr/log_ppl_diff:np.float64(0.0004386729906400433) - rollout_corr/log_ppl_abs_diff:np.float64(0.0016436664354841923) - rollout_corr/log_ppl_diff_max:np.float64(0.013000249862670898) - rollout_corr/log_ppl_diff_min:np.float64(-0.008548349142074585) - rollout_corr/ppl_ratio:np.float64(1.0004414792638272) - rollout_corr/chi2_token:np.float64(0.0005525860469788313) - rollout_corr/chi2_seq:np.float64(0.04466355126351118) - actor/pg_loss:np.float64(-0.0944358478382128) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.10016463336069137) - srpo/sdpo_loss:np.float64(0.005728785522478574) - srpo/routed_loss:np.float64(-0.0944358478382128) - srpo/routed_token_count:np.float64(118.4140625) - srpo/correct_sample_fraction:np.float64(0.66015625) - srpo/sdpo_sample_fraction:np.float64(0.21484375) - srpo/grpo_sample_fraction:np.float64(0.78515625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.010715416441144043) - self_distillation/dynamic_weight_mean:np.float64(0.21484375512227416) - self_distillation/dynamic_weight_max:np.float64(1.185659408569336) - self_distillation/empty_target_batch:np.float64(0.12890625) - actor/grad_norm:np.float64(0.23672513663768768) - perf/mfu/actor:np.float64(0.022593837692275857) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(148.87750244140625) - actor/lr:np.float64(5e-06) - training/global_step:79 - training/epoch:1 - critic/score/mean:0.66015625 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.66015625 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.005954344756901264 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.005954344756901264 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:118.4140625 - response_length/max:374.0 - response_length/min:53.0 - respon
test_gen_batch meta info: {'eos_token_id': 151645, 'pad_token_id': 151643, 'recompute_log_prob': False, 'do_sample': True, 'validate': True, 'global_steps': 80}
validation generation end
[prompt] system

Given a question and four options, please select the right answer. Respond in the following format:
<reasoning>
...
</reasoning>
<answer>
...
</answer>

For the answer, only output the letter corresponding to the correct option (A, B, C, or D), and nothing else. Do not restate the answer text. For example, if the answer is "A", just output:
<answer>
A
</answer>

user
Which of the following molecules is the only correct reactant used in the synthesis of the product "Nc1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1" ?

A: O=C1c2ccccc2C(=O)N1c1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1
B: O=C1c2ccccc2C(Cl)N1c1ccc2ccc(Cl)nc2n1
C: O=C1c2ccccc2C(=O)N1Cc1ccc(Oc2ccc(Cl)c(C(F)(F)F)c2)cc1
D: O=C1c2ccccc2C(=O)N1CC=C(Cc1ccccc1)C(F)(F)F
Please reason step by step.
assistant
<think>

</think>


[response] <reasoning>
The product "Nc1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1" contains a nitrogen-containing group, a chlorine-containing group, and a trifluorinated group. The correct reactant must include these functional groups. Option A includes a nitrogen-containing group, a chlorine-containing group, and a trifluorinated group, making it the correct answer. Therefore, the correct answer is A.
</reasoning>
<answer>
A
</answer>
[ground_truth] A
[score] 1.0
[acc] 1.0
[pred] A
[incorrect_format] 1
[feedback]
local_global_step_folder: /workspace/L2T/checkpoints/datasets/sciknoweval/chemistry/qwen3gen-chemistry-SRPO_TR-Qwen-Qwen3-4B-mbs32-tr0.1-dwtrue-train32-rollout8-lr5e-6-vllm0.8/global_step_80
step:80 - global_seqlen/min:10902 - global_seqlen/max:17302 - global_seqlen/minmax_diff:6400 - global_seqlen/balanced_min:13346 - global_seqlen/balanced_max:13346 - global_seqlen/mean:13346.0 - actor/entropy:0.0792277455329895 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.1791682243347168 - training/rollout_probs_diff_mean:0.0024412365164607763 - training/rollout_probs_diff_std:0.010290739126503468 - training/rollout_actor_probs_pearson_corr:0.9975858330726624 - self_distillation/success_group_fraction:0.90625 - self_distillation/success_sample_fraction:0.90625 - self_distillation/correct_sample_fraction:0.72265625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.90625 - rollout_corr/rollout_is_mean:0.9998809099197388 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:3.7787183828186244e-05 - rollout_corr/rollout_is_max:1.7928087711334229 - rollout_corr/rollout_is_min:0.43323153257369995 - rollout_corr/rollout_is_std:0.028045371174812317 - rollout_corr/rollout_is_eff_sample_size:0.9992140752710535 - rollout_corr/rollout_is_seq_mean:0.9998906254768372 - rollout_corr/rollout_is_seq_std:0.002873089164495468 - rollout_corr/rollout_is_seq_max:1.0095385313034058 - rollout_corr/rollout_is_seq_min:0.9849433302879333 - rollout_corr/rollout_is_seq_max_deviation:0.01505666971206665 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.0822319672442973) - rollout_corr/training_log_ppl:np.float64(0.07772880301672558) - rollout_corr/kl:np.float64(0.0004939299538424535) - rollout_corr/k3_kl:np.float64(0.0003846082520695959) - rollout_corr/rollout_ppl:np.float64(1.0816740836016834) - rollout_corr/rollout_log_ppl:np.float64(0.07723487233488413) - rollout_corr/log_ppl_diff:np.float64(0.000493930681841448) - rollout_corr/log_ppl_abs_diff:np.float64(0.002141821925761178) - rollout_corr/log_ppl_diff_max:np.float64(0.018234044313430786) - rollout_corr/log_ppl_diff_min:np.float64(-0.008881479501724243) - rollout_corr/ppl_ratio:np.float64(1.0004986070562154) - rollout_corr/chi2_token:np.float64(0.0005350830033421516) - rollout_corr/chi2_seq:np.float64(0.05080895754508674) - actor/pg_loss:np.float64(-0.07214670345638297) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.07764237260562368) - srpo/sdpo_loss:np.float64(0.005495669149240712) - srpo/routed_loss:np.float64(-0.07214670345638297) - srpo/routed_token_count:np.float64(103.375) - srpo/correct_sample_fraction:np.float64(0.72265625) - srpo/sdpo_sample_fraction:np.float64(0.18359375) - srpo/grpo_sample_fraction:np.float64(0.81640625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.009288684247394485) - self_distillation/dynamic_weight_mean:np.float64(0.18359375093132257) - self_distillation/dynamic_weight_max:np.float64(1.1205753087997437) - self_distillation/empty_target_batch:np.float64(0.09375) - actor/grad_norm:np.float64(0.23049889504909515) - perf/mfu/actor:np.float64(0.02460045352716278) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(148.94375610351562) - actor/lr:np.float64(5e-06) - val-aux/sciknoweval/reward/mean@16:np.float64(0.7297619047619047) - val-aux/sciknoweval/reward/std@16:np.float64(0.16846229992150755) - val-aux/sciknoweval/reward/best@2/mean:np.float64(0.7984380952380952) - val-aux/sciknoweval/reward/best@2/std:np.float64(0.14399044642631648) - val-aux/sciknoweval/reward/worst@2/mean:np.float64(0.6603095238095239) - val-aux/sciknoweval/reward/worst@2/std:np.float64(0.14126866800550886) - val-aux/sciknoweval/reward/maj@2/mean:np.float64(0.7289190476190475) - val-aux/sciknoweval/
step:81 - global_seqlen/min:9813 - global_seqlen/max:16246 - global_seqlen/minmax_diff:6433 - global_seqlen/balanced_min:13161 - global_seqlen/balanced_max:13162 - global_seqlen/mean:13161.875 - actor/entropy:0.07348481565713882 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.347068190574646 - training/rollout_probs_diff_mean:0.002137346426025033 - training/rollout_probs_diff_std:0.009722309187054634 - training/rollout_actor_probs_pearson_corr:0.9977238774299622 - self_distillation/success_group_fraction:0.875 - self_distillation/success_sample_fraction:0.859375 - self_distillation/correct_sample_fraction:0.57421875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.859375 - rollout_corr/rollout_is_mean:0.9999052286148071 - rollout_corr/rollout_is_ratio_fraction_high:3.772019044845365e-05 - rollout_corr/rollout_is_ratio_fraction_low:3.772019044845365e-05 - rollout_corr/rollout_is_max:2.9073822498321533 - rollout_corr/rollout_is_min:0.2241167426109314 - rollout_corr/rollout_is_std:0.02701038494706154 - rollout_corr/rollout_is_eff_sample_size:0.9992706139124649 - rollout_corr/rollout_is_seq_mean:0.9998422861099243 - rollout_corr/rollout_is_seq_std:0.0026016186457127333 - rollout_corr/rollout_is_seq_max:1.0158658027648926 - rollout_corr/rollout_is_seq_min:0.9922592043876648 - rollout_corr/rollout_is_seq_max_deviation:0.015865802764892578 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.0765355075709522) - rollout_corr/training_log_ppl:np.float64(0.07240884040584206) - rollout_corr/kl:np.float64(0.0005156485322121984) - rollout_corr/k3_kl:np.float64(0.00035794944963194776) - rollout_corr/rollout_ppl:np.float64(1.075964877847582) - rollout_corr/rollout_log_ppl:np.float64(0.07189319134340622) - rollout_corr/log_ppl_diff:np.float64(0.0005156490624358412) - rollout_corr/log_ppl_abs_diff:np.float64(0.0018730323172349017) - rollout_corr/log_ppl_diff_max:np.float64(0.009854093194007874) - rollout_corr/log_ppl_diff_min:np.float64(-0.008874692022800446) - rollout_corr/ppl_ratio:np.float64(1.0005191133823246) - rollout_corr/chi2_token:np.float64(0.0004306891933083534) - rollout_corr/chi2_seq:np.float64(0.06263559544458985) - actor/pg_loss:np.float64(-0.09007504786177378) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.09518340177601203) - srpo/sdpo_loss:np.float64(0.005108353914238251) - srpo/routed_loss:np.float64(-0.09007504786177378) - srpo/routed_token_count:np.float64(103.55859375) - srpo/correct_sample_fraction:np.float64(0.57421875) - srpo/sdpo_sample_fraction:np.float64(0.30078125) - srpo/grpo_sample_fraction:np.float64(0.69921875) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.008746350949337511) - self_distillation/dynamic_weight_mean:np.float64(0.30078125069849193) - self_distillation/dynamic_weight_max:np.float64(1.1100587844848633) - self_distillation/empty_target_batch:np.float64(0.140625) - actor/grad_norm:np.float64(0.25355294346809387) - perf/mfu/actor:np.float64(0.02393713844865898) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(150.25606155395508) - actor/lr:np.float64(5e-06) - training/global_step:81 - training/epoch:1 - critic/score/mean:0.57421875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.57421875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.0028950246050953865 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.0028950246050953865 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:103.55859375 - response_length/max:258.0 - response_lengt
step:82 - global_seqlen/min:10278 - global_seqlen/max:13365 - global_seqlen/minmax_diff:3087 - global_seqlen/balanced_min:11764 - global_seqlen/balanced_max:11764 - global_seqlen/mean:11764.0 - actor/entropy:0.08316215872764587 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.17870813608169556 - training/rollout_probs_diff_mean:0.002568744355812669 - training/rollout_probs_diff_std:0.010389648377895355 - training/rollout_actor_probs_pearson_corr:0.9975401759147644 - self_distillation/success_group_fraction:0.9375 - self_distillation/success_sample_fraction:0.92578125 - self_distillation/correct_sample_fraction:0.63671875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.92578125 - rollout_corr/rollout_is_mean:0.9999285340309143 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.624503254890442 - rollout_corr/rollout_is_min:0.6282551288604736 - rollout_corr/rollout_is_std:0.02856549620628357 - rollout_corr/rollout_is_eff_sample_size:0.9991844396759914 - rollout_corr/rollout_is_seq_mean:0.9999452829360962 - rollout_corr/rollout_is_seq_std:0.003069821512326598 - rollout_corr/rollout_is_seq_max:1.0134748220443726 - rollout_corr/rollout_is_seq_min:0.9887793660163879 - rollout_corr/rollout_is_seq_max_deviation:0.013474822044372559 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.0843495279550552) - rollout_corr/training_log_ppl:np.float64(0.08000823870315799) - rollout_corr/kl:np.float64(0.00045289268040349384) - rollout_corr/k3_kl:np.float64(0.0003981553314727983) - rollout_corr/rollout_ppl:np.float64(1.0838531539775431) - rollout_corr/rollout_log_ppl:np.float64(0.079555345542758) - rollout_corr/log_ppl_diff:np.float64(0.00045289316039998084) - rollout_corr/log_ppl_abs_diff:np.float64(0.0023346303059952334) - rollout_corr/log_ppl_diff_max:np.float64(0.012745559215545654) - rollout_corr/log_ppl_diff_min:np.float64(-0.011450842022895813) - rollout_corr/ppl_ratio:np.float64(1.0004577136132866) - rollout_corr/chi2_token:np.float64(0.0006887775380164385) - rollout_corr/chi2_seq:np.float64(0.07982906117103994) - actor/pg_loss:np.float64(-0.09984756185622246) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.10689236177131534) - srpo/sdpo_loss:np.float64(0.007044799915092881) - srpo/routed_loss:np.float64(-0.09984756185622246) - srpo/routed_token_count:np.float64(101.5625) - srpo/correct_sample_fraction:np.float64(0.63671875) - srpo/sdpo_sample_fraction:np.float64(0.30078125) - srpo/grpo_sample_fraction:np.float64(0.69921875) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.014435893238783137) - self_distillation/dynamic_weight_mean:np.float64(0.3007812516298145) - self_distillation/dynamic_weight_max:np.float64(1.1144254207611084) - self_distillation/empty_target_batch:np.float64(0.07421875) - actor/grad_norm:np.float64(0.4343135952949524) - perf/mfu/actor:np.float64(0.02182563686529571) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(150.03274154663086) - actor/lr:np.float64(5e-06) - training/global_step:82 - training/epoch:1 - critic/score/mean:0.63671875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.63671875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.007168269250541925 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.007168269250541925 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:101.5625 - response_length/max:192.0 - response_length/min:56.0 - response_length/clip_ratio:0.
step:83 - global_seqlen/min:10415 - global_seqlen/max:14436 - global_seqlen/minmax_diff:4021 - global_seqlen/balanced_min:12308 - global_seqlen/balanced_max:12309 - global_seqlen/mean:12308.625 - actor/entropy:0.07909444719552994 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.18865349888801575 - training/rollout_probs_diff_mean:0.0023771459236741066 - training/rollout_probs_diff_std:0.009836291894316673 - training/rollout_actor_probs_pearson_corr:0.9976748824119568 - self_distillation/success_group_fraction:0.84375 - self_distillation/success_sample_fraction:0.83203125 - self_distillation/correct_sample_fraction:0.6171875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.83203125 - rollout_corr/rollout_is_mean:1.0002893209457397 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.6759071350097656 - rollout_corr/rollout_is_min:0.6251932978630066 - rollout_corr/rollout_is_std:0.0264979749917984 - rollout_corr/rollout_is_eff_sample_size:0.9992988261114965 - rollout_corr/rollout_is_seq_mean:1.0003459453582764 - rollout_corr/rollout_is_seq_std:0.003097602864727378 - rollout_corr/rollout_is_seq_max:1.012476921081543 - rollout_corr/rollout_is_seq_min:0.9905458092689514 - rollout_corr/rollout_is_seq_max_deviation:0.012476921081542969 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.0762789640575647) - rollout_corr/training_log_ppl:np.float64(0.07206299771769409) - rollout_corr/kl:np.float64(-1.567936966750949e-05) - rollout_corr/k3_kl:np.float64(0.0003303609572927746) - rollout_corr/rollout_ppl:np.float64(1.0762940663844347) - rollout_corr/rollout_log_ppl:np.float64(0.07207867698070913) - rollout_corr/log_ppl_diff:np.float64(-1.5679263015044853e-05) - rollout_corr/log_ppl_abs_diff:np.float64(0.0022263024548010435) - rollout_corr/log_ppl_diff_max:np.float64(0.01034681499004364) - rollout_corr/log_ppl_diff_min:np.float64(-0.011252529919147491) - rollout_corr/ppl_ratio:np.float64(0.9999889875762165) - rollout_corr/chi2_token:np.float64(0.0013725715689361095) - rollout_corr/chi2_seq:np.float64(0.12682225415483117) - actor/pg_loss:np.float64(-0.09000486146578623) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.09476127481320873) - srpo/sdpo_loss:np.float64(0.0047564133474224946) - srpo/routed_loss:np.float64(-0.09000486146578623) - srpo/routed_token_count:np.float64(85.73828125) - srpo/correct_sample_fraction:np.float64(0.6171875) - srpo/sdpo_sample_fraction:np.float64(0.2265625) - srpo/grpo_sample_fraction:np.float64(0.7734375) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.006679293537047215) - self_distillation/dynamic_weight_mean:np.float64(0.22656250116415322) - self_distillation/dynamic_weight_max:np.float64(1.099640965461731) - self_distillation/empty_target_batch:np.float64(0.16796875) - actor/grad_norm:np.float64(0.22374041378498077) - perf/mfu/actor:np.float64(0.02275217005116409) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(149.5258445739746) - actor/lr:np.float64(5e-06) - training/global_step:83 - training/epoch:1 - critic/score/mean:0.6171875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.6171875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.0030753109604120255 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.0030753109604120255 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:85.73828125 - response_length/max:198.0 - response_length/min:40.0 - response_length/clip_ra
step:84 - global_seqlen/min:9379 - global_seqlen/max:13777 - global_seqlen/minmax_diff:4398 - global_seqlen/balanced_min:10774 - global_seqlen/balanced_max:10775 - global_seqlen/mean:10774.75 - actor/entropy:0.06580796092748642 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.1777186393737793 - training/rollout_probs_diff_mean:0.001965097850188613 - training/rollout_probs_diff_std:0.008992645889520645 - training/rollout_actor_probs_pearson_corr:0.9978021383285522 - self_distillation/success_group_fraction:1.0 - self_distillation/success_sample_fraction:0.99609375 - self_distillation/correct_sample_fraction:0.78515625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.99609375 - rollout_corr/rollout_is_mean:0.9998197555541992 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.6179242134094238 - rollout_corr/rollout_is_min:0.5374007821083069 - rollout_corr/rollout_is_std:0.02525528334081173 - rollout_corr/rollout_is_eff_sample_size:0.9993622796198363 - rollout_corr/rollout_is_seq_mean:0.9997678995132446 - rollout_corr/rollout_is_seq_std:0.00277720857411623 - rollout_corr/rollout_is_seq_max:1.0114127397537231 - rollout_corr/rollout_is_seq_min:0.9875314831733704 - rollout_corr/rollout_is_seq_max_deviation:0.012468516826629639 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.0702485120855272) - rollout_corr/training_log_ppl:np.float64(0.06685088950325735) - rollout_corr/kl:np.float64(0.0005534151690831379) - rollout_corr/k3_kl:np.float64(0.0003213346852004896) - rollout_corr/rollout_ppl:np.float64(1.069643889553845) - rollout_corr/rollout_log_ppl:np.float64(0.06629747333863634) - rollout_corr/log_ppl_diff:np.float64(0.0005534161646210123) - rollout_corr/log_ppl_abs_diff:np.float64(0.002140419830539031) - rollout_corr/log_ppl_diff_max:np.float64(0.014341384172439575) - rollout_corr/log_ppl_diff_min:np.float64(-0.010289102792739868) - rollout_corr/ppl_ratio:np.float64(1.0005577022675425) - rollout_corr/chi2_token:np.float64(0.00015612738206982613) - rollout_corr/chi2_seq:np.float64(0.021115493029356003) - actor/pg_loss:np.float64(-0.07807571551074943) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.08154095939244144) - srpo/sdpo_loss:np.float64(0.003465243881692004) - srpo/routed_loss:np.float64(-0.07807571551074943) - srpo/routed_token_count:np.float64(82.6796875) - srpo/correct_sample_fraction:np.float64(0.78515625) - srpo/sdpo_sample_fraction:np.float64(0.21484375) - srpo/grpo_sample_fraction:np.float64(0.78515625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.0038342640308002274) - self_distillation/dynamic_weight_mean:np.float64(0.21484375395812094) - self_distillation/dynamic_weight_max:np.float64(1.0777287483215332) - self_distillation/empty_target_batch:np.float64(0.00390625) - actor/grad_norm:np.float64(0.3104580342769623) - perf/mfu/actor:np.float64(0.01958911499748557) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(149.4275779724121) - actor/lr:np.float64(5e-06) - training/global_step:84 - training/epoch:1 - critic/score/mean:0.78515625 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.78515625 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.0036851554177701473 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.0036851554177701473 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:82.6796875 - response_length/max:143.0 - response_length/min:34.0 - response_length/clip_ratio
step:85 - global_seqlen/min:9940 - global_seqlen/max:15496 - global_seqlen/minmax_diff:5556 - global_seqlen/balanced_min:11801 - global_seqlen/balanced_max:11802 - global_seqlen/mean:11801.75 - actor/entropy:0.06800810247659683 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.1791505217552185 - training/rollout_probs_diff_mean:0.002521419432014227 - training/rollout_probs_diff_std:0.011691744439303875 - training/rollout_actor_probs_pearson_corr:0.9962990283966064 - self_distillation/success_group_fraction:1.0 - self_distillation/success_sample_fraction:0.99609375 - self_distillation/correct_sample_fraction:0.81640625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.99609375 - rollout_corr/rollout_is_mean:0.9998829960823059 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.5228978395462036 - rollout_corr/rollout_is_min:0.5504919290542603 - rollout_corr/rollout_is_std:0.029497602954506874 - rollout_corr/rollout_is_eff_sample_size:0.9991302908201731 - rollout_corr/rollout_is_seq_mean:0.9994535446166992 - rollout_corr/rollout_is_seq_std:0.004763338249176741 - rollout_corr/rollout_is_seq_max:1.0149345397949219 - rollout_corr/rollout_is_seq_min:0.9799094796180725 - rollout_corr/rollout_is_seq_max_deviation:0.02009052038192749 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.065216934774071) - rollout_corr/training_log_ppl:np.float64(0.06203309536067536) - rollout_corr/kl:np.float64(0.0010624460230346244) - rollout_corr/k3_kl:np.float64(0.0005159919557974207) - rollout_corr/rollout_ppl:np.float64(1.0640755323693156) - rollout_corr/rollout_log_ppl:np.float64(0.060970647566136904) - rollout_corr/log_ppl_diff:np.float64(0.001062447794538457) - rollout_corr/log_ppl_abs_diff:np.float64(0.0032615314703434706) - rollout_corr/log_ppl_diff_max:np.float64(0.025949347764253616) - rollout_corr/log_ppl_diff_min:np.float64(-0.013035841286182404) - rollout_corr/ppl_ratio:np.float64(1.0010764228645712) - rollout_corr/chi2_token:np.float64(-0.00012046704068779945) - rollout_corr/chi2_seq:np.float64(0.06827747984789312) - actor/pg_loss:np.float64(-0.06619404083971858) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.06991090776864439) - srpo/sdpo_loss:np.float64(0.0037168669289258105) - srpo/routed_loss:np.float64(-0.06619404083971858) - srpo/routed_token_count:np.float64(68.9921875) - srpo/correct_sample_fraction:np.float64(0.81640625) - srpo/sdpo_sample_fraction:np.float64(0.18359375) - srpo/grpo_sample_fraction:np.float64(0.81640625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.006575970779067575) - self_distillation/dynamic_weight_mean:np.float64(0.1835937537252903) - self_distillation/dynamic_weight_max:np.float64(1.1395808458328247) - self_distillation/empty_target_batch:np.float64(0.00390625) - actor/grad_norm:np.float64(0.32647716999053955) - perf/mfu/actor:np.float64(0.02168861307003504) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(149.32513046264648) - actor/lr:np.float64(5e-06) - training/global_step:85 - training/epoch:1 - critic/score/mean:0.81640625 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.81640625 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.003736836137250066 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.003736836137250066 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:68.9921875 - response_length/max:195.0 - response_length/min:23.0 - response_length/clip_rati
step:86 - global_seqlen/min:9265 - global_seqlen/max:11708 - global_seqlen/minmax_diff:2443 - global_seqlen/balanced_min:10221 - global_seqlen/balanced_max:10222 - global_seqlen/mean:10221.25 - actor/entropy:0.054270923137664795 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.27730315923690796 - training/rollout_probs_diff_mean:0.0022144122049212456 - training/rollout_probs_diff_std:0.012745820917189121 - training/rollout_actor_probs_pearson_corr:0.9945066571235657 - self_distillation/success_group_fraction:0.96875 - self_distillation/success_sample_fraction:0.953125 - self_distillation/correct_sample_fraction:0.6796875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.953125 - rollout_corr/rollout_is_mean:1.0000319480895996 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0003716781211551279 - rollout_corr/rollout_is_max:1.5546504259109497 - rollout_corr/rollout_is_min:0.44541478157043457 - rollout_corr/rollout_is_std:0.029084516689181328 - rollout_corr/rollout_is_eff_sample_size:0.9991548058387957 - rollout_corr/rollout_is_seq_mean:1.000037431716919 - rollout_corr/rollout_is_seq_std:0.0070443907752633095 - rollout_corr/rollout_is_seq_max:1.029192328453064 - rollout_corr/rollout_is_seq_min:0.9758851528167725 - rollout_corr/rollout_is_seq_max_deviation:0.029192328453063965 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.0564633077010512) - rollout_corr/training_log_ppl:np.float64(0.0537344533331634) - rollout_corr/kl:np.float64(0.0006043339651558011) - rollout_corr/k3_kl:np.float64(0.0006418050531819611) - rollout_corr/rollout_ppl:np.float64(1.0557562881149352) - rollout_corr/rollout_log_ppl:np.float64(0.053130117659748066) - rollout_corr/log_ppl_diff:np.float64(0.0006043357134331018) - rollout_corr/log_ppl_abs_diff:np.float64(0.00452952820705832) - rollout_corr/log_ppl_diff_max:np.float64(0.0351654514670372) - rollout_corr/log_ppl_diff_min:np.float64(-0.02322380617260933) - rollout_corr/ppl_ratio:np.float64(1.0006317517254502) - rollout_corr/chi2_token:np.float64(0.0012955761048942804) - rollout_corr/chi2_seq:np.float64(0.035272367764264345) - actor/pg_loss:np.float64(-0.0889014573476743) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.09944611025275663) - srpo/sdpo_loss:np.float64(0.01054465290508233) - srpo/routed_loss:np.float64(-0.0889014573476743) - srpo/routed_token_count:np.float64(42.0390625) - srpo/correct_sample_fraction:np.float64(0.6796875) - srpo/sdpo_sample_fraction:np.float64(0.2890625) - srpo/grpo_sample_fraction:np.float64(0.7109375) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.011092479211633588) - self_distillation/dynamic_weight_mean:np.float64(0.28906250884756446) - self_distillation/dynamic_weight_max:np.float64(1.1367307901382446) - self_distillation/empty_target_batch:np.float64(0.046875) - actor/grad_norm:np.float64(1.5443015098571777) - perf/mfu/actor:np.float64(0.018852890603117893) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(149.25756072998047) - actor/lr:np.float64(5e-06) - training/global_step:86 - training/epoch:1 - critic/score/mean:0.6796875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.6796875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.005319643300026655 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.005319643300026655 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:42.0390625 - response_length/max:130.0 - response_length/min:19.0 - response_length/clip_
step:87 - global_seqlen/min:9365 - global_seqlen/max:13996 - global_seqlen/minmax_diff:4631 - global_seqlen/balanced_min:11012 - global_seqlen/balanced_max:11013 - global_seqlen/mean:11012.125 - actor/entropy:0.05640158802270889 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.1243477463722229 - training/rollout_probs_diff_mean:0.0016045940574258566 - training/rollout_probs_diff_std:0.007987619377672672 - training/rollout_actor_probs_pearson_corr:0.9980933666229248 - self_distillation/success_group_fraction:0.9375 - self_distillation/success_sample_fraction:0.93359375 - self_distillation/correct_sample_fraction:0.65234375 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.93359375 - rollout_corr/rollout_is_mean:0.9997854828834534 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.5868679285049438 - rollout_corr/rollout_is_min:0.6087791323661804 - rollout_corr/rollout_is_std:0.020954890176653862 - rollout_corr/rollout_is_eff_sample_size:0.9995607875510558 - rollout_corr/rollout_is_seq_mean:0.9997974634170532 - rollout_corr/rollout_is_seq_std:0.00247522653080523 - rollout_corr/rollout_is_seq_max:1.00944185256958 - rollout_corr/rollout_is_seq_min:0.9893996715545654 - rollout_corr/rollout_is_seq_max_deviation:0.01060032844543457 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.0503069758415222) - rollout_corr/training_log_ppl:np.float64(0.047689266299130395) - rollout_corr/kl:np.float64(0.00036339917098127117) - rollout_corr/k3_kl:np.float64(0.00016090433209517419) - rollout_corr/rollout_ppl:np.float64(1.049904995597899) - rollout_corr/rollout_log_ppl:np.float64(0.04732586655063642) - rollout_corr/log_ppl_diff:np.float64(0.0003633997484939755) - rollout_corr/log_ppl_abs_diff:np.float64(0.0016135515079440665) - rollout_corr/log_ppl_diff_max:np.float64(0.01158994436264038) - rollout_corr/log_ppl_diff_min:np.float64(-0.008541479706764221) - rollout_corr/ppl_ratio:np.float64(1.0003665830008686) - rollout_corr/chi2_token:np.float64(-8.454406633973122e-05) - rollout_corr/chi2_seq:np.float64(-0.012059166096150875) - actor/pg_loss:np.float64(-0.10845235665874497) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.11576932697789744) - srpo/sdpo_loss:np.float64(0.00731697031915246) - srpo/routed_loss:np.float64(-0.10845235665874497) - srpo/routed_token_count:np.float64(48.66015625) - srpo/correct_sample_fraction:np.float64(0.65234375) - srpo/sdpo_sample_fraction:np.float64(0.28515625) - srpo/grpo_sample_fraction:np.float64(0.71484375) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.008207479996940492) - self_distillation/dynamic_weight_mean:np.float64(0.28515626047737896) - self_distillation/dynamic_weight_max:np.float64(1.0771037340164185) - self_distillation/empty_target_batch:np.float64(0.06640625) - actor/grad_norm:np.float64(0.5186740159988403) - perf/mfu/actor:np.float64(0.019873088604782964) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(149.2928123474121) - actor/lr:np.float64(5e-06) - training/global_step:87 - training/epoch:1 - critic/score/mean:0.65234375 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.65234375 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.0008027615258470178 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.0008027615258470178 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:48.66015625 - response_length/max:185.0 - response_length/min:23.0 - response_length/cl
step:88 - global_seqlen/min:8980 - global_seqlen/max:11157 - global_seqlen/minmax_diff:2177 - global_seqlen/balanced_min:10022 - global_seqlen/balanced_max:10022 - global_seqlen/mean:10022.0 - actor/entropy:0.06112293899059296 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.11700010299682617 - training/rollout_probs_diff_mean:0.0018625702941790223 - training/rollout_probs_diff_std:0.008895128034055233 - training/rollout_actor_probs_pearson_corr:0.9978421926498413 - self_distillation/success_group_fraction:0.9375 - self_distillation/success_sample_fraction:0.92578125 - self_distillation/correct_sample_fraction:0.62109375 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.92578125 - rollout_corr/rollout_is_mean:1.0000560283660889 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.4741427898406982 - rollout_corr/rollout_is_min:0.5226935744285583 - rollout_corr/rollout_is_std:0.02337128482758999 - rollout_corr/rollout_is_eff_sample_size:0.9994539621462206 - rollout_corr/rollout_is_seq_mean:1.0000741481781006 - rollout_corr/rollout_is_seq_std:0.0032028015702962875 - rollout_corr/rollout_is_seq_max:1.0208666324615479 - rollout_corr/rollout_is_seq_min:0.9822154641151428 - rollout_corr/rollout_is_seq_max_deviation:0.02086663246154785 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.05361582711339) - rollout_corr/training_log_ppl:np.float64(0.050667653704294935) - rollout_corr/kl:np.float64(0.00014744740528413458) - rollout_corr/k3_kl:np.float64(0.00022159051046601822) - rollout_corr/rollout_ppl:np.float64(1.053438679780811) - rollout_corr/rollout_log_ppl:np.float64(0.050520205406655805) - rollout_corr/log_ppl_diff:np.float64(0.00014744829763913003) - rollout_corr/log_ppl_abs_diff:np.float64(0.00189981911103132) - rollout_corr/log_ppl_diff_max:np.float64(0.021829113364219666) - rollout_corr/log_ppl_diff_min:np.float64(-0.017112985253334045) - rollout_corr/ppl_ratio:np.float64(1.0001527501735836) - rollout_corr/chi2_token:np.float64(0.0005885227583348751) - rollout_corr/chi2_seq:np.float64(0.03134301328100264) - actor/pg_loss:np.float64(-0.08731241254281485) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.09629750388558023) - srpo/sdpo_loss:np.float64(0.008985091342765372) - srpo/routed_loss:np.float64(-0.08731241254281485) - srpo/routed_token_count:np.float64(44.375) - srpo/correct_sample_fraction:np.float64(0.62109375) - srpo/sdpo_sample_fraction:np.float64(0.31640625) - srpo/grpo_sample_fraction:np.float64(0.68359375) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.013783988112209045) - self_distillation/dynamic_weight_mean:np.float64(0.3164062520954758) - self_distillation/dynamic_weight_max:np.float64(1.1933176517486572) - self_distillation/empty_target_batch:np.float64(0.07421875) - actor/grad_norm:np.float64(0.38641032576560974) - perf/mfu/actor:np.float64(0.01848338937422251) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(149.23738479614258) - actor/lr:np.float64(5e-06) - training/global_step:88 - training/epoch:1 - critic/score/mean:0.62109375 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.62109375 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.00042913731886073947 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.00042913731886073947 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:44.375 - response_length/max:139.0 - response_length/min:23.0 - response_length/clip_ratio:0.
step:89 - global_seqlen/min:8448 - global_seqlen/max:13428 - global_seqlen/minmax_diff:4980 - global_seqlen/balanced_min:10734 - global_seqlen/balanced_max:10735 - global_seqlen/mean:10734.75 - actor/entropy:0.06133819371461868 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.12433949112892151 - training/rollout_probs_diff_mean:0.001804135157726705 - training/rollout_probs_diff_std:0.00893300399184227 - training/rollout_actor_probs_pearson_corr:0.99761962890625 - self_distillation/success_group_fraction:0.9375 - self_distillation/success_sample_fraction:0.93359375 - self_distillation/correct_sample_fraction:0.671875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.93359375 - rollout_corr/rollout_is_mean:1.0003291368484497 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.5629695653915405 - rollout_corr/rollout_is_min:0.778823971748352 - rollout_corr/rollout_is_std:0.024428706616163254 - rollout_corr/rollout_is_eff_sample_size:0.9994039514101857 - rollout_corr/rollout_is_seq_mean:1.0002192258834839 - rollout_corr/rollout_is_seq_std:0.003044999437406659 - rollout_corr/rollout_is_seq_max:1.014376163482666 - rollout_corr/rollout_is_seq_min:0.9909852743148804 - rollout_corr/rollout_is_seq_max_deviation:0.014376163482666016 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.0406372211873531) - rollout_corr/training_log_ppl:np.float64(0.03863605880746945) - rollout_corr/kl:np.float64(-3.801287108018414e-05) - rollout_corr/k3_kl:np.float64(0.00018123817688081578) - rollout_corr/rollout_ppl:np.float64(1.040678933262825) - rollout_corr/rollout_log_ppl:np.float64(0.03867407105190068) - rollout_corr/log_ppl_diff:np.float64(-3.801224443122919e-05) - rollout_corr/log_ppl_abs_diff:np.float64(0.0015884851206919848) - rollout_corr/log_ppl_diff_max:np.float64(0.01014639437198639) - rollout_corr/log_ppl_diff_min:np.float64(-0.012935452163219452) - rollout_corr/ppl_ratio:np.float64(0.9999661855399609) - rollout_corr/chi2_token:np.float64(0.0008213554974645376) - rollout_corr/chi2_seq:np.float64(0.06126519222743809) - actor/pg_loss:np.float64(-0.1089662706344825) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.11632535816170275) - srpo/sdpo_loss:np.float64(0.007359087527220254) - srpo/routed_loss:np.float64(-0.1089662706344825) - srpo/routed_token_count:np.float64(44.1171875) - srpo/correct_sample_fraction:np.float64(0.671875) - srpo/sdpo_sample_fraction:np.float64(0.265625) - srpo/grpo_sample_fraction:np.float64(0.734375) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.007995159414690534) - self_distillation/dynamic_weight_mean:np.float64(0.2656250046566129) - self_distillation/dynamic_weight_max:np.float64(1.159695029258728) - self_distillation/empty_target_batch:np.float64(0.06640625) - actor/grad_norm:np.float64(0.5853164792060852) - perf/mfu/actor:np.float64(0.01944073692802734) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(149.2334771156311) - actor/lr:np.float64(5e-06) - training/global_step:89 - training/epoch:1 - critic/score/mean:0.671875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.671875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.015351071022450924 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.015351071022450924 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:44.1171875 - response_length/max:318.0 - response_length/min:23.0 - response_length/clip_ratio:0.0 - response_l
test_gen_batch meta info: {'eos_token_id': 151645, 'pad_token_id': 151643, 'recompute_log_prob': False, 'do_sample': True, 'validate': True, 'global_steps': 90}
validation generation end
[prompt] system

Given a question and four options, please select the right answer. Respond in the following format:
<reasoning>
...
</reasoning>
<answer>
...
</answer>

For the answer, only output the letter corresponding to the correct option (A, B, C, or D), and nothing else. Do not restate the answer text. For example, if the answer is "A", just output:
<answer>
A
</answer>

user
Which of the following molecules is the only correct reactant used in the synthesis of the product "Nc1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1" ?

A: O=C1c2ccccc2C(=O)N1c1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1
B: O=C1c2ccccc2C(Cl)N1c1ccc2ccc(Cl)nc2n1
C: O=C1c2ccccc2C(=O)N1Cc1ccc(Oc2ccc(Cl)c(C(F)(F)F)c2)cc1
D: O=C1c2ccccc2C(=O)N1CC=C(Cc1ccccc1)C(F)(F)F
Please reason step by step.
assistant
<think>

</think>


[response] <reasoning>
The molecule "Nc1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1" is a nitrile group with a chlorine-containing benzene ring and a trifluorinated group. The correct reactant is option A, which includes the nitrile group, a chlorine-containing benzene ring, and a trifluorinated group. The correct answer is A.
</reasoning>
<answer>
A
</answer>
[ground_truth] A
[score] 1.0
[acc] 1.0
[pred] A
[incorrect_format] 1
[feedback]
local_global_step_folder: /workspace/L2T/checkpoints/datasets/sciknoweval/chemistry/qwen3gen-chemistry-SRPO_TR-Qwen-Qwen3-4B-mbs32-tr0.1-dwtrue-train32-rollout8-lr5e-6-vllm0.8/global_step_90
step:90 - global_seqlen/min:8644 - global_seqlen/max:13546 - global_seqlen/minmax_diff:4902 - global_seqlen/balanced_min:10357 - global_seqlen/balanced_max:10358 - global_seqlen/mean:10357.75 - actor/entropy:0.050647325813770294 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.12433096766471863 - training/rollout_probs_diff_mean:0.0010999547084793448 - training/rollout_probs_diff_std:0.006013025064021349 - training/rollout_actor_probs_pearson_corr:0.9986728429794312 - self_distillation/success_group_fraction:0.96875 - self_distillation/success_sample_fraction:0.96875 - self_distillation/correct_sample_fraction:0.74609375 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.96875 - rollout_corr/rollout_is_mean:1.0000684261322021 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.2624804973602295 - rollout_corr/rollout_is_min:0.6087054014205933 - rollout_corr/rollout_is_std:0.015765512362122536 - rollout_corr/rollout_is_eff_sample_size:0.9997515103937183 - rollout_corr/rollout_is_seq_mean:1.0000416040420532 - rollout_corr/rollout_is_seq_std:0.001838645082898438 - rollout_corr/rollout_is_seq_max:1.0083354711532593 - rollout_corr/rollout_is_seq_min:0.9938872456550598 - rollout_corr/rollout_is_seq_max_deviation:0.008335471153259277 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.0376063282601535) - rollout_corr/training_log_ppl:np.float64(0.03615942332456257) - rollout_corr/kl:np.float64(4.7036995986626806e-05) - rollout_corr/k3_kl:np.float64(8.866228545657698e-05) - rollout_corr/rollout_ppl:np.float64(1.0375485867261887) - rollout_corr/rollout_log_ppl:np.float64(0.03611238591574306) - rollout_corr/log_ppl_diff:np.float64(4.70374088195058e-05) - rollout_corr/log_ppl_abs_diff:np.float64(0.0011756470902923866) - rollout_corr/log_ppl_diff_max:np.float64(0.0073887258768081665) - rollout_corr/log_ppl_diff_min:np.float64(-0.0076553598046302795) - rollout_corr/ppl_ratio:np.float64(1.0000487987417728) - rollout_corr/chi2_token:np.float64(0.0002544571179896593) - rollout_corr/chi2_seq:np.float64(0.02463704999536276) - actor/pg_loss:np.float64(-0.10643497332421248) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.1117768238473218) - srpo/sdpo_loss:np.float64(0.005341850523109315) - srpo/routed_loss:np.float64(-0.10643497332421248) - srpo/routed_token_count:np.float64(45.8046875) - srpo/correct_sample_fraction:np.float64(0.74609375) - srpo/sdpo_sample_fraction:np.float64(0.22265625) - srpo/grpo_sample_fraction:np.float64(0.77734375) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.007365271829257836) - self_distillation/dynamic_weight_mean:np.float64(0.22265625116415322) - self_distillation/dynamic_weight_max:np.float64(1.0523344278335571) - self_distillation/empty_target_batch:np.float64(0.03125) - actor/grad_norm:np.float64(0.23571494221687317) - perf/mfu/actor:np.float64(0.01913403029337096) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(149.2429542541504) - actor/lr:np.float64(5e-06) - val-aux/sciknoweval/reward/mean@16:np.float64(0.7348214285714286) - val-aux/sciknoweval/reward/std@16:np.float64(0.13294413219085893) - val-aux/sciknoweval/reward/best@2/mean:np.float64(0.7899809523809525) - val-aux/sciknoweval/reward/best@2/std:np.float64(0.11468444295124176) - val-aux/sciknoweval/reward/worst@2/mean:np.float64(0.680457142857143) - val-aux/sciknoweval/reward/worst@2/std:np.float64(0.10994216014457459) - val-aux/sciknoweval/reward/maj@2/mean:np.float64(0.735295238095238) - val-aux/sciknoweval/reward/maj@2/
step:91 - global_seqlen/min:8397 - global_seqlen/max:16973 - global_seqlen/minmax_diff:8576 - global_seqlen/balanced_min:10570 - global_seqlen/balanced_max:10571 - global_seqlen/mean:10570.75 - actor/entropy:0.045913517475128174 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.18239784240722656 - training/rollout_probs_diff_mean:0.0011290523689240217 - training/rollout_probs_diff_std:0.006749647669494152 - training/rollout_actor_probs_pearson_corr:0.9982606172561646 - self_distillation/success_group_fraction:1.0 - self_distillation/success_sample_fraction:0.99609375 - self_distillation/correct_sample_fraction:0.77734375 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.99609375 - rollout_corr/rollout_is_mean:1.0000507831573486 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.404425859451294 - rollout_corr/rollout_is_min:0.640227198600769 - rollout_corr/rollout_is_std:0.018168393522500992 - rollout_corr/rollout_is_eff_sample_size:0.9996697801398317 - rollout_corr/rollout_is_seq_mean:1.00014328956604 - rollout_corr/rollout_is_seq_std:0.0021028094924986362 - rollout_corr/rollout_is_seq_max:1.0117266178131104 - rollout_corr/rollout_is_seq_min:0.9924172759056091 - rollout_corr/rollout_is_seq_max_deviation:0.011726617813110352 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.0355504644103348) - rollout_corr/training_log_ppl:np.float64(0.033772075178205796) - rollout_corr/kl:np.float64(-3.1552499545783874e-05) - rollout_corr/k3_kl:np.float64(0.00011171491833161969) - rollout_corr/rollout_ppl:np.float64(1.0355919506400824) - rollout_corr/rollout_log_ppl:np.float64(0.03380362737621567) - rollout_corr/log_ppl_diff:np.float64(-3.155219800987652e-05) - rollout_corr/log_ppl_abs_diff:np.float64(0.0011742089249366927) - rollout_corr/log_ppl_diff_max:np.float64(0.009898841381072998) - rollout_corr/log_ppl_diff_min:np.float64(-0.010418757796287537) - rollout_corr/ppl_ratio:np.float64(0.9999705930240452) - rollout_corr/chi2_token:np.float64(0.0005139871500432491) - rollout_corr/chi2_seq:np.float64(0.014233705587685108) - actor/pg_loss:np.float64(-0.08923335366262108) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.09424315884825774) - srpo/sdpo_loss:np.float64(0.005009805185636651) - srpo/routed_loss:np.float64(-0.08923335366262108) - srpo/routed_token_count:np.float64(44.2109375) - srpo/correct_sample_fraction:np.float64(0.77734375) - srpo/sdpo_sample_fraction:np.float64(0.22265625) - srpo/grpo_sample_fraction:np.float64(0.77734375) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.009194281774625779) - self_distillation/dynamic_weight_mean:np.float64(0.22265624650754035) - self_distillation/dynamic_weight_max:np.float64(1.1660722494125366) - self_distillation/empty_target_batch:np.float64(0.00390625) - actor/grad_norm:np.float64(0.18805918097496033) - perf/mfu/actor:np.float64(0.019317535741041995) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(150.32583236694336) - actor/lr:np.float64(5e-06) - training/global_step:91 - training/epoch:1 - critic/score/mean:0.77734375 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.77734375 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.008294309489428997 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.008294309489428997 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:44.2109375 - response_length/max:226.0 - response_length/min:23.0 - response_length/cl
step:92 - global_seqlen/min:9516 - global_seqlen/max:11575 - global_seqlen/minmax_diff:2059 - global_seqlen/balanced_min:10667 - global_seqlen/balanced_max:10667 - global_seqlen/mean:10667.0 - actor/entropy:0.05356821045279503 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.12265193462371826 - training/rollout_probs_diff_mean:0.001480194739997387 - training/rollout_probs_diff_std:0.007430844008922577 - training/rollout_actor_probs_pearson_corr:0.998096764087677 - self_distillation/success_group_fraction:0.90625 - self_distillation/success_sample_fraction:0.89453125 - self_distillation/correct_sample_fraction:0.6875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.89453125 - rollout_corr/rollout_is_mean:1.000179648399353 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.440192461013794 - rollout_corr/rollout_is_min:0.6785665154457092 - rollout_corr/rollout_is_std:0.01947011984884739 - rollout_corr/rollout_is_eff_sample_size:0.9996211772295936 - rollout_corr/rollout_is_seq_mean:1.0000736713409424 - rollout_corr/rollout_is_seq_std:0.0020817688200622797 - rollout_corr/rollout_is_seq_max:1.0087515115737915 - rollout_corr/rollout_is_seq_min:0.986434280872345 - rollout_corr/rollout_is_seq_max_deviation:0.01356571912765503 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.0390669042244554) - rollout_corr/training_log_ppl:np.float64(0.037364652206633764) - rollout_corr/kl:np.float64(4.6983253202625974e-05) - rollout_corr/k3_kl:np.float64(0.00012067534343655414) - rollout_corr/rollout_ppl:np.float64(1.0390021805651486) - rollout_corr/rollout_log_ppl:np.float64(0.0373176684615828) - rollout_corr/log_ppl_diff:np.float64(4.698374505096581e-05) - rollout_corr/log_ppl_abs_diff:np.float64(0.0013520465181500185) - rollout_corr/log_ppl_diff_max:np.float64(0.0158633291721344) - rollout_corr/log_ppl_diff_min:np.float64(-0.00813911110162735) - rollout_corr/ppl_ratio:np.float64(1.0000492511317134) - rollout_corr/chi2_token:np.float64(0.00038604438304901123) - rollout_corr/chi2_seq:np.float64(0.03082447638735175) - actor/pg_loss:np.float64(-0.06937670237357452) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.07422164085437544) - srpo/sdpo_loss:np.float64(0.00484493848080092) - srpo/routed_loss:np.float64(-0.06937670237357452) - srpo/routed_token_count:np.float64(45.84375) - srpo/correct_sample_fraction:np.float64(0.6875) - srpo/sdpo_sample_fraction:np.float64(0.21875) - srpo/grpo_sample_fraction:np.float64(0.78125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.0080922304273372) - self_distillation/dynamic_weight_mean:np.float64(0.21874999883584678) - self_distillation/dynamic_weight_max:np.float64(1.0980885028839111) - self_distillation/empty_target_batch:np.float64(0.10546875) - actor/grad_norm:np.float64(0.256654292345047) - perf/mfu/actor:np.float64(0.019798272550026898) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(150.10909271240234) - actor/lr:np.float64(5e-06) - training/global_step:92 - training/epoch:1 - critic/score/mean:0.6875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.6875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.0097669567912817 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.0097669567912817 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:45.84375 - response_length/max:174.0 - response_length/min:23.0 - response_length/clip_ratio:0.0 - response_length_non_aborted/m
step:93 - global_seqlen/min:7860 - global_seqlen/max:14215 - global_seqlen/minmax_diff:6355 - global_seqlen/balanced_min:9860 - global_seqlen/balanced_max:9861 - global_seqlen/mean:9860.75 - actor/entropy:0.03537750989198685 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.11699521541595459 - training/rollout_probs_diff_mean:0.0007877535535953939 - training/rollout_probs_diff_std:0.005052958149462938 - training/rollout_actor_probs_pearson_corr:0.9988642334938049 - self_distillation/success_group_fraction:0.9375 - self_distillation/success_sample_fraction:0.93359375 - self_distillation/correct_sample_fraction:0.77734375 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.93359375 - rollout_corr/rollout_is_mean:0.9999171495437622 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.2373543977737427 - rollout_corr/rollout_is_min:0.5575910210609436 - rollout_corr/rollout_is_std:0.014393948018550873 - rollout_corr/rollout_is_eff_sample_size:0.999792857171802 - rollout_corr/rollout_is_seq_mean:0.9998306035995483 - rollout_corr/rollout_is_seq_std:0.001692939200438559 - rollout_corr/rollout_is_seq_max:1.0051791667938232 - rollout_corr/rollout_is_seq_min:0.9898188710212708 - rollout_corr/rollout_is_seq_max_deviation:0.010181128978729248 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.033122431486845) - rollout_corr/training_log_ppl:np.float64(0.031127349807434257) - rollout_corr/kl:np.float64(0.0002478227741659067) - rollout_corr/k3_kl:np.float64(7.84277350363168e-05) - rollout_corr/rollout_ppl:np.float64(1.032839698716998) - rollout_corr/rollout_log_ppl:np.float64(0.030879526691705905) - rollout_corr/log_ppl_diff:np.float64(0.0002478231157283517) - rollout_corr/log_ppl_abs_diff:np.float64(0.0008903827999517944) - rollout_corr/log_ppl_diff_max:np.float64(0.011361658573150635) - rollout_corr/log_ppl_diff_min:np.float64(-0.004744093865156174) - rollout_corr/ppl_ratio:np.float64(1.0002494438085705) - rollout_corr/chi2_token:np.float64(-0.00019223429262638092) - rollout_corr/chi2_seq:np.float64(0.001075179548934102) - actor/pg_loss:np.float64(-0.058158593599273445) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.06393161907908507) - srpo/sdpo_loss:np.float64(0.005773025479811622) - srpo/routed_loss:np.float64(-0.058158593599273445) - srpo/routed_token_count:np.float64(41.6796875) - srpo/correct_sample_fraction:np.float64(0.77734375) - srpo/sdpo_sample_fraction:np.float64(0.16015625) - srpo/grpo_sample_fraction:np.float64(0.83984375) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.0056535559747317166) - self_distillation/dynamic_weight_mean:np.float64(0.16015625395812094) - self_distillation/dynamic_weight_max:np.float64(1.1655592918395996) - self_distillation/empty_target_batch:np.float64(0.06640625) - actor/grad_norm:np.float64(0.3563731610774994) - perf/mfu/actor:np.float64(0.018203451569450062) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(149.78484344482422) - actor/lr:np.float64(5e-06) - training/global_step:93 - training/epoch:1 - critic/score/mean:0.77734375 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.77734375 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.0035145266447216272 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.0035145266447216272 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:41.6796875 - response_length/max:118.0 - response_length/min:23.0 - response_length/c
step:94 - global_seqlen/min:8396 - global_seqlen/max:17424 - global_seqlen/minmax_diff:9028 - global_seqlen/balanced_min:9377 - global_seqlen/balanced_max:17323 - global_seqlen/mean:10370.5 - actor/entropy:0.01709718443453312 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.06593501567840576 - training/rollout_probs_diff_mean:0.0003720529202837497 - training/rollout_probs_diff_std:0.003450233256444335 - training/rollout_actor_probs_pearson_corr:0.9987819194793701 - self_distillation/success_group_fraction:0.96875 - self_distillation/success_sample_fraction:0.9609375 - self_distillation/correct_sample_fraction:0.8203125 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.9609375 - rollout_corr/rollout_is_mean:1.0000293254852295 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.6460747718811035 - rollout_corr/rollout_is_min:0.7791339755058289 - rollout_corr/rollout_is_std:0.010675395838916302 - rollout_corr/rollout_is_eff_sample_size:0.9998860489055212 - rollout_corr/rollout_is_seq_mean:1.0000336170196533 - rollout_corr/rollout_is_seq_std:0.0016057805623859167 - rollout_corr/rollout_is_seq_max:1.0111995935440063 - rollout_corr/rollout_is_seq_min:0.994225800037384 - rollout_corr/rollout_is_seq_max_deviation:0.011199593544006348 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.023848693817854) - rollout_corr/training_log_ppl:np.float64(0.02259953225446054) - rollout_corr/kl:np.float64(3.330237713068673e-05) - rollout_corr/k3_kl:np.float64(6.687154982689889e-05) - rollout_corr/rollout_ppl:np.float64(1.023817187640816) - rollout_corr/rollout_log_ppl:np.float64(0.022566229374973545) - rollout_corr/log_ppl_diff:np.float64(3.330287948699606e-05) - rollout_corr/log_ppl_abs_diff:np.float64(0.000667275483962726) - rollout_corr/log_ppl_diff_max:np.float64(0.006298858672380447) - rollout_corr/log_ppl_diff_min:np.float64(-0.00886402279138565) - rollout_corr/ppl_ratio:np.float64(1.0000344694126397) - rollout_corr/chi2_token:np.float64(0.00021080649457871914) - rollout_corr/chi2_seq:np.float64(0.009841396240517497) - actor/pg_loss:np.float64(-0.048936141346985096) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.05268722178880125) - srpo/sdpo_loss:np.float64(0.003751080441816157) - srpo/routed_loss:np.float64(-0.048936141346985096) - srpo/routed_token_count:np.float64(71.265625) - srpo/correct_sample_fraction:np.float64(0.8203125) - srpo/sdpo_sample_fraction:np.float64(0.1484375) - srpo/grpo_sample_fraction:np.float64(0.8515625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.006305594161947425) - self_distillation/dynamic_weight_mean:np.float64(0.1484375) - self_distillation/dynamic_weight_max:np.float64(1.1505732536315918) - self_distillation/empty_target_batch:np.float64(0.0390625) - actor/grad_norm:np.float64(0.16345210373401642) - perf/mfu/actor:np.float64(0.018066219747790796) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(149.76529693603516) - actor/lr:np.float64(5e-06) - training/global_step:94 - training/epoch:1 - critic/score/mean:0.8203125 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.8203125 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.3908408284187317 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.3908408284187317 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:71.265625 - response_length/max:8192.0 - response_length/min:23.0 - response_length/clip_ratio:0.00390625 - res
step:95 - global_seqlen/min:8021 - global_seqlen/max:13895 - global_seqlen/minmax_diff:5874 - global_seqlen/balanced_min:9787 - global_seqlen/balanced_max:9788 - global_seqlen/mean:9787.125 - actor/entropy:0.03951212391257286 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.10855144262313843 - training/rollout_probs_diff_mean:0.0009532520780339837 - training/rollout_probs_diff_std:0.005827426444739103 - training/rollout_actor_probs_pearson_corr:0.9985724687576294 - self_distillation/success_group_fraction:0.875 - self_distillation/success_sample_fraction:0.859375 - self_distillation/correct_sample_fraction:0.5859375 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.859375 - rollout_corr/rollout_is_mean:1.0000454187393188 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.639607310295105 - rollout_corr/rollout_is_min:0.6534220576286316 - rollout_corr/rollout_is_std:0.018695497885346413 - rollout_corr/rollout_is_eff_sample_size:0.9996507196117422 - rollout_corr/rollout_is_seq_mean:1.0001091957092285 - rollout_corr/rollout_is_seq_std:0.0029185484163463116 - rollout_corr/rollout_is_seq_max:1.027817726135254 - rollout_corr/rollout_is_seq_min:0.990573525428772 - rollout_corr/rollout_is_seq_max_deviation:0.027817726135253906 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.040347813628614) - rollout_corr/training_log_ppl:np.float64(0.038059175688431424) - rollout_corr/kl:np.float64(3.8311105930777245e-05) - rollout_corr/k3_kl:np.float64(0.00014752023374336705) - rollout_corr/rollout_ppl:np.float64(1.0403347304090858) - rollout_corr/rollout_log_ppl:np.float64(0.03802086352538936) - rollout_corr/log_ppl_diff:np.float64(3.831216304206464e-05) - rollout_corr/log_ppl_abs_diff:np.float64(0.0013346494031338807) - rollout_corr/log_ppl_diff_max:np.float64(0.010093554854393005) - rollout_corr/log_ppl_diff_min:np.float64(-0.02150675654411316) - rollout_corr/ppl_ratio:np.float64(1.0000418983399868) - rollout_corr/chi2_token:np.float64(0.000539576169103384) - rollout_corr/chi2_seq:np.float64(0.01673857681453228) - actor/pg_loss:np.float64(-0.0979807126104788) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.10349075408885255) - srpo/sdpo_loss:np.float64(0.005510041478373751) - srpo/routed_loss:np.float64(-0.0979807126104788) - srpo/routed_token_count:np.float64(42.66015625) - srpo/correct_sample_fraction:np.float64(0.5859375) - srpo/sdpo_sample_fraction:np.float64(0.2890625) - srpo/grpo_sample_fraction:np.float64(0.7109375) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.005359917184364349) - self_distillation/dynamic_weight_mean:np.float64(0.28906251047737896) - self_distillation/dynamic_weight_max:np.float64(1.0897819995880127) - self_distillation/empty_target_batch:np.float64(0.140625) - actor/grad_norm:np.float64(0.41541552543640137) - perf/mfu/actor:np.float64(0.018097834272512908) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(149.71403121948242) - actor/lr:np.float64(5e-06) - training/global_step:95 - training/epoch:1 - critic/score/mean:0.5859375 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.5859375 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.0013277172110974789 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.0013277172110974789 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:42.66015625 - response_length/max:105.0 - response_length/min:23.0 - response_length/clip_ratio:0.0 - res
step:96 - global_seqlen/min:8583 - global_seqlen/max:11657 - global_seqlen/minmax_diff:3074 - global_seqlen/balanced_min:9851 - global_seqlen/balanced_max:9851 - global_seqlen/mean:9851.0 - actor/entropy:0.028916779905557632 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.08651548624038696 - training/rollout_probs_diff_mean:0.0006886937189847231 - training/rollout_probs_diff_std:0.004749463405460119 - training/rollout_actor_probs_pearson_corr:0.9985435605049133 - self_distillation/success_group_fraction:0.90625 - self_distillation/success_sample_fraction:0.8984375 - self_distillation/correct_sample_fraction:0.7109375 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.8984375 - rollout_corr/rollout_is_mean:0.9998740553855896 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.2815817594528198 - rollout_corr/rollout_is_min:0.6252192258834839 - rollout_corr/rollout_is_std:0.011570308357477188 - rollout_corr/rollout_is_eff_sample_size:0.9998661458871566 - rollout_corr/rollout_is_seq_mean:0.9999244809150696 - rollout_corr/rollout_is_seq_std:0.0018198898760601878 - rollout_corr/rollout_is_seq_max:1.0042718648910522 - rollout_corr/rollout_is_seq_min:0.99330735206604 - rollout_corr/rollout_is_seq_max_deviation:0.006692647933959961 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.025174856185913) - rollout_corr/training_log_ppl:np.float64(0.02430382329509939) - rollout_corr/kl:np.float64(0.00013897788870775862) - rollout_corr/k3_kl:np.float64(6.350690899730978e-05) - rollout_corr/rollout_ppl:np.float64(1.0250237314030528) - rollout_corr/rollout_log_ppl:np.float64(0.024164845190796314) - rollout_corr/log_ppl_diff:np.float64(0.000138978104303078) - rollout_corr/log_ppl_abs_diff:np.float64(0.0010294890007003232) - rollout_corr/log_ppl_diff_max:np.float64(0.007239937782287598) - rollout_corr/log_ppl_diff_min:np.float64(-0.003980681300163269) - rollout_corr/ppl_ratio:np.float64(1.0001407612580806) - rollout_corr/chi2_token:np.float64(-2.9427697882056236e-05) - rollout_corr/chi2_seq:np.float64(-0.004812988685443997) - actor/pg_loss:np.float64(-0.0680088848566811) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.07126857689581811) - srpo/sdpo_loss:np.float64(0.0032596920391370077) - srpo/routed_loss:np.float64(-0.0680088848566811) - srpo/routed_token_count:np.float64(41.1875) - srpo/correct_sample_fraction:np.float64(0.7109375) - srpo/sdpo_sample_fraction:np.float64(0.1953125) - srpo/grpo_sample_fraction:np.float64(0.8046875) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.005766489975485456) - self_distillation/dynamic_weight_mean:np.float64(0.19531250232830644) - self_distillation/dynamic_weight_max:np.float64(1.0971882343292236) - self_distillation/empty_target_batch:np.float64(0.1015625) - actor/grad_norm:np.float64(0.4241745173931122) - perf/mfu/actor:np.float64(0.01817929308889902) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(149.6992530822754) - actor/lr:np.float64(5e-06) - training/global_step:96 - training/epoch:1 - critic/score/mean:0.7109375 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.7109375 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.0012092185206711292 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.0012092185206711292 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:41.1875 - response_length/max:112.0 - response_length/min:23.0 - response_length/clip_ratio:0.0 - resp
step:97 - global_seqlen/min:8915 - global_seqlen/max:15212 - global_seqlen/minmax_diff:6297 - global_seqlen/balanced_min:10801 - global_seqlen/balanced_max:10802 - global_seqlen/mean:10801.375 - actor/entropy:0.03493117541074753 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.09514814615249634 - training/rollout_probs_diff_mean:0.0007625046418979764 - training/rollout_probs_diff_std:0.005210522562265396 - training/rollout_actor_probs_pearson_corr:0.9985776543617249 - self_distillation/success_group_fraction:0.84375 - self_distillation/success_sample_fraction:0.8359375 - self_distillation/correct_sample_fraction:0.60546875 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.8359375 - rollout_corr/rollout_is_mean:1.0000402927398682 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.5994430780410767 - rollout_corr/rollout_is_min:0.7657786011695862 - rollout_corr/rollout_is_std:0.01429004967212677 - rollout_corr/rollout_is_eff_sample_size:0.9997957170177736 - rollout_corr/rollout_is_seq_mean:1.000033974647522 - rollout_corr/rollout_is_seq_std:0.0020038574002683163 - rollout_corr/rollout_is_seq_max:1.0136959552764893 - rollout_corr/rollout_is_seq_min:0.9948128461837769 - rollout_corr/rollout_is_seq_max_deviation:0.013695955276489258 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.0315454411320388) - rollout_corr/training_log_ppl:np.float64(0.029900110626527976) - rollout_corr/kl:np.float64(4.717255652500474e-05) - rollout_corr/k3_kl:np.float64(8.122715516936056e-05) - rollout_corr/rollout_ppl:np.float64(1.0315131559036672) - rollout_corr/rollout_log_ppl:np.float64(0.02985293752053053) - rollout_corr/log_ppl_diff:np.float64(4.717310599744451e-05) - rollout_corr/log_ppl_abs_diff:np.float64(0.0010645087408107656) - rollout_corr/log_ppl_diff_max:np.float64(0.005778759717941284) - rollout_corr/log_ppl_diff_min:np.float64(-0.01260647177696228) - rollout_corr/ppl_ratio:np.float64(1.0000491121318191) - rollout_corr/chi2_token:np.float64(0.0002365082036703825) - rollout_corr/chi2_seq:np.float64(0.011892619775608182) - actor/pg_loss:np.float64(-0.07281802542047444) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.07856842808541842) - srpo/sdpo_loss:np.float64(0.0057504026649439766) - srpo/routed_loss:np.float64(-0.07281802542047444) - srpo/routed_token_count:np.float64(43.13671875) - srpo/correct_sample_fraction:np.float64(0.60546875) - srpo/sdpo_sample_fraction:np.float64(0.23828125) - srpo/grpo_sample_fraction:np.float64(0.76171875) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.006044773168468964) - self_distillation/dynamic_weight_mean:np.float64(0.2382812574505806) - self_distillation/dynamic_weight_max:np.float64(1.1180425882339478) - self_distillation/empty_target_batch:np.float64(0.1640625) - actor/grad_norm:np.float64(0.38245293498039246) - perf/mfu/actor:np.float64(0.019929897066596546) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(149.7938346862793) - actor/lr:np.float64(5e-06) - training/global_step:97 - training/epoch:1 - critic/score/mean:0.60546875 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.60546875 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:0.0023544325958937407 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:0.0023544325958937407 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:43.13671875 - response_length/max:94.0 - response_length/min:23.0 - response_length/clip
step:98 - global_seqlen/min:8168 - global_seqlen/max:13742 - global_seqlen/minmax_diff:5574 - global_seqlen/balanced_min:11068 - global_seqlen/balanced_max:11069 - global_seqlen/mean:11068.875 - actor/entropy:0.049632541835308075 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.12028932571411133 - training/rollout_probs_diff_mean:0.0011203591711819172 - training/rollout_probs_diff_std:0.006207714788615704 - training/rollout_actor_probs_pearson_corr:0.9985822439193726 - self_distillation/success_group_fraction:0.84375 - self_distillation/success_sample_fraction:0.84375 - self_distillation/correct_sample_fraction:0.75 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.84375 - rollout_corr/rollout_is_mean:1.0000295639038086 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:0.0 - rollout_corr/rollout_is_max:1.43240487575531 - rollout_corr/rollout_is_min:0.6283036470413208 - rollout_corr/rollout_is_std:0.017005931586027145 - rollout_corr/rollout_is_eff_sample_size:0.9997110010173956 - rollout_corr/rollout_is_seq_mean:0.9999811053276062 - rollout_corr/rollout_is_seq_std:0.0022579727228730917 - rollout_corr/rollout_is_seq_max:1.013187050819397 - rollout_corr/rollout_is_seq_min:0.9912670254707336 - rollout_corr/rollout_is_seq_max_deviation:0.013187050819396973 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.04276253702119) - rollout_corr/training_log_ppl:np.float64(0.039936935934292705) - rollout_corr/kl:np.float64(0.0001362871292077683) - rollout_corr/k3_kl:np.float64(0.0001173268217014873) - rollout_corr/rollout_ppl:np.float64(1.0426045977510512) - rollout_corr/rollout_log_ppl:np.float64(0.0398006483901554) - rollout_corr/log_ppl_diff:np.float64(0.00013628754413730348) - rollout_corr/log_ppl_abs_diff:np.float64(0.0011757521674553573) - rollout_corr/log_ppl_diff_max:np.float64(0.009629026055335999) - rollout_corr/log_ppl_diff_min:np.float64(-0.012113258242607117) - rollout_corr/ppl_ratio:np.float64(1.0001388329546899) - rollout_corr/chi2_token:np.float64(0.00020000548101961613) - rollout_corr/chi2_seq:np.float64(0.026782590430229902) - actor/pg_loss:np.float64(-0.05598723360162694) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.05760814537643455) - srpo/sdpo_loss:np.float64(0.0016209117748076096) - srpo/routed_loss:np.float64(-0.05598723360162694) - srpo/routed_token_count:np.float64(47.18359375) - srpo/correct_sample_fraction:np.float64(0.75) - srpo/sdpo_sample_fraction:np.float64(0.09375) - srpo/grpo_sample_fraction:np.float64(0.90625) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.003746190428500995) - self_distillation/dynamic_weight_mean:np.float64(0.09375000325962901) - self_distillation/dynamic_weight_max:np.float64(1.1569981575012207) - self_distillation/empty_target_batch:np.float64(0.15625) - actor/grad_norm:np.float64(0.11126992106437683) - perf/mfu/actor:np.float64(0.020451989281423506) - perf/max_memory_allocated_gb:np.float64(138.30242013931274) - perf/max_memory_reserved_gb:np.float64(142.8125) - perf/cpu_memory_used_gb:np.float64(149.59305953979492) - actor/lr:np.float64(5e-06) - training/global_step:98 - training/epoch:1 - critic/score/mean:0.75 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.75 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.0044395229779183865 - critic/advantages/max:0.625 - critic/advantages/min:-0.875 - critic/returns/mean:-0.0044395229779183865 - critic/returns/max:0.625 - critic/returns/min:-0.875 - response_length/mean:47.18359375 - response_length/max:121.0 - response_length/min:23.0 - response_length/clip_ratio:0.0 - response_length_no
step:99 - global_seqlen/min:9429 - global_seqlen/max:19950 - global_seqlen/minmax_diff:10521 - global_seqlen/balanced_min:11601 - global_seqlen/balanced_max:19199 - global_seqlen/mean:12574.5 - actor/entropy:0.03956972062587738 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.21881121397018433 - training/rollout_probs_diff_mean:0.0011209510266780853 - training/rollout_probs_diff_std:0.007301102392375469 - training/rollout_actor_probs_pearson_corr:0.9977377653121948 - self_distillation/success_group_fraction:0.90625 - self_distillation/success_sample_fraction:0.89453125 - self_distillation/correct_sample_fraction:0.62109375 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.89453125 - rollout_corr/rollout_is_mean:1.000081181526184 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:9.695559128886089e-05 - rollout_corr/rollout_is_max:1.5607579946517944 - rollout_corr/rollout_is_min:0.19671852886676788 - rollout_corr/rollout_is_std:0.019350359216332436 - rollout_corr/rollout_is_eff_sample_size:0.9996257037716452 - rollout_corr/rollout_is_seq_mean:1.0000147819519043 - rollout_corr/rollout_is_seq_std:0.00336979771964252 - rollout_corr/rollout_is_seq_max:1.015538215637207 - rollout_corr/rollout_is_seq_min:0.986192524433136 - rollout_corr/rollout_is_seq_max_deviation:0.015538215637207031 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.0623197718523443) - rollout_corr/training_log_ppl:np.float64(0.05860299618666431) - rollout_corr/kl:np.float64(0.0003086193539305615) - rollout_corr/k3_kl:np.float64(0.00032341648815276385) - rollout_corr/rollout_ppl:np.float64(1.062003269791603) - rollout_corr/rollout_log_ppl:np.float64(0.05829437606735155) - rollout_corr/log_ppl_diff:np.float64(0.00030862011931276356) - rollout_corr/log_ppl_abs_diff:np.float64(0.0021399275553903863) - rollout_corr/log_ppl_diff_max:np.float64(0.03011687844991684) - rollout_corr/log_ppl_diff_min:np.float64(-0.012398652732372284) - rollout_corr/ppl_ratio:np.float64(1.0003158629406244) - rollout_corr/chi2_token:np.float64(0.0005820468068122864) - rollout_corr/chi2_seq:np.float64(0.06017608195543289) - actor/pg_loss:np.float64(-0.0785270437327199) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.08643104159273207) - srpo/sdpo_loss:np.float64(0.007903997860012169) - srpo/routed_loss:np.float64(-0.0785270437327199) - srpo/routed_token_count:np.float64(80.578125) - srpo/correct_sample_fraction:np.float64(0.62109375) - srpo/sdpo_sample_fraction:np.float64(0.28515625) - srpo/grpo_sample_fraction:np.float64(0.71484375) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.010530682531680213) - self_distillation/dynamic_weight_mean:np.float64(0.28515625232830644) - self_distillation/dynamic_weight_max:np.float64(1.1747715473175049) - self_distillation/empty_target_batch:np.float64(0.10546875) - actor/grad_norm:np.float64(0.3736087381839752) - perf/mfu/actor:np.float64(0.021766593212641704) - perf/max_memory_allocated_gb:np.float64(138.39261770248413) - perf/max_memory_reserved_gb:np.float64(143.0078125) - perf/cpu_memory_used_gb:np.float64(149.70141983032227) - actor/lr:np.float64(5e-06) - training/global_step:99 - training/epoch:1 - critic/score/mean:0.62109375 - critic/score/max:1.0 - critic/score/min:0.0 - critic/rewards/mean:0.62109375 - critic/rewards/max:1.0 - critic/rewards/min:0.0 - critic/advantages/mean:-0.0011816462501883507 - critic/advantages/max:0.875 - critic/advantages/min:-0.875 - critic/returns/mean:-0.0011816462501883507 - critic/returns/max:0.875 - critic/returns/min:-0.875 - response_length/mean:80.578125 - response_length/max:8192.0 - response_length/min:23.0 - re
test_gen_batch meta info: {'eos_token_id': 151645, 'pad_token_id': 151643, 'recompute_log_prob': False, 'do_sample': True, 'validate': True, 'global_steps': 100}
validation generation end
[prompt] system

Given a question and four options, please select the right answer. Respond in the following format:
<reasoning>
...
</reasoning>
<answer>
...
</answer>

For the answer, only output the letter corresponding to the correct option (A, B, C, or D), and nothing else. Do not restate the answer text. For example, if the answer is "A", just output:
<answer>
A
</answer>

user
Which of the following molecules is the only correct reactant used in the synthesis of the product "Nc1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1" ?

A: O=C1c2ccccc2C(=O)N1c1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1
B: O=C1c2ccccc2C(Cl)N1c1ccc2ccc(Cl)nc2n1
C: O=C1c2ccccc2C(=O)N1Cc1ccc(Oc2ccc(Cl)c(C(F)(F)F)c2)cc1
D: O=C1c2ccccc2C(=O)N1CC=C(Cc1ccccc1)C(F)(F)F
Please reason step by step.
assistant
<think>

</think>


[response] <reasoning>
The correct answer is A. The molecule in option A contains the structure "Nc1ccn(Cc2ccc(Cl)cc2C(F)(F)F)n1", which is the correct reactant for the synthesis of the product. The other options do not contain the correct structure. The correct answer is A.
</reasoning>
<answer>
A
</answer>
[ground_truth] A
[score] 1.0
[acc] 1.0
[pred] A
[incorrect_format] 1
[feedback]
local_global_step_folder: /workspace/L2T/checkpoints/datasets/sciknoweval/chemistry/qwen3gen-chemistry-SRPO_TR-Qwen-Qwen3-4B-mbs32-tr0.1-dwtrue-train32-rollout8-lr5e-6-vllm0.8/global_step_100
step:100 - global_seqlen/min:9718 - global_seqlen/max:14844 - global_seqlen/minmax_diff:5126 - global_seqlen/balanced_min:11430 - global_seqlen/balanced_max:11431 - global_seqlen/mean:11430.625 - actor/entropy:0.07664129883050919 - perf/mfu/actor_infer:0 - training/rollout_probs_diff_valid:1 - training/rollout_probs_diff_max:0.41022980213165283 - training/rollout_probs_diff_mean:0.0021808038000017405 - training/rollout_probs_diff_std:0.01022244431078434 - training/rollout_actor_probs_pearson_corr:0.9975420832633972 - self_distillation/success_group_fraction:0.84375 - self_distillation/success_sample_fraction:0.8359375 - self_distillation/correct_sample_fraction:0.6640625 - self_distillation/feedback_available_fraction:0.0 - self_distillation/feedback_used_fraction:0.0 - self_distillation/reprompt_sample_fraction:0.8359375 - rollout_corr/rollout_is_mean:1.0001804828643799 - rollout_corr/rollout_is_ratio_fraction_high:0.0 - rollout_corr/rollout_is_ratio_fraction_low:6.3983621657826e-05 - rollout_corr/rollout_is_max:1.6167027950286865 - rollout_corr/rollout_is_min:0.43883800506591797 - rollout_corr/rollout_is_std:0.027851298451423645 - rollout_corr/rollout_is_eff_sample_size:0.9992250254135918 - rollout_corr/rollout_is_seq_mean:1.0000782012939453 - rollout_corr/rollout_is_seq_std:0.003016919596120715 - rollout_corr/rollout_is_seq_max:1.0158183574676514 - rollout_corr/rollout_is_seq_min:0.9890013337135315 - rollout_corr/rollout_is_seq_max_deviation:0.015818357467651367 - rollout_corr/rollout_is_seq_fraction_high:0.0 - rollout_corr/rollout_is_seq_fraction_low:0.0 - rollout_corr/training_ppl:np.float64(1.081399030983448) - rollout_corr/training_log_ppl:np.float64(0.07520282005225454) - rollout_corr/kl:np.float64(0.000273317242802662) - rollout_corr/k3_kl:np.float64(0.00035141411022021965) - rollout_corr/rollout_ppl:np.float64(1.08108503324911) - rollout_corr/rollout_log_ppl:np.float64(0.07492950143864618) - rollout_corr/log_ppl_diff:np.float64(0.0002733186136083532) - rollout_corr/log_ppl_abs_diff:np.float64(0.001995735538230292) - rollout_corr/log_ppl_diff_max:np.float64(0.012438356876373291) - rollout_corr/log_ppl_diff_min:np.float64(-0.014117404818534851) - rollout_corr/ppl_ratio:np.float64(1.000278169522062) - rollout_corr/chi2_token:np.float64(0.0008595744147896767) - rollout_corr/chi2_seq:np.float64(0.05469026346690953) - actor/pg_loss:np.float64(-0.056587886293527845) - actor/kl_loss:np.float64(0.0) - srpo/grpo_loss:np.float64(-0.06346910382853821) - srpo/sdpo_loss:np.float64(0.006881217535010364) - srpo/routed_loss:np.float64(-0.056587886293527845) - srpo/routed_token_count:np.float64(61.05078125) - srpo/correct_sample_fraction:np.float64(0.6640625) - srpo/sdpo_sample_fraction:np.float64(0.1796875) - srpo/grpo_sample_fraction:np.float64(0.8203125) - srpo/grpo_actor/pg_clipfrac:np.float64(0.0) - srpo/grpo_actor/ppo_kl:np.float64(0.0) - srpo/grpo_actor/pg_clipfrac_lower:np.float64(0.0) - self_distillation/dynamic_weighting_enabled:np.float64(1.0) - self_distillation/dynamic_weight_entropy_mean:np.float64(0.007747210298703067) - self_distillation/dynamic_weight_mean:np.float64(0.17968750419095159) - self_distillation/dynamic_weight_max:np.float64(1.1291440725326538) - self_distillation/empty_target_batch:np.float64(0.1640625) - actor/grad_norm:np.float64(0.2391292154788971) - perf/mfu/actor:np.float64(0.020568645471119033) - perf/max_memory_allocated_gb:np.float64(138.39261770248413) - perf/max_memory_reserved_gb:np.float64(143.0078125) - perf/cpu_memory_used_gb:np.float64(149.58447647094727) - actor/lr:np.float64(5e-06) - val-aux/sciknoweval/reward/mean@16:np.float64(0.7660714285714286) - val-aux/sciknoweval/reward/std@16:np.float64(0.08180825117524429) - val-aux/sciknoweval/reward/best@2/mean:np.float64(0.7995714285714286) - val-aux/sciknoweval/reward/best@2/std:np.float64(0.07120304368358653) - val-aux/sciknoweval/reward/worst@2/mean:np.float64(0.7324333333333333) - val-aux/sciknoweval/reward/worst@2/std:np.float64(0.06710901416436343) - val-aux/sciknoweval/reward/maj@2/mean:np.float64(0.7658571428571429) - val-aux/sci
("Final validation metrics: {'val-aux/sciknoweval/reward/mean@16': "
 "np.float64(0.7660714285714286), 'val-aux/sciknoweval/reward/std@16': "
 "np.float64(0.08180825117524429), 'val-aux/sciknoweval/reward/best@2/mean': "
 "np.float64(0.7995714285714286), 'val-aux/sciknoweval/reward/best@2/std': "
 "np.float64(0.07120304368358653), 'val-aux/sciknoweval/reward/worst@2/mean': "
 "np.float64(0.7324333333333333), 'val-aux/sciknoweval/reward/worst@2/std': "
 "np.float64(0.06710901416436343), 'val-aux/sciknoweval/reward/maj@2/mean': "
 "np.float64(0.7658571428571429), 'val-aux/sciknoweval/reward/maj@2/std': "
 "np.float64(0.08209907766512188), 'val-aux/sciknoweval/reward/best@4/mean': "
 "np.float64(0.828), 'val-aux/sciknoweval/reward/best@4/std': "
 "np.float64(0.05458358345213735), 'val-aux/sciknoweval/reward/worst@4/mean': "
 "np.float64(0.7069809523809524), 'val-aux/sciknoweval/reward/worst@4/std': "
 "np.float64(0.04900531157529151), 'val-aux/sciknoweval/reward/maj@4/mean': "
 "np.float64(0.767504761904762), 'val-aux/sciknoweval/reward/maj@4/std': "
 "np.float64(0.06469916434542414), 'val-aux/sciknoweval/reward/best@8/mean': "
 "np.float64(0.8498714285714286), 'val-aux/sciknoweval/reward/best@8/std': "
 "np.float64(0.03842860858872272), 'val-aux/sciknoweval/reward/worst@8/mean': "
 "np.float64(0.6884666666666667), 'val-aux/sciknoweval/reward/worst@8/std': "
 "np.float64(0.034267210500084945), 'val-aux/sciknoweval/reward/maj@8/mean': "
 "np.float64(0.7679904761904761), 'val-aux/sciknoweval/reward/maj@8/std': "
 "np.float64(0.04659246363692196), 'val-aux/sciknoweval/reward/best@16/mean': "
 "np.float64(0.8647904761904761), 'val-aux/sciknoweval/reward/best@16/std': "
 'np.float64(0.022818983837850748), '
 "'val-aux/sciknoweval/reward/worst@16/mean': np.float64(0.6746142857142856), "
 "'val-aux/sciknoweval/reward/worst@16/std': np.float64(0.022450981651084227), "
 "'val-aux/sciknoweval/reward/maj@16/mean': np.float64(0.7674333333333333), "
 "'val-aux/sciknoweval/reward/maj@16/std': np.float64(0.033146197360321726), "
 "'val-aux/sciknoweval/score/mean@16': np.float64(0.7660714285714286), "
 "'val-aux/sciknoweval/score/std@16': np.float64(0.08180825117524429), "
 "'val-aux/sciknoweval/score/best@2/mean': np.float64(0.7995714285714286), "
 "'val-aux/sciknoweval/score/best@2/std': np.float64(0.07120304368358653), "
 "'val-aux/sciknoweval/score/worst@2/mean': np.float64(0.7324333333333333), "
 "'val-aux/sciknoweval/score/worst@2/std': np.float64(0.06710901416436343), "
 "'val-aux/sciknoweval/score/maj@2/mean': np.float64(0.7658571428571429), "
 "'val-aux/sciknoweval/score/maj@2/std': np.float64(0.08209907766512188), "
 "'val-aux/sciknoweval/score/best@4/mean': np.float64(0.828), "
 "'val-aux/sciknoweval/score/best@4/std': np.float64(0.05458358345213735), "
 "'val-aux/sciknoweval/score/worst@4/mean': np.float64(0.7069809523809524), "
 "'val-aux/sciknoweval/score/worst@4/std': np.float64(0.04900531157529151), "
 "'val-aux/sciknoweval/score/maj@4/mean': np.float64(0.767504761904762), "
 "'val-aux/sciknoweval/score/maj@4/std': np.float64(0.06469916434542414), "
 "'val-aux/sciknoweval/score/best@8/mean': np.float64(0.8498714285714286), "
 "'val-aux/sciknoweval/score/best@8/std': np.float64(0.03842860858872272), "
 "'val-aux/sciknoweval/score/worst@8/mean': np.float64(0.6884666666666667), "
 "'val-aux/sciknoweval/score/worst@8/std': np.float64(0.034267210500084945), "
 "'val-aux/sciknoweval/score/maj@8/mean': np.float64(0.7679904761904761), "
 "'val-aux/sciknoweval/score/maj@8/std': np.float64(0.04659246363692196), "
 "'val-aux/sciknoweval/score/best@16/mean': np.float64(0.8647904761904761), "
 "'val-aux/sciknoweval/score/best@16/std': np.float64(0.022818983837850748), "
 "'val-aux/sciknoweval/score/worst@16/mean': np.float64(0.6746142857142856), "
 "'val-aux/sciknoweval/score/worst@16/std': np.float64(0.022450981651084227), "
 "'val-aux/sciknoweval/score/maj@16/mean': np.float64(0.7674333333333333), "
 "'val-aux/sciknoweval/score/maj@16/std': np.float64(0.033146197360321726), "
 "'val-core/sciknoweval/acc/mean@16': np.float64(0.7660714285714286), "
 "'val-aux/sciknoweval/acc/std@16': np.float64(0.08180825117524429), "
 "'val-aux/sciknoweval/acc/best@2/mean': np.float64(0.7995714285714286), "
 "'val-aux/sciknoweval/acc/best@2/std': np.float64(0.07120304368358653), "
 "'val-aux/sciknoweval/acc/worst@2/mean': np.float64(0.7324333333333333), "
 "'val-aux/sciknoweval/acc/worst@2/std': np.float64(0.06710901416436343), "
 "'val-aux/sciknoweval/acc/maj@2/mean': np.float64(0.7658571428571429), "
 "'val-aux/sciknoweval/acc/maj@2/std': np.float64(0.08209907766512188), "
 "'val-aux/sciknoweval/acc/best@4/mean': np.float64(0.828), "
 "'val-aux/sciknoweval/acc/best@4/std': np.float64(0.05458358345213735), "
 "'val-aux/sciknoweval/acc/worst@4/mean': np.float64(0.7069809523809524), "
 "'val-aux/sciknoweval/acc/worst@4/std': np.float64(0.04900531157529151), "
 "'val-aux/sciknoweval/acc/maj@4/mean': np.float64(0.767504761904762), "
 "'val-aux/sciknoweval/acc/maj@4/std': np.float64(0.06469916434542414), "
 "'val-aux/sciknoweval/acc/best@8/mean': np.float64(0.8498714285714286), "
 "'val-aux/sciknoweval/acc/best@8/std': np.float64(0.03842860858872272), "
 "'val-aux/sciknoweval/acc/worst@8/mean': np.float64(0.6884666666666667), "
 "'val-aux/sciknoweval/acc/worst@8/std': np.float64(0.034267210500084945), "
 "'val-aux/sciknoweval/acc/maj@8/mean': np.float64(0.7679904761904761), "
 "'val-aux/sciknoweval/acc/maj@8/std': np.float64(0.04659246363692196), "
 "'val-core/sciknoweval/acc/best@16/mean': np.float64(0.8647904761904761), "
 "'val-core/sciknoweval/acc/best@16/std': np.float64(0.022818983837850748), "
 "'val-aux/sciknoweval/acc/worst@16/mean': np.float64(0.6746142857142856), "
 "'val-aux/sciknoweval/acc/worst@16/std': np.float64(0.022450981651084227), "
 "'val-core/sciknoweval/acc/maj@16/mean': np.float64(0.7674333333333333), "
 "'val-core/sciknoweval/acc/maj@16/std': np.float64(0.033146197360321726), "
 "'val-aux/sciknoweval/incorrect_format/mean@16': "
 'np.float64(0.9982142857142857), '
 "'val-aux/sciknoweval/incorrect_format/std@16': "
 'np.float64(0.005316651201494241), '
 "'val-aux/sciknoweval/incorrect_format/best@2/mean': "
 'np.float64(0.9998142857142858), '
 "'val-aux/sciknoweval/incorrect_format/best@2/std': "
 'np.float64(0.0016347712194112146), '
 "'val-aux/sciknoweval/incorrect_format/worst@2/mean': "
 'np.float64(0.9967047619047619), '
 "'val-aux/sciknoweval/incorrect_format/worst@2/std': "
 'np.float64(0.006964443486271992), '
 "'val-aux/sciknoweval/incorrect_format/maj@2/mean': "
 'np.float64(0.9983523809523809), '
 "'val-aux/sciknoweval/incorrect_format/maj@2/std': "
 'np.float64(0.005210571230613174), '
 "'val-aux/sciknoweval/incorrect_format/best@4/mean': "
 'np.float64(0.9999904761904761), '
 "'val-aux/sciknoweval/incorrect_format/best@4/std': "
 'np.float64(0.0002127457895589398), '
 "'val-aux/sciknoweval/incorrect_format/worst@4/mean': "
 'np.float64(0.9940714285714285), '
 "'val-aux/sciknoweval/incorrect_format/worst@4/std': "
 'np.float64(0.008456497157340367), '
 "'val-aux/sciknoweval/incorrect_format/maj@4/mean': "
 'np.float64(0.9994285714285714), '
 "'val-aux/sciknoweval/incorrect_format/maj@4/std': "
 'np.float64(0.0029882373225892547), '
 "'val-aux/sciknoweval/incorrect_format/best@8/mean': np.float64(1.0), "
 "'val-aux/sciknoweval/incorrect_format/best@8/std': np.float64(0.0), "
 "'val-aux/sciknoweval/incorrect_format/worst@8/mean': "
 'np.float64(0.9903190476190475), '
 "'val-aux/sciknoweval/incorrect_format/worst@8/std': "
 'np.float64(0.008910847165478613), '
 "'val-aux/sciknoweval/incorrect_format/maj@8/mean': "
 'np.float64(0.999804761904762), '
 "'val-aux/sciknoweval/incorrect_format/maj@8/std': "
 'np.float64(0.0015690976922436521), '
 "'val-aux/sciknoweval/incorrect_format/best@16/mean': np.float64(1.0), "
 "'val-aux/sciknoweval/incorrect_format/best@16/std': np.float64(0.0), "
 "'val-aux/sciknoweval/incorrect_format/worst@16/mean': "
 'np.float64(0.9860095238095239), '
 "'val-aux/sciknoweval/incorrect_format/worst@16/std': "
 'np.float64(0.00755187697793512), '
 "'val-aux/sciknoweval/incorrect_format/maj@16/mean': "
 'np.float64(0.9999857142857144), '
 "'val-aux/sciknoweval/incorrect_format/maj@16/std': "
 "np.float64(0.0002604287412166727), 'val-aux/num_turns/min': np.int32(2), "
 "'val-aux/num_turns/max': np.int32(2), 'val-aux/num_turns/mean': "
 'np.float64(2.0)}')