{ "plain": { "rollouts": [ { "rollout": 0, "rollout/raw_reward_unfiltered": 0.62890625, "rollout/response_len/mean": 739.92578125, "rollout/truncated_ratio": 0.3203125, "rollout/prefix_cache_hit_rate": 0.24518696778970753, "perf/rollout_time": 61.951833724975586, "perf/tokens_per_gpu_per_sec": 3057.5527568869334, "train/train_rollout_logprob_abs_diff": 0.015336457639932632, "train/train_rollout_kl": 0.0006840902497060597, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5216571092605591 }, { "rollout": 1, "rollout/raw_reward_unfiltered": 0.59375, "rollout/response_len/mean": 768.90625, "rollout/truncated_ratio": 0.40234375, "rollout/prefix_cache_hit_rate": 0.20739650249821556, "perf/rollout_time": 65.47558307647705, "perf/tokens_per_gpu_per_sec": 3006.3115248639506, "train/train_rollout_logprob_abs_diff": 0.01610221341252327, "train/train_rollout_kl": 0.0007360127056017518, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5375333428382874 }, { "rollout": 2, "rollout/raw_reward_unfiltered": 0.7578125, "rollout/response_len/mean": 707.6875, "rollout/truncated_ratio": 0.234375, "rollout/prefix_cache_hit_rate": 0.24773878437047755, "perf/rollout_time": 57.12712788581848, "perf/tokens_per_gpu_per_sec": 3171.312941937241, "train/train_rollout_logprob_abs_diff": 0.014312651939690113, "train/train_rollout_kl": 0.0006478244322352111, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.3654817044734955 }, { "rollout": 3, "rollout/raw_reward_unfiltered": 0.69140625, "rollout/response_len/mean": 683.7265625, "rollout/truncated_ratio": 0.2890625, "rollout/prefix_cache_hit_rate": 0.23570526695526695, "perf/rollout_time": 56.08247256278992, "perf/tokens_per_gpu_per_sec": 3121.0107543677213, "train/train_rollout_logprob_abs_diff": 0.015629252418875694, "train/train_rollout_kl": 0.0007079748320393264, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.42627525329589844 }, { "rollout": 4, "rollout/raw_reward_unfiltered": 0.58984375, "rollout/response_len/mean": 710.7578125, "rollout/truncated_ratio": 0.3046875, "rollout/prefix_cache_hit_rate": 0.20591985428051002, "perf/rollout_time": 58.59470510482788, "perf/tokens_per_gpu_per_sec": 3105.2976488998147, "train/train_rollout_logprob_abs_diff": 0.015349497087299824, "train/train_rollout_kl": 0.000685779785271734, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.420428991317749 }, { "rollout": 5, "rollout/raw_reward_unfiltered": 0.5859375, "rollout/response_len/mean": 711.59375, "rollout/truncated_ratio": 0.35546875, "rollout/prefix_cache_hit_rate": 0.20338674435542609, "perf/rollout_time": 59.59472179412842, "perf/tokens_per_gpu_per_sec": 3056.78077715178, "train/train_rollout_logprob_abs_diff": 0.01628461480140686, "train/train_rollout_kl": 0.0007354679401032627, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4396167993545532 }, { "rollout": 6, "rollout/raw_reward_unfiltered": 0.6875, "rollout/response_len/mean": 646.3046875, "rollout/truncated_ratio": 0.21875, "rollout/prefix_cache_hit_rate": 0.24956995412844038, "perf/rollout_time": 51.69342064857483, "perf/tokens_per_gpu_per_sec": 3200.6781119167727, "train/train_rollout_logprob_abs_diff": 0.014715931378304958, "train/train_rollout_kl": 0.0006737026269547641, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.3854154348373413 }, { "rollout": 7, "rollout/raw_reward_unfiltered": 0.67578125, "rollout/response_len/mean": 615.28515625, "rollout/truncated_ratio": 0.23828125, "rollout/prefix_cache_hit_rate": 0.22218610698365526, "perf/rollout_time": 49.27047061920166, "perf/tokens_per_gpu_per_sec": 3196.90471839362, "train/train_rollout_logprob_abs_diff": 0.015263394452631474, "train/train_rollout_kl": 0.000698999734595418, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.429172158241272 }, { "rollout": 8, "rollout/raw_reward_unfiltered": 0.74609375, "rollout/response_len/mean": 595.91015625, "rollout/truncated_ratio": 0.171875, "rollout/prefix_cache_hit_rate": 0.2520673076923077, "perf/rollout_time": 46.150370836257935, "perf/tokens_per_gpu_per_sec": 3305.563904161461, "train/train_rollout_logprob_abs_diff": 0.01513341348618269, "train/train_rollout_kl": 0.0007040104828774929, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.40924978256225586 }, { "rollout": 9, "rollout/raw_reward_unfiltered": 0.6484375, "rollout/response_len/mean": 695.35546875, "rollout/truncated_ratio": 0.31640625, "rollout/prefix_cache_hit_rate": 0.22655444864307797, "perf/rollout_time": 57.43663716316223, "perf/tokens_per_gpu_per_sec": 3099.258744802869, "train/train_rollout_logprob_abs_diff": 0.016122153028845787, "train/train_rollout_kl": 0.0007552009192295372, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5382251143455505 }, { "rollout": 10, "rollout/raw_reward_unfiltered": 0.65625, "rollout/response_len/mean": 691.3203125, "rollout/truncated_ratio": 0.31640625, "rollout/prefix_cache_hit_rate": 0.24109100877192982, "perf/rollout_time": 57.25109601020813, "perf/tokens_per_gpu_per_sec": 3091.2595973436737, "train/train_rollout_logprob_abs_diff": 0.015402797609567642, "train/train_rollout_kl": 0.0007068103877827525, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5083921551704407 }, { "rollout": 11, "rollout/raw_reward_unfiltered": 0.6640625, "rollout/response_len/mean": 632.4921875, "rollout/truncated_ratio": 0.2109375, "rollout/prefix_cache_hit_rate": 0.2019092219020173, "perf/rollout_time": 51.1147837638855, "perf/tokens_per_gpu_per_sec": 3167.7332481332164, "train/train_rollout_logprob_abs_diff": 0.015569799579679966, "train/train_rollout_kl": 0.000722734082955867, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5278587341308594 } ], "timers_s": { "actor_train": { "median": 80.25, "max": 87.4, "n": 12 }, "finalize_and_resume_engines": { "median": 0.0, "max": 0.0, "n": 12 }, "log_probs": { "median": 25.15, "max": 31.5, "n": 12 }, "sleep": { "median": 0.8, "max": 4.0, "n": 12 }, "train": { "median": 105.5, "max": 119.0, "n": 12 }, "train_wait": { "median": 61.75, "max": 114.3, "n": 12 }, "update_weights": { "median": 0.65, "max": 0.7, "n": 12 }, "update_weights_implementation": { "median": 0.2, "max": 0.3, "n": 12 }, "wake_up": { "median": 0.9, "max": 0.9, "n": 12 } } }, "true-on-policy": { "rollouts": [], "timers_s": {} }, "true-on-policy-fa3": { "rollouts": [], "timers_s": { "finalize_and_resume_engines": { "median": 0.0, "max": 0.0, "n": 1 }, "sleep": { "median": 2.1, "max": 2.1, "n": 1 }, "update_weights": { "median": 0.4, "max": 0.4, "n": 1 }, "update_weights_implementation": { "median": 0.1, "max": 0.1, "n": 1 } } }, "plain-48": { "rollouts": [ { "rollout": 0, "rollout/raw_reward_unfiltered": 0.58203125, "rollout/response_len/mean": 759.08984375, "rollout/truncated_ratio": 0.34765625, "rollout/prefix_cache_hit_rate": 0.2423176601258793, "perf/rollout_time": 63.933146953582764, "perf/tokens_per_gpu_per_sec": 3039.534408357636, "train/train_rollout_logprob_abs_diff": 0.015669148415327072, "train/train_rollout_kl": 0.0007089389255270362, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5211054682731628 }, { "rollout": 1, "rollout/raw_reward_unfiltered": 0.60546875, "rollout/response_len/mean": 771.06640625, "rollout/truncated_ratio": 0.43359375, "rollout/prefix_cache_hit_rate": 0.23166488222698073, "perf/rollout_time": 65.76358842849731, "perf/tokens_per_gpu_per_sec": 3001.554579318907, "train/train_rollout_logprob_abs_diff": 0.01593092456459999, "train/train_rollout_kl": 0.0007229857728816569, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.43598222732543945 }, { "rollout": 2, "rollout/raw_reward_unfiltered": 0.7265625, "rollout/response_len/mean": 727.77734375, "rollout/truncated_ratio": 0.23828125, "rollout/prefix_cache_hit_rate": 0.1978111432706223, "perf/rollout_time": 59.2400918006897, "perf/tokens_per_gpu_per_sec": 3145.015382940897, "train/train_rollout_logprob_abs_diff": 0.014127157628536224, "train/train_rollout_kl": 0.0006314609199762344, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.3464278280735016 }, { "rollout": 3, "rollout/raw_reward_unfiltered": 0.65625, "rollout/response_len/mean": 709.69921875, "rollout/truncated_ratio": 0.36328125, "rollout/prefix_cache_hit_rate": 0.202245670995671, "perf/rollout_time": 59.173675775527954, "perf/tokens_per_gpu_per_sec": 3070.3348679775168, "train/train_rollout_logprob_abs_diff": 0.015861723572015762, "train/train_rollout_kl": 0.0007229630136862397, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.41638296842575073 }, { "rollout": 4, "rollout/raw_reward_unfiltered": 0.5859375, "rollout/response_len/mean": 718.2265625, "rollout/truncated_ratio": 0.30859375, "rollout/prefix_cache_hit_rate": 0.24298724954462658, "perf/rollout_time": 58.97963237762451, "perf/tokens_per_gpu_per_sec": 3117.449068227059, "train/train_rollout_logprob_abs_diff": 0.014987394213676453, "train/train_rollout_kl": 0.0006829750491306186, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4993102252483368 }, { "rollout": 5, "rollout/raw_reward_unfiltered": 0.55859375, "rollout/response_len/mean": 770.828125, "rollout/truncated_ratio": 0.41015625, "rollout/prefix_cache_hit_rate": 0.19883466860888566, "perf/rollout_time": 65.65153884887695, "perf/tokens_per_gpu_per_sec": 3005.7482803904695, "train/train_rollout_logprob_abs_diff": 0.016160542145371437, "train/train_rollout_kl": 0.0007167896837927401, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.40144261717796326 }, { "rollout": 6, "rollout/raw_reward_unfiltered": 0.69921875, "rollout/response_len/mean": 673.1796875, "rollout/truncated_ratio": 0.2421875, "rollout/prefix_cache_hit_rate": 0.24746750764525993, "perf/rollout_time": 54.31332349777222, "perf/tokens_per_gpu_per_sec": 3172.9599461367643, "train/train_rollout_logprob_abs_diff": 0.014528200961649418, "train/train_rollout_kl": 0.0006798460963182151, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4285953938961029 }, { "rollout": 7, "rollout/raw_reward_unfiltered": 0.67578125, "rollout/response_len/mean": 663.15625, "rollout/truncated_ratio": 0.265625, "rollout/prefix_cache_hit_rate": 0.24312778603268945, "perf/rollout_time": 54.35955619812012, "perf/tokens_per_gpu_per_sec": 3123.0571379438707, "train/train_rollout_logprob_abs_diff": 0.01530056819319725, "train/train_rollout_kl": 0.0007084502140060067, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.51386559009552 }, { "rollout": 8, "rollout/raw_reward_unfiltered": 0.734375, "rollout/response_len/mean": 626.171875, "rollout/truncated_ratio": 0.1875, "rollout/prefix_cache_hit_rate": 0.21557692307692308, "perf/rollout_time": 49.095505475997925, "perf/tokens_per_gpu_per_sec": 3265.0646621485203, "train/train_rollout_logprob_abs_diff": 0.014684163965284824, "train/train_rollout_kl": 0.0006792584317736328, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.42341604828834534 }, { "rollout": 9, "rollout/raw_reward_unfiltered": 0.63671875, "rollout/response_len/mean": 750.3125, "rollout/truncated_ratio": 0.359375, "rollout/prefix_cache_hit_rate": 0.22560975609756098, "perf/rollout_time": 62.93436813354492, "perf/tokens_per_gpu_per_sec": 3052.068459516615, "train/train_rollout_logprob_abs_diff": 0.015497742220759392, "train/train_rollout_kl": 0.0006951651303097606, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.42883360385894775 }, { "rollout": 10, "rollout/raw_reward_unfiltered": 0.58984375, "rollout/response_len/mean": 742.8515625, "rollout/truncated_ratio": 0.39453125, "rollout/prefix_cache_hit_rate": 0.20317068713450293, "perf/rollout_time": 62.600292682647705, "perf/tokens_per_gpu_per_sec": 3037.8452216520323, "train/train_rollout_logprob_abs_diff": 0.015224173665046692, "train/train_rollout_kl": 0.0006856130203232169, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.46757036447525024 }, { "rollout": 11, "rollout/raw_reward_unfiltered": 0.65234375, "rollout/response_len/mean": 685.140625, "rollout/truncated_ratio": 0.27734375, "rollout/prefix_cache_hit_rate": 0.24081412103746397, "perf/rollout_time": 56.547826051712036, "perf/tokens_per_gpu_per_sec": 3101.7284349641186, "train/train_rollout_logprob_abs_diff": 0.015468996949493885, "train/train_rollout_kl": 0.0007174779311753809, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.45263758301734924 }, { "rollout": 12, "rollout/raw_reward_unfiltered": 0.703125, "rollout/response_len/mean": 661.1171875, "rollout/truncated_ratio": 0.19921875, "rollout/prefix_cache_hit_rate": 0.23867786705624544, "perf/rollout_time": 52.82849717140198, "perf/tokens_per_gpu_per_sec": 3203.687575114651, "train/train_rollout_logprob_abs_diff": 0.015438497997820377, "train/train_rollout_kl": 0.0007292819209396839, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4147093594074249 }, { "rollout": 13, "rollout/raw_reward_unfiltered": 0.7265625, "rollout/response_len/mean": 660.94140625, "rollout/truncated_ratio": 0.22265625, "rollout/prefix_cache_hit_rate": 0.2537686160552125, "perf/rollout_time": 53.621577501297, "perf/tokens_per_gpu_per_sec": 3155.464793923815, "train/train_rollout_logprob_abs_diff": 0.014749601483345032, "train/train_rollout_kl": 0.0006858782144263387, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.45241424441337585 }, { "rollout": 14, "rollout/raw_reward_unfiltered": 0.57421875, "rollout/response_len/mean": 726.29296875, "rollout/truncated_ratio": 0.34765625, "rollout/prefix_cache_hit_rate": 0.25854627071823205, "perf/rollout_time": 61.09412479400635, "perf/tokens_per_gpu_per_sec": 3043.353196840309, "train/train_rollout_logprob_abs_diff": 0.016038861125707626, "train/train_rollout_kl": 0.0007261380087584257, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4682966470718384 }, { "rollout": 15, "rollout/raw_reward_unfiltered": 0.609375, "rollout/response_len/mean": 703.61328125, "rollout/truncated_ratio": 0.2734375, "rollout/prefix_cache_hit_rate": 0.2397160538769567, "perf/rollout_time": 57.86472249031067, "perf/tokens_per_gpu_per_sec": 3112.8638010864315, "train/train_rollout_logprob_abs_diff": 0.015107162296772003, "train/train_rollout_kl": 0.0006795918452553451, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5199260711669922 }, { "rollout": 16, "rollout/raw_reward_unfiltered": 0.68359375, "rollout/response_len/mean": 734.9296875, "rollout/truncated_ratio": 0.30078125, "rollout/prefix_cache_hit_rate": 0.18760899895361005, "perf/rollout_time": 61.46616291999817, "perf/tokens_per_gpu_per_sec": 3060.903610412088, "train/train_rollout_logprob_abs_diff": 0.014878937043249607, "train/train_rollout_kl": 0.0006919661536812782, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.48708242177963257 }, { "rollout": 17, "rollout/raw_reward_unfiltered": 0.75390625, "rollout/response_len/mean": 576.9453125, "rollout/truncated_ratio": 0.15234375, "rollout/prefix_cache_hit_rate": 0.24171852122986823, "perf/rollout_time": 45.04358148574829, "perf/tokens_per_gpu_per_sec": 3279.0021381122056, "train/train_rollout_logprob_abs_diff": 0.015163729898631573, "train/train_rollout_kl": 0.0007196383085101843, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 4.950849533081055 }, { "rollout": 18, "rollout/raw_reward_unfiltered": 0.671875, "rollout/response_len/mean": 637.9140625, "rollout/truncated_ratio": 0.2265625, "rollout/prefix_cache_hit_rate": 0.23619469026548673, "perf/rollout_time": 51.04841208457947, "perf/tokens_per_gpu_per_sec": 3199.041720032873, "train/train_rollout_logprob_abs_diff": 0.015150027349591255, "train/train_rollout_kl": 0.0006801282870583236, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5013094544410706 }, { "rollout": 19, "rollout/raw_reward_unfiltered": 0.6484375, "rollout/response_len/mean": 652.97265625, "rollout/truncated_ratio": 0.3046875, "rollout/prefix_cache_hit_rate": 0.23965669014084506, "perf/rollout_time": 53.80374813079834, "perf/tokens_per_gpu_per_sec": 3106.8653357313915, "train/train_rollout_logprob_abs_diff": 0.015863237902522087, "train/train_rollout_kl": 0.0007217709207907319, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.42389675974845886 }, { "rollout": 20, "rollout/raw_reward_unfiltered": 0.7109375, "rollout/response_len/mean": 632.65234375, "rollout/truncated_ratio": 0.20703125, "rollout/prefix_cache_hit_rate": 0.226882373351839, "perf/rollout_time": 50.64891338348389, "perf/tokens_per_gpu_per_sec": 3197.6796574832983, "train/train_rollout_logprob_abs_diff": 0.015775462612509727, "train/train_rollout_kl": 0.000748142774682492, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4796660542488098 }, { "rollout": 21, "rollout/raw_reward_unfiltered": 0.71484375, "rollout/response_len/mean": 625.30078125, "rollout/truncated_ratio": 0.2421875, "rollout/prefix_cache_hit_rate": 0.18811336717428087, "perf/rollout_time": 50.376752853393555, "perf/tokens_per_gpu_per_sec": 3177.5966280688267, "train/train_rollout_logprob_abs_diff": 0.015374618582427502, "train/train_rollout_kl": 0.000709490617737174, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.43412938714027405 }, { "rollout": 22, "rollout/raw_reward_unfiltered": 0.65625, "rollout/response_len/mean": 619.24609375, "rollout/truncated_ratio": 0.21875, "rollout/prefix_cache_hit_rate": 0.2128212396069539, "perf/rollout_time": 49.62475562095642, "perf/tokens_per_gpu_per_sec": 3194.5144719877353, "train/train_rollout_logprob_abs_diff": 0.015395589172840118, "train/train_rollout_kl": 0.0007065099198371172, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5716469287872314 }, { "rollout": 23, "rollout/raw_reward_unfiltered": 0.6484375, "rollout/response_len/mean": 618.3515625, "rollout/truncated_ratio": 0.23828125, "rollout/prefix_cache_hit_rate": 0.2579686910599774, "perf/rollout_time": 49.85119986534119, "perf/tokens_per_gpu_per_sec": 3175.4100287976407, "train/train_rollout_logprob_abs_diff": 0.01520753838121891, "train/train_rollout_kl": 0.0006942453910596669, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.3798649311065674 }, { "rollout": 24, "rollout/raw_reward_unfiltered": 0.75390625, "rollout/response_len/mean": 563.6015625, "rollout/truncated_ratio": 0.15625, "rollout/prefix_cache_hit_rate": 0.22847682119205298, "perf/rollout_time": 43.23245334625244, "perf/tokens_per_gpu_per_sec": 3337.3539744421405, "train/train_rollout_logprob_abs_diff": 0.014738596975803375, "train/train_rollout_kl": 0.0006925351335667074, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5025835633277893 }, { "rollout": 25, "rollout/raw_reward_unfiltered": 0.7265625, "rollout/response_len/mean": 545.9453125, "rollout/truncated_ratio": 0.13671875, "rollout/prefix_cache_hit_rate": 0.2443096313017306, "perf/rollout_time": 41.20708894729614, "perf/tokens_per_gpu_per_sec": 3391.69797164647, "train/train_rollout_logprob_abs_diff": 0.015290481969714165, "train/train_rollout_kl": 0.0007087115081958473, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5568088889122009 }, { "rollout": 26, "rollout/raw_reward_unfiltered": 0.6640625, "rollout/response_len/mean": 569.09375, "rollout/truncated_ratio": 0.171875, "rollout/prefix_cache_hit_rate": 0.2284603285564488, "perf/rollout_time": 44.25398302078247, "perf/tokens_per_gpu_per_sec": 3292.087854139192, "train/train_rollout_logprob_abs_diff": 0.015299706719815731, "train/train_rollout_kl": 0.000722345954272896, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5254616141319275 }, { "rollout": 27, "rollout/raw_reward_unfiltered": 0.72265625, "rollout/response_len/mean": 500.86328125, "rollout/truncated_ratio": 0.11328125, "rollout/prefix_cache_hit_rate": 0.2365811499272198, "perf/rollout_time": 37.74953556060791, "perf/tokens_per_gpu_per_sec": 3396.624570231803, "train/train_rollout_logprob_abs_diff": 0.01499457098543644, "train/train_rollout_kl": 0.0006976217264309525, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4675930142402649 }, { "rollout": 28, "rollout/raw_reward_unfiltered": 0.70703125, "rollout/response_len/mean": 574.51953125, "rollout/truncated_ratio": 0.17578125, "rollout/prefix_cache_hit_rate": 0.2143697935530605, "perf/rollout_time": 44.872830867767334, "perf/tokens_per_gpu_per_sec": 3277.64032613434, "train/train_rollout_logprob_abs_diff": 0.01564876176416874, "train/train_rollout_kl": 0.000717950111720711, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5080634355545044 }, { "rollout": 29, "rollout/raw_reward_unfiltered": 0.765625, "rollout/response_len/mean": 471.33203125, "rollout/truncated_ratio": 0.0703125, "rollout/prefix_cache_hit_rate": 0.19373691556175854, "perf/rollout_time": 35.005818605422974, "perf/tokens_per_gpu_per_sec": 3446.884112611714, "train/train_rollout_logprob_abs_diff": 0.013833332806825638, "train/train_rollout_kl": 0.0006709957960993052, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.46245670318603516 }, { "rollout": 30, "rollout/raw_reward_unfiltered": 0.80078125, "rollout/response_len/mean": 526.01953125, "rollout/truncated_ratio": 0.12890625, "rollout/prefix_cache_hit_rate": 0.20605946556266466, "perf/rollout_time": 40.34177255630493, "perf/tokens_per_gpu_per_sec": 3338.0040456093966, "train/train_rollout_logprob_abs_diff": 0.014549431391060352, "train/train_rollout_kl": 0.000702361052390188, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.47933024168014526 }, { "rollout": 31, "rollout/raw_reward_unfiltered": 0.59375, "rollout/response_len/mean": 628.3515625, "rollout/truncated_ratio": 0.21875, "rollout/prefix_cache_hit_rate": 0.18922345483359745, "perf/rollout_time": 50.3379909992218, "perf/tokens_per_gpu_per_sec": 3195.5585991202306, "train/train_rollout_logprob_abs_diff": 0.015586714260280132, "train/train_rollout_kl": 0.0007328960346058011, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.49036097526550293 }, { "rollout": 32, "rollout/raw_reward_unfiltered": 0.71484375, "rollout/response_len/mean": 581.1796875, "rollout/truncated_ratio": 0.1796875, "rollout/prefix_cache_hit_rate": 0.1664617276360283, "perf/rollout_time": 45.89498496055603, "perf/tokens_per_gpu_per_sec": 3241.792107086845, "train/train_rollout_logprob_abs_diff": 0.014495952054858208, "train/train_rollout_kl": 0.0006580839981324971, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5358617901802063 }, { "rollout": 33, "rollout/raw_reward_unfiltered": 0.70703125, "rollout/response_len/mean": 519.8671875, "rollout/truncated_ratio": 0.12890625, "rollout/prefix_cache_hit_rate": 0.23064663023679416, "perf/rollout_time": 39.56239366531372, "perf/tokens_per_gpu_per_sec": 3363.952169473582, "train/train_rollout_logprob_abs_diff": 0.015362434089183807, "train/train_rollout_kl": 0.0006999865872785449, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.589840292930603 }, { "rollout": 34, "rollout/raw_reward_unfiltered": 0.67578125, "rollout/response_len/mean": 500.3046875, "rollout/truncated_ratio": 0.1171875, "rollout/prefix_cache_hit_rate": 0.23051146384479718, "perf/rollout_time": 37.79030656814575, "perf/tokens_per_gpu_per_sec": 3389.1759985868875, "train/train_rollout_logprob_abs_diff": 0.014537480659782887, "train/train_rollout_kl": 0.0006739323143847287, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5997372269630432 }, { "rollout": 35, "rollout/raw_reward_unfiltered": 0.7578125, "rollout/response_len/mean": 525.79296875, "rollout/truncated_ratio": 0.1171875, "rollout/prefix_cache_hit_rate": 0.22375856164383562, "perf/rollout_time": 39.78631353378296, "perf/tokens_per_gpu_per_sec": 3383.148325257811, "train/train_rollout_logprob_abs_diff": 0.015031552873551846, "train/train_rollout_kl": 0.0007121246890164912, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.6063504219055176 }, { "rollout": 36, "rollout/raw_reward_unfiltered": 0.87109375, "rollout/response_len/mean": 396.3125, "rollout/truncated_ratio": 0.015625, "rollout/prefix_cache_hit_rate": 0.23825435398946943, "perf/rollout_time": 27.54421329498291, "perf/tokens_per_gpu_per_sec": 3683.3871025272624, "train/train_rollout_logprob_abs_diff": 0.014479887671768665, "train/train_rollout_kl": 0.0006955971475690603, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5015864372253418 }, { "rollout": 37, "rollout/raw_reward_unfiltered": 0.67578125, "rollout/response_len/mean": 539.265625, "rollout/truncated_ratio": 0.125, "rollout/prefix_cache_hit_rate": 0.23913621851720468, "perf/rollout_time": 40.821311235427856, "perf/tokens_per_gpu_per_sec": 3381.8609893203998, "train/train_rollout_logprob_abs_diff": 0.01507849432528019, "train/train_rollout_kl": 0.0006932957330718637, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4862067401409149 }, { "rollout": 38, "rollout/raw_reward_unfiltered": 0.6796875, "rollout/response_len/mean": 478.72265625, "rollout/truncated_ratio": 0.12890625, "rollout/prefix_cache_hit_rate": 0.24073741007194244, "perf/rollout_time": 36.48499608039856, "perf/tokens_per_gpu_per_sec": 3358.997208878451, "train/train_rollout_logprob_abs_diff": 0.014453672803938389, "train/train_rollout_kl": 0.0006773276254534721, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5477573871612549 }, { "rollout": 39, "rollout/raw_reward_unfiltered": 0.8203125, "rollout/response_len/mean": 499.27734375, "rollout/truncated_ratio": 0.09375, "rollout/prefix_cache_hit_rate": 0.21176312005751258, "perf/rollout_time": 37.675957441329956, "perf/tokens_per_gpu_per_sec": 3392.481802195394, "train/train_rollout_logprob_abs_diff": 0.014326857402920723, "train/train_rollout_kl": 0.0006868948112241924, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.6488147974014282 }, { "rollout": 40, "rollout/raw_reward_unfiltered": 0.76953125, "rollout/response_len/mean": 436.6796875, "rollout/truncated_ratio": 0.0703125, "rollout/prefix_cache_hit_rate": 0.24143145161290322, "perf/rollout_time": 31.195338010787964, "perf/tokens_per_gpu_per_sec": 3583.5482840846544, "train/train_rollout_logprob_abs_diff": 0.014349369332194328, "train/train_rollout_kl": 0.0006853842060081661, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5389475226402283 }, { "rollout": 41, "rollout/raw_reward_unfiltered": 0.77734375, "rollout/response_len/mean": 450.6015625, "rollout/truncated_ratio": 0.0625, "rollout/prefix_cache_hit_rate": 0.20229915433403806, "perf/rollout_time": 32.97510242462158, "perf/tokens_per_gpu_per_sec": 3498.2150628247455, "train/train_rollout_logprob_abs_diff": 0.014434712938964367, "train/train_rollout_kl": 0.0007003873470239341, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.6502324938774109 }, { "rollout": 42, "rollout/raw_reward_unfiltered": 0.79296875, "rollout/response_len/mean": 472.62109375, "rollout/truncated_ratio": 0.08984375, "rollout/prefix_cache_hit_rate": 0.20496258018531718, "perf/rollout_time": 35.311248540878296, "perf/tokens_per_gpu_per_sec": 3426.415235925005, "train/train_rollout_logprob_abs_diff": 0.014533978886902332, "train/train_rollout_kl": 0.0006908613722771406, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4830622971057892 }, { "rollout": 43, "rollout/raw_reward_unfiltered": 0.67578125, "rollout/response_len/mean": 483.69921875, "rollout/truncated_ratio": 0.05859375, "rollout/prefix_cache_hit_rate": 0.23367198838896952, "perf/rollout_time": 35.148743629455566, "perf/tokens_per_gpu_per_sec": 3522.942421652583, "train/train_rollout_logprob_abs_diff": 0.014406603761017323, "train/train_rollout_kl": 0.0006802676361985505, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5409393310546875 }, { "rollout": 44, "rollout/raw_reward_unfiltered": 0.8125, "rollout/response_len/mean": 425.10546875, "rollout/truncated_ratio": 0.08984375, "rollout/prefix_cache_hit_rate": 0.2479245283018868, "perf/rollout_time": 31.510847568511963, "perf/tokens_per_gpu_per_sec": 3453.6360776518186, "train/train_rollout_logprob_abs_diff": 0.015361673198640347, "train/train_rollout_kl": 0.0007005876977927983, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.6032417416572571 }, { "rollout": 45, "rollout/raw_reward_unfiltered": 0.74609375, "rollout/response_len/mean": 416.96875, "rollout/truncated_ratio": 0.08203125, "rollout/prefix_cache_hit_rate": 0.23746821649110061, "perf/rollout_time": 30.356738090515137, "perf/tokens_per_gpu_per_sec": 3516.3198259878855, "train/train_rollout_logprob_abs_diff": 0.014661542139947414, "train/train_rollout_kl": 0.0007001106860116124, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.6004100441932678 }, { "rollout": 46, "rollout/raw_reward_unfiltered": 0.6953125, "rollout/response_len/mean": 470.0390625, "rollout/truncated_ratio": 0.11328125, "rollout/prefix_cache_hit_rate": 0.25413223140495866, "perf/rollout_time": 35.380285024642944, "perf/tokens_per_gpu_per_sec": 3401.0466539822446, "train/train_rollout_logprob_abs_diff": 0.015030866488814354, "train/train_rollout_kl": 0.0006984873325563967, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.6678542494773865 }, { "rollout": 47, "rollout/raw_reward_unfiltered": 0.79296875, "rollout/response_len/mean": 421.73046875, "rollout/truncated_ratio": 0.03515625, "rollout/prefix_cache_hit_rate": 0.20817997791682002, "perf/rollout_time": 29.969306468963623, "perf/tokens_per_gpu_per_sec": 3602.452399484355, "train/train_rollout_logprob_abs_diff": 0.014126655645668507, "train/train_rollout_kl": 0.0006787325837649405, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.47494393587112427 } ], "timers_s": { "actor_train": { "median": 76.85, "max": 88.3, "n": 48 }, "finalize_and_resume_engines": { "median": 0.0, "max": 0.0, "n": 48 }, "log_probs": { "median": 23.35, "max": 31.4, "n": 48 }, "sleep": { "median": 0.8, "max": 3.7, "n": 48 }, "train": { "median": 100.3, "max": 119.8, "n": 48 }, "train_wait": { "median": 52.0, "max": 121.3, "n": 48 }, "update_weights": { "median": 0.6, "max": 0.7, "n": 48 }, "update_weights_implementation": { "median": 0.2, "max": 0.3, "n": 48 }, "wake_up": { "median": 0.9, "max": 0.9, "n": 48 } } }, "plain-48-s2": { "rollouts": [ { "rollout": 0, "rollout/raw_reward_unfiltered": 0.58203125, "rollout/response_len/mean": 752.55859375, "rollout/truncated_ratio": 0.38671875, "rollout/prefix_cache_hit_rate": 0.16303956834532374, "perf/rollout_time": 61.408626079559326, "perf/tokens_per_gpu_per_sec": 3137.2628293360854, "train/train_rollout_logprob_abs_diff": 0.01525168213993311, "train/train_rollout_kl": 0.0006787290913052857, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.34759414196014404 }, { "rollout": 1, "rollout/raw_reward_unfiltered": 0.66015625, "rollout/response_len/mean": 742.77734375, "rollout/truncated_ratio": 0.30859375, "rollout/prefix_cache_hit_rate": 0.2695176288260364, "perf/rollout_time": 1606.3647954463959, "perf/tokens_per_gpu_per_sec": 118.37348561113016, "train/train_rollout_logprob_abs_diff": 0.015045513398945332, "train/train_rollout_kl": 0.0006847805925644934, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4764772057533264 }, { "rollout": 2, "rollout/raw_reward_unfiltered": 0.64453125, "rollout/response_len/mean": 702.234375, "rollout/truncated_ratio": 0.27734375, "rollout/prefix_cache_hit_rate": 0.24550898203592814, "perf/rollout_time": 101.79760932922363, "perf/tokens_per_gpu_per_sec": 1765.974674499471, "train/train_rollout_logprob_abs_diff": 0.015226084738969803, "train/train_rollout_kl": 0.0006906750495545566, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.47439974546432495 }, { "rollout": 3, "rollout/raw_reward_unfiltered": 0.73828125, "rollout/response_len/mean": 697.0390625, "rollout/truncated_ratio": 0.25, "rollout/prefix_cache_hit_rate": 0.2602759009009009, "perf/rollout_time": 101.53148579597473, "perf/tokens_per_gpu_per_sec": 1757.5040747318053, "train/train_rollout_logprob_abs_diff": 0.015218325890600681, "train/train_rollout_kl": 0.0006967055960558355, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4580579698085785 }, { "rollout": 4, "rollout/raw_reward_unfiltered": 0.609375, "rollout/response_len/mean": 666.0703125, "rollout/truncated_ratio": 0.28125, "rollout/prefix_cache_hit_rate": 0.18254558987718644, "perf/rollout_time": 144.50163316726685, "perf/tokens_per_gpu_per_sec": 1180.014344907941, "train/train_rollout_logprob_abs_diff": 0.01580769568681717, "train/train_rollout_kl": 0.0007261783466674387, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.3864086866378784 }, { "rollout": 5, "rollout/raw_reward_unfiltered": 0.68359375, "rollout/response_len/mean": 724.23046875, "rollout/truncated_ratio": 0.2890625, "rollout/prefix_cache_hit_rate": 0.23129133310128785, "perf/rollout_time": 104.90688920021057, "perf/tokens_per_gpu_per_sec": 1767.3100538341753, "train/train_rollout_logprob_abs_diff": 0.015242335386574268, "train/train_rollout_kl": 0.0006913143442943692, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4605361819267273 }, { "rollout": 6, "rollout/raw_reward_unfiltered": 0.6796875, "rollout/response_len/mean": 700.234375, "rollout/truncated_ratio": 0.25390625, "rollout/prefix_cache_hit_rate": 0.253406613372093, "perf/rollout_time": 147.47803354263306, "perf/tokens_per_gpu_per_sec": 1215.5030528541688, "train/train_rollout_logprob_abs_diff": 0.015092396177351475, "train/train_rollout_kl": 0.0006816089735366404, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4870711863040924 }, { "rollout": 7, "rollout/raw_reward_unfiltered": 0.62890625, "rollout/response_len/mean": 712.49609375, "rollout/truncated_ratio": 0.33203125, "rollout/prefix_cache_hit_rate": 0.2355091613812544, "perf/rollout_time": 59.11879277229309, "perf/tokens_per_gpu_per_sec": 3085.2964251577887, "train/train_rollout_logprob_abs_diff": 0.01573433168232441, "train/train_rollout_kl": 0.0007396869477815926, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.510966956615448 }, { "rollout": 8, "rollout/raw_reward_unfiltered": 0.65234375, "rollout/response_len/mean": 686.1015625, "rollout/truncated_ratio": 0.25, "rollout/prefix_cache_hit_rate": 0.20538965768390385, "perf/rollout_time": 100.6363456249237, "perf/tokens_per_gpu_per_sec": 1745.3137721696078, "train/train_rollout_logprob_abs_diff": 0.01612410694360733, "train/train_rollout_kl": 0.0007455660379491746, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5015780329704285 }, { "rollout": 9, "rollout/raw_reward_unfiltered": 0.53515625, "rollout/response_len/mean": 787.3828125, "rollout/truncated_ratio": 0.3671875, "rollout/prefix_cache_hit_rate": 0.23397893609425205, "perf/rollout_time": 157.28912830352783, "perf/tokens_per_gpu_per_sec": 1281.5253169374898, "train/train_rollout_logprob_abs_diff": 0.015281174331903458, "train/train_rollout_kl": 0.0006972383125685155, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.48450276255607605 }, { "rollout": 10, "rollout/raw_reward_unfiltered": 0.6953125, "rollout/response_len/mean": 704.5703125, "rollout/truncated_ratio": 0.30859375, "rollout/prefix_cache_hit_rate": 0.23884083044982699, "perf/rollout_time": 103.2744767665863, "perf/tokens_per_gpu_per_sec": 1746.5109061521518, "train/train_rollout_logprob_abs_diff": 0.015055423602461815, "train/train_rollout_kl": 0.0007040108321234584, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5049099326133728 }, { "rollout": 11, "rollout/raw_reward_unfiltered": 0.63671875, "rollout/response_len/mean": 661.28515625, "rollout/truncated_ratio": 0.24609375, "rollout/prefix_cache_hit_rate": 0.24781517292673857, "perf/rollout_time": 98.02447414398193, "perf/tokens_per_gpu_per_sec": 1727.0074792887144, "train/train_rollout_logprob_abs_diff": 0.015128152444958687, "train/train_rollout_kl": 0.0006990459514781833, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5105209350585938 }, { "rollout": 12, "rollout/raw_reward_unfiltered": 0.6171875, "rollout/response_len/mean": 649.5625, "rollout/truncated_ratio": 0.23828125, "rollout/prefix_cache_hit_rate": 0.21186815129020856, "perf/rollout_time": 142.44463086128235, "perf/tokens_per_gpu_per_sec": 1167.3869277806418, "train/train_rollout_logprob_abs_diff": 0.015790827572345734, "train/train_rollout_kl": 0.0007497806800529361, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4742433428764343 }, { "rollout": 13, "rollout/raw_reward_unfiltered": 0.61328125, "rollout/response_len/mean": 655.0859375, "rollout/truncated_ratio": 0.28125, "rollout/prefix_cache_hit_rate": 0.24522760646108663, "perf/rollout_time": 143.48572993278503, "perf/tokens_per_gpu_per_sec": 1168.7712783602865, "train/train_rollout_logprob_abs_diff": 0.014979574829339981, "train/train_rollout_kl": 0.0006651494768448174, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.44260600209236145 }, { "rollout": 14, "rollout/raw_reward_unfiltered": 0.7421875, "rollout/response_len/mean": 588.62109375, "rollout/truncated_ratio": 0.13671875, "rollout/prefix_cache_hit_rate": 0.22747333578521514, "perf/rollout_time": 44.96698045730591, "perf/tokens_per_gpu_per_sec": 3351.0588984970964, "train/train_rollout_logprob_abs_diff": 0.014972541481256485, "train/train_rollout_kl": 0.0007144781993702054, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5255929231643677 }, { "rollout": 15, "rollout/raw_reward_unfiltered": 0.71875, "rollout/response_len/mean": 591.85546875, "rollout/truncated_ratio": 0.20703125, "rollout/prefix_cache_hit_rate": 0.2521217712177122, "perf/rollout_time": 137.11588740348816, "perf/tokens_per_gpu_per_sec": 1105.0141808449948, "train/train_rollout_logprob_abs_diff": 0.014856726862490177, "train/train_rollout_kl": 0.0006898688152432442, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4714270234107971 }, { "rollout": 16, "rollout/raw_reward_unfiltered": 0.5703125, "rollout/response_len/mean": 657.21484375, "rollout/truncated_ratio": 0.29296875, "rollout/prefix_cache_hit_rate": 0.24358497582744515, "perf/rollout_time": 144.01592206954956, "perf/tokens_per_gpu_per_sec": 1168.252770820358, "train/train_rollout_logprob_abs_diff": 0.015482616610825062, "train/train_rollout_kl": 0.0007058379123918712, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5370127558708191 }, { "rollout": 17, "rollout/raw_reward_unfiltered": 0.62109375, "rollout/response_len/mean": 674.39453125, "rollout/truncated_ratio": 0.3046875, "rollout/prefix_cache_hit_rate": 0.2308327538247566, "perf/rollout_time": 146.4970350265503, "perf/tokens_per_gpu_per_sec": 1178.4880149192836, "train/train_rollout_logprob_abs_diff": 0.015472915954887867, "train/train_rollout_kl": 0.0007193459314294159, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5597505569458008 }, { "rollout": 18, "rollout/raw_reward_unfiltered": 0.70703125, "rollout/response_len/mean": 612.0625, "rollout/truncated_ratio": 0.25, "rollout/prefix_cache_hit_rate": 0.2516968325791855, "perf/rollout_time": 94.28405499458313, "perf/tokens_per_gpu_per_sec": 1661.8716707613196, "train/train_rollout_logprob_abs_diff": 0.015468839555978775, "train/train_rollout_kl": 0.0007178616360761225, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4608649015426636 }, { "rollout": 19, "rollout/raw_reward_unfiltered": 0.6328125, "rollout/response_len/mean": 646.64453125, "rollout/truncated_ratio": 0.29296875, "rollout/prefix_cache_hit_rate": 0.2315421388101983, "perf/rollout_time": 143.3345627784729, "perf/tokens_per_gpu_per_sec": 1154.9273028854018, "train/train_rollout_logprob_abs_diff": 0.016102394089102745, "train/train_rollout_kl": 0.0007444372749887407, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.49612510204315186 }, { "rollout": 20, "rollout/raw_reward_unfiltered": 0.73046875, "rollout/response_len/mean": 573.22265625, "rollout/truncated_ratio": 0.17578125, "rollout/prefix_cache_hit_rate": 0.23220577871740664, "perf/rollout_time": 90.12307786941528, "perf/tokens_per_gpu_per_sec": 1628.2732843703764, "train/train_rollout_logprob_abs_diff": 0.014415111392736435, "train/train_rollout_kl": 0.000662067555822432, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5239095091819763 }, { "rollout": 21, "rollout/raw_reward_unfiltered": 0.7578125, "rollout/response_len/mean": 566.8515625, "rollout/truncated_ratio": 0.10546875, "rollout/prefix_cache_hit_rate": 0.2399749373433584, "perf/rollout_time": 88.31594371795654, "perf/tokens_per_gpu_per_sec": 1643.1234711529803, "train/train_rollout_logprob_abs_diff": 0.01402817852795124, "train/train_rollout_kl": 0.0006489473744295537, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4951937198638916 }, { "rollout": 22, "rollout/raw_reward_unfiltered": 0.74609375, "rollout/response_len/mean": 518.421875, "rollout/truncated_ratio": 0.109375, "rollout/prefix_cache_hit_rate": 0.20911086717892427, "perf/rollout_time": 83.963543176651, "perf/tokens_per_gpu_per_sec": 1580.6383935082235, "train/train_rollout_logprob_abs_diff": 0.0151652367785573, "train/train_rollout_kl": 0.0007157651125453413, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5014925599098206 }, { "rollout": 23, "rollout/raw_reward_unfiltered": 0.76171875, "rollout/response_len/mean": 568.76953125, "rollout/truncated_ratio": 0.16015625, "rollout/prefix_cache_hit_rate": 0.1982116920842411, "perf/rollout_time": 89.44818496704102, "perf/tokens_per_gpu_per_sec": 1627.8139132018284, "train/train_rollout_logprob_abs_diff": 0.01462508924305439, "train/train_rollout_kl": 0.0006826756871305406, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.481305330991745 }, { "rollout": 24, "rollout/raw_reward_unfiltered": 0.69140625, "rollout/response_len/mean": 527.33203125, "rollout/truncated_ratio": 0.1640625, "rollout/prefix_cache_hit_rate": 0.19335016835016836, "perf/rollout_time": 131.06620860099792, "perf/tokens_per_gpu_per_sec": 1029.990883546258, "train/train_rollout_logprob_abs_diff": 0.014917274937033653, "train/train_rollout_kl": 0.0006857553962618113, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4051929712295532 }, { "rollout": 25, "rollout/raw_reward_unfiltered": 0.6796875, "rollout/response_len/mean": 550.96875, "rollout/truncated_ratio": 0.1640625, "rollout/prefix_cache_hit_rate": 0.23017601951899616, "perf/rollout_time": 88.39088344573975, "perf/tokens_per_gpu_per_sec": 1595.7301760264081, "train/train_rollout_logprob_abs_diff": 0.015060408972203732, "train/train_rollout_kl": 0.0007041142089292407, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5626997351646423 }, { "rollout": 26, "rollout/raw_reward_unfiltered": 0.6953125, "rollout/response_len/mean": 514.984375, "rollout/truncated_ratio": 0.1015625, "rollout/prefix_cache_hit_rate": 0.20023958717287135, "perf/rollout_time": 83.74388194084167, "perf/tokens_per_gpu_per_sec": 1574.2761971929071, "train/train_rollout_logprob_abs_diff": 0.014844565652310848, "train/train_rollout_kl": 0.0006915725534781814, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5518562197685242 }, { "rollout": 27, "rollout/raw_reward_unfiltered": 0.75, "rollout/response_len/mean": 544.328125, "rollout/truncated_ratio": 0.1328125, "rollout/prefix_cache_hit_rate": 0.234049245147376, "perf/rollout_time": 86.5747299194336, "perf/tokens_per_gpu_per_sec": 1609.5689831164034, "train/train_rollout_logprob_abs_diff": 0.013762340880930424, "train/train_rollout_kl": 0.00062677072128281, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4400290846824646 }, { "rollout": 28, "rollout/raw_reward_unfiltered": 0.80078125, "rollout/response_len/mean": 512.2265625, "rollout/truncated_ratio": 0.09375, "rollout/prefix_cache_hit_rate": 0.19399289985052318, "perf/rollout_time": 37.824491024017334, "perf/tokens_per_gpu_per_sec": 3466.8014413395986, "train/train_rollout_logprob_abs_diff": 0.015247168950736523, "train/train_rollout_kl": 0.0007237099343910813, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4728754162788391 }, { "rollout": 29, "rollout/raw_reward_unfiltered": 0.74609375, "rollout/response_len/mean": 496.24609375, "rollout/truncated_ratio": 0.13671875, "rollout/prefix_cache_hit_rate": 0.24126294378698224, "perf/rollout_time": 82.36993789672852, "perf/tokens_per_gpu_per_sec": 1542.2981155974092, "train/train_rollout_logprob_abs_diff": 0.015082410536706448, "train/train_rollout_kl": 0.0006957974401302636, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.6027895212173462 }, { "rollout": 30, "rollout/raw_reward_unfiltered": 0.74609375, "rollout/response_len/mean": 494.734375, "rollout/truncated_ratio": 0.125, "rollout/prefix_cache_hit_rate": 0.2553585346843336, "perf/rollout_time": 82.99393916130066, "perf/tokens_per_gpu_per_sec": 1526.039145507347, "train/train_rollout_logprob_abs_diff": 0.014500949531793594, "train/train_rollout_kl": 0.0006667535053566098, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.461325466632843 }, { "rollout": 31, "rollout/raw_reward_unfiltered": 0.71875, "rollout/response_len/mean": 495.0859375, "rollout/truncated_ratio": 0.109375, "rollout/prefix_cache_hit_rate": 0.229957805907173, "perf/rollout_time": 82.36445569992065, "perf/tokens_per_gpu_per_sec": 1538.7948469150401, "train/train_rollout_logprob_abs_diff": 0.014625342562794685, "train/train_rollout_kl": 0.0006811160710640252, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.49958011507987976 }, { "rollout": 32, "rollout/raw_reward_unfiltered": 0.78515625, "rollout/response_len/mean": 442.71875, "rollout/truncated_ratio": 0.0859375, "rollout/prefix_cache_hit_rate": 0.18976565360673747, "perf/rollout_time": 77.15389490127563, "perf/tokens_per_gpu_per_sec": 1468.9601885299785, "train/train_rollout_logprob_abs_diff": 0.014607522636651993, "train/train_rollout_kl": 0.0006796806119382381, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5931206941604614 }, { "rollout": 33, "rollout/raw_reward_unfiltered": 0.81640625, "rollout/response_len/mean": 463.49609375, "rollout/truncated_ratio": 0.05078125, "rollout/prefix_cache_hit_rate": 0.2010290334435869, "perf/rollout_time": 78.7480058670044, "perf/tokens_per_gpu_per_sec": 1506.7683136052176, "train/train_rollout_logprob_abs_diff": 0.01432132814079523, "train/train_rollout_kl": 0.0006773420609533787, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.6347492337226868 }, { "rollout": 34, "rollout/raw_reward_unfiltered": 0.7109375, "rollout/response_len/mean": 503.97265625, "rollout/truncated_ratio": 0.1171875, "rollout/prefix_cache_hit_rate": 0.21078607322325915, "perf/rollout_time": 83.6487283706665, "perf/tokens_per_gpu_per_sec": 1542.3665429592234, "train/train_rollout_logprob_abs_diff": 0.0157795250415802, "train/train_rollout_kl": 0.0007246676250360906, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5765215158462524 }, { "rollout": 35, "rollout/raw_reward_unfiltered": 0.71875, "rollout/response_len/mean": 518.3515625, "rollout/truncated_ratio": 0.14453125, "rollout/prefix_cache_hit_rate": 0.23101155057752887, "perf/rollout_time": 130.5885510444641, "perf/tokens_per_gpu_per_sec": 1016.153398890364, "train/train_rollout_logprob_abs_diff": 0.015291202813386917, "train/train_rollout_kl": 0.0007091988809406757, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.49425268173217773 }, { "rollout": 36, "rollout/raw_reward_unfiltered": 0.8046875, "rollout/response_len/mean": 443.57421875, "rollout/truncated_ratio": 0.06640625, "rollout/prefix_cache_hit_rate": 0.23414376321353064, "perf/rollout_time": 77.23959970474243, "perf/tokens_per_gpu_per_sec": 1470.1655683623103, "train/train_rollout_logprob_abs_diff": 0.01409775298088789, "train/train_rollout_kl": 0.0006506751524284482, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.6474598050117493 }, { "rollout": 37, "rollout/raw_reward_unfiltered": 0.77734375, "rollout/response_len/mean": 532.76953125, "rollout/truncated_ratio": 0.109375, "rollout/prefix_cache_hit_rate": 0.18905709342560553, "perf/rollout_time": 85.463383436203, "perf/tokens_per_gpu_per_sec": 1595.8764387301858, "train/train_rollout_logprob_abs_diff": 0.015058303251862526, "train/train_rollout_kl": 0.0006966817891225219, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.575229287147522 }, { "rollout": 38, "rollout/raw_reward_unfiltered": 0.77734375, "rollout/response_len/mean": 441.67578125, "rollout/truncated_ratio": 0.03515625, "rollout/prefix_cache_hit_rate": 0.21361940298507462, "perf/rollout_time": 76.49131917953491, "perf/tokens_per_gpu_per_sec": 1478.1938815123922, "train/train_rollout_logprob_abs_diff": 0.013843555003404617, "train/train_rollout_kl": 0.0006425278261303902, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5271176099777222 }, { "rollout": 39, "rollout/raw_reward_unfiltered": 0.71875, "rollout/response_len/mean": 483.2890625, "rollout/truncated_ratio": 0.09375, "rollout/prefix_cache_hit_rate": 0.2302879670894755, "perf/rollout_time": 81.33856987953186, "perf/tokens_per_gpu_per_sec": 1521.0741986641883, "train/train_rollout_logprob_abs_diff": 0.014759664423763752, "train/train_rollout_kl": 0.0006905515328980982, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4597359001636505 }, { "rollout": 40, "rollout/raw_reward_unfiltered": 0.69921875, "rollout/response_len/mean": 489.3203125, "rollout/truncated_ratio": 0.15234375, "rollout/prefix_cache_hit_rate": 0.22920979986197376, "perf/rollout_time": 82.89181303977966, "perf/tokens_per_gpu_per_sec": 1511.198698716905, "train/train_rollout_logprob_abs_diff": 0.015280772931873798, "train/train_rollout_kl": 0.0007182887638919055, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.575655996799469 }, { "rollout": 41, "rollout/raw_reward_unfiltered": 0.78515625, "rollout/response_len/mean": 438.00390625, "rollout/truncated_ratio": 0.078125, "rollout/prefix_cache_hit_rate": 0.2331033262561925, "perf/rollout_time": 77.37404346466064, "perf/tokens_per_gpu_per_sec": 1449.1810816532177, "train/train_rollout_logprob_abs_diff": 0.014812755398452282, "train/train_rollout_kl": 0.0006882799789309502, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5869902968406677 }, { "rollout": 42, "rollout/raw_reward_unfiltered": 0.76171875, "rollout/response_len/mean": 480.53125, "rollout/truncated_ratio": 0.13671875, "rollout/prefix_cache_hit_rate": 0.2526429809358752, "perf/rollout_time": 81.9634108543396, "perf/tokens_per_gpu_per_sec": 1500.8648214825585, "train/train_rollout_logprob_abs_diff": 0.014492746442556381, "train/train_rollout_kl": 0.0006682407693006098, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4923684597015381 }, { "rollout": 43, "rollout/raw_reward_unfiltered": 0.83984375, "rollout/response_len/mean": 429.27734375, "rollout/truncated_ratio": 0.046875, "rollout/prefix_cache_hit_rate": 0.2028250633828323, "perf/rollout_time": 75.92094612121582, "perf/tokens_per_gpu_per_sec": 1447.4924986385313, "train/train_rollout_logprob_abs_diff": 0.013985930010676384, "train/train_rollout_kl": 0.0006443631718866527, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5254744291305542 }, { "rollout": 44, "rollout/raw_reward_unfiltered": 0.6875, "rollout/response_len/mean": 521.57421875, "rollout/truncated_ratio": 0.18359375, "rollout/prefix_cache_hit_rate": 0.2289136302294197, "perf/rollout_time": 85.90356636047363, "perf/tokens_per_gpu_per_sec": 1554.3359333849178, "train/train_rollout_logprob_abs_diff": 0.015992218628525734, "train/train_rollout_kl": 0.0007573578041046858, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5052977204322815 }, { "rollout": 45, "rollout/raw_reward_unfiltered": 0.7734375, "rollout/response_len/mean": 476.28515625, "rollout/truncated_ratio": 0.10546875, "rollout/prefix_cache_hit_rate": 0.20151228733459356, "perf/rollout_time": 80.49055051803589, "perf/tokens_per_gpu_per_sec": 1514.823780124088, "train/train_rollout_logprob_abs_diff": 0.015417185612022877, "train/train_rollout_kl": 0.0007304837345145643, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5095674991607666 }, { "rollout": 46, "rollout/raw_reward_unfiltered": 0.6953125, "rollout/response_len/mean": 494.171875, "rollout/truncated_ratio": 0.12109375, "rollout/prefix_cache_hit_rate": 0.24352377572746628, "perf/rollout_time": 82.78148221969604, "perf/tokens_per_gpu_per_sec": 1528.2161735671384, "train/train_rollout_logprob_abs_diff": 0.015080174431204796, "train/train_rollout_kl": 0.000698937801644206, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5794944167137146 }, { "rollout": 47, "rollout/raw_reward_unfiltered": 0.6484375, "rollout/response_len/mean": 514.0390625, "rollout/truncated_ratio": 0.12109375, "rollout/prefix_cache_hit_rate": 0.24259425493716338, "perf/rollout_time": 83.99452042579651, "perf/tokens_per_gpu_per_sec": 1566.6974385103422, "train/train_rollout_logprob_abs_diff": 0.014963770285248756, "train/train_rollout_kl": 0.000676291820127517, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5826910734176636 } ], "timers_s": { "actor_train": { "median": 164.3, "max": 265.4, "n": 48 }, "finalize_and_resume_engines": { "median": 0.0, "max": 0.0, "n": 48 }, "log_probs": { "median": 66.1, "max": 115.2, "n": 48 }, "sleep": { "median": 0.7, "max": 4.9, "n": 48 }, "train": { "median": 229.8, "max": 337.6, "n": 48 }, "train_wait": { "median": 91.65, "max": 1614.8, "n": 48 }, "update_weights": { "median": 0.6, "max": 0.7, "n": 48 }, "update_weights_implementation": { "median": 0.2, "max": 0.3, "n": 48 }, "wake_up": { "median": 0.8, "max": 0.8, "n": 48 } } }, "plain-48-s3": { "rollouts": [ { "rollout": 0, "rollout/raw_reward_unfiltered": 0.6328125, "rollout/response_len/mean": 707.78125, "rollout/truncated_ratio": 0.3203125, "rollout/prefix_cache_hit_rate": 0.16882011217948717, "perf/rollout_time": 104.43908667564392, "perf/tokens_per_gpu_per_sec": 1734.9060181149162, "train/train_rollout_logprob_abs_diff": 0.014946463517844677, "train/train_rollout_kl": 0.0006613255827687681, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.429164320230484 }, { "rollout": 1, "rollout/raw_reward_unfiltered": 0.63671875, "rollout/response_len/mean": 733.1953125, "rollout/truncated_ratio": 0.3046875, "rollout/prefix_cache_hit_rate": 0.23004238784881667, "perf/rollout_time": 151.98661851882935, "perf/tokens_per_gpu_per_sec": 1234.9639845217455, "train/train_rollout_logprob_abs_diff": 0.015562252141535282, "train/train_rollout_kl": 0.0007177075021900237, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.47511425614356995 }, { "rollout": 2, "rollout/raw_reward_unfiltered": 0.5546875, "rollout/response_len/mean": 758.94140625, "rollout/truncated_ratio": 0.41796875, "rollout/prefix_cache_hit_rate": 0.25, "perf/rollout_time": 155.03394651412964, "perf/tokens_per_gpu_per_sec": 1253.2029556655368, "train/train_rollout_logprob_abs_diff": 0.01599210500717163, "train/train_rollout_kl": 0.0007346115307882428, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4447347819805145 }, { "rollout": 3, "rollout/raw_reward_unfiltered": 0.609375, "rollout/response_len/mean": 750.8671875, "rollout/truncated_ratio": 0.34375, "rollout/prefix_cache_hit_rate": 0.2025643349039507, "perf/rollout_time": 107.86071109771729, "perf/tokens_per_gpu_per_sec": 1782.131770166571, "train/train_rollout_logprob_abs_diff": 0.015438200905919075, "train/train_rollout_kl": 0.0006837508408352733, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5516325235366821 }, { "rollout": 4, "rollout/raw_reward_unfiltered": 0.52734375, "rollout/response_len/mean": 780.46484375, "rollout/truncated_ratio": 0.42578125, "rollout/prefix_cache_hit_rate": 0.2121226239284383, "perf/rollout_time": 157.1394350528717, "perf/tokens_per_gpu_per_sec": 1271.4758706671873, "train/train_rollout_logprob_abs_diff": 0.01635424792766571, "train/train_rollout_kl": 0.000739757320843637, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5109831094741821 }, { "rollout": 5, "rollout/raw_reward_unfiltered": 0.59375, "rollout/response_len/mean": 767.5546875, "rollout/truncated_ratio": 0.40625, "rollout/prefix_cache_hit_rate": 0.18763029881862406, "perf/rollout_time": 156.1864833831787, "perf/tokens_per_gpu_per_sec": 1258.0730146662768, "train/train_rollout_logprob_abs_diff": 0.015741078183054924, "train/train_rollout_kl": 0.0007413745624944568, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5047527551651001 }, { "rollout": 6, "rollout/raw_reward_unfiltered": 0.65234375, "rollout/response_len/mean": 688.67578125, "rollout/truncated_ratio": 0.26171875, "rollout/prefix_cache_hit_rate": 0.22339039821362114, "perf/rollout_time": 147.2279553413391, "perf/tokens_per_gpu_per_sec": 1197.469594624586, "train/train_rollout_logprob_abs_diff": 0.015382969751954079, "train/train_rollout_kl": 0.0006961311446502805, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5190079212188721 }, { "rollout": 7, "rollout/raw_reward_unfiltered": 0.70703125, "rollout/response_len/mean": 681.41796875, "rollout/truncated_ratio": 0.26171875, "rollout/prefix_cache_hit_rate": 0.2562862669245648, "perf/rollout_time": 145.750559091568, "perf/tokens_per_gpu_per_sec": 1196.859902886588, "train/train_rollout_logprob_abs_diff": 0.015760013833642006, "train/train_rollout_kl": 0.0007182589615695179, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.49740099906921387 }, { "rollout": 8, "rollout/raw_reward_unfiltered": 0.68359375, "rollout/response_len/mean": 633.76171875, "rollout/truncated_ratio": 0.2578125, "rollout/prefix_cache_hit_rate": 0.2478424015009381, "perf/rollout_time": 141.32547616958618, "perf/tokens_per_gpu_per_sec": 1148.0095761737498, "train/train_rollout_logprob_abs_diff": 0.015256368555128574, "train/train_rollout_kl": 0.0007115419721230865, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4274096190929413 }, { "rollout": 9, "rollout/raw_reward_unfiltered": 0.64453125, "rollout/response_len/mean": 697.0703125, "rollout/truncated_ratio": 0.3046875, "rollout/prefix_cache_hit_rate": 0.23143912737508796, "perf/rollout_time": 103.22877168655396, "perf/tokens_per_gpu_per_sec": 1728.6847172980938, "train/train_rollout_logprob_abs_diff": 0.015476806089282036, "train/train_rollout_kl": 0.0007131364545784891, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.39073795080184937 }, { "rollout": 10, "rollout/raw_reward_unfiltered": 0.75390625, "rollout/response_len/mean": 616.12890625, "rollout/truncated_ratio": 0.22265625, "rollout/prefix_cache_hit_rate": 0.2373630258041711, "perf/rollout_time": 139.76897263526917, "perf/tokens_per_gpu_per_sec": 1128.4979564928049, "train/train_rollout_logprob_abs_diff": 0.015178020112216473, "train/train_rollout_kl": 0.0007148389704525471, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4220947325229645 }, { "rollout": 11, "rollout/raw_reward_unfiltered": 0.6640625, "rollout/response_len/mean": 656.87109375, "rollout/truncated_ratio": 0.26953125, "rollout/prefix_cache_hit_rate": 0.23137039540385265, "perf/rollout_time": 99.2485420703888, "perf/tokens_per_gpu_per_sec": 1694.3221179081775, "train/train_rollout_logprob_abs_diff": 0.015490984544157982, "train/train_rollout_kl": 0.0007082612719386816, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.44377627968788147 }, { "rollout": 12, "rollout/raw_reward_unfiltered": 0.55859375, "rollout/response_len/mean": 687.109375, "rollout/truncated_ratio": 0.33984375, "rollout/prefix_cache_hit_rate": 0.22418200408997954, "perf/rollout_time": 102.4974012374878, "perf/tokens_per_gpu_per_sec": 1716.1410716398307, "train/train_rollout_logprob_abs_diff": 0.015560918487608433, "train/train_rollout_kl": 0.0007186852162703872, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.46735328435897827 }, { "rollout": 13, "rollout/raw_reward_unfiltered": 0.63671875, "rollout/response_len/mean": 613.76953125, "rollout/truncated_ratio": 0.19921875, "rollout/prefix_cache_hit_rate": 0.22123773173391495, "perf/rollout_time": 139.71479606628418, "perf/tokens_per_gpu_per_sec": 1124.6124564033719, "train/train_rollout_logprob_abs_diff": 0.015888521447777748, "train/train_rollout_kl": 0.0007299978751689196, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5154780149459839 }, { "rollout": 14, "rollout/raw_reward_unfiltered": 0.7265625, "rollout/response_len/mean": 617.6640625, "rollout/truncated_ratio": 0.19140625, "rollout/prefix_cache_hit_rate": 0.19795825771324863, "perf/rollout_time": 94.02757358551025, "perf/tokens_per_gpu_per_sec": 1681.6556460026186, "train/train_rollout_logprob_abs_diff": 0.01559443399310112, "train/train_rollout_kl": 0.0007275853422470391, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4724407196044922 }, { "rollout": 15, "rollout/raw_reward_unfiltered": 0.81640625, "rollout/response_len/mean": 542.21875, "rollout/truncated_ratio": 0.09375, "rollout/prefix_cache_hit_rate": 0.24740859404447796, "perf/rollout_time": 131.60531783103943, "perf/tokens_per_gpu_per_sec": 1054.729415859986, "train/train_rollout_logprob_abs_diff": 0.014334797859191895, "train/train_rollout_kl": 0.0006671622395515442, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5148454904556274 }, { "rollout": 16, "rollout/raw_reward_unfiltered": 0.71875, "rollout/response_len/mean": 557.921875, "rollout/truncated_ratio": 0.203125, "rollout/prefix_cache_hit_rate": 0.2333479020979021, "perf/rollout_time": 43.056248903274536, "perf/tokens_per_gpu_per_sec": 3317.2420644645977, "train/train_rollout_logprob_abs_diff": 0.01460379920899868, "train/train_rollout_kl": 0.0006806293386034667, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4549553096294403 }, { "rollout": 17, "rollout/raw_reward_unfiltered": 0.67578125, "rollout/response_len/mean": 593.5625, "rollout/truncated_ratio": 0.2109375, "rollout/prefix_cache_hit_rate": 0.24507617985878855, "perf/rollout_time": 92.03353762626648, "perf/tokens_per_gpu_per_sec": 1651.0503010006291, "train/train_rollout_logprob_abs_diff": 0.015744350850582123, "train/train_rollout_kl": 0.0007203637505881488, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5955234169960022 }, { "rollout": 18, "rollout/raw_reward_unfiltered": 0.68359375, "rollout/response_len/mean": 606.48046875, "rollout/truncated_ratio": 0.17578125, "rollout/prefix_cache_hit_rate": 0.2350685673042223, "perf/rollout_time": 93.03625154495239, "perf/tokens_per_gpu_per_sec": 1668.8011116288728, "train/train_rollout_logprob_abs_diff": 0.01499306783080101, "train/train_rollout_kl": 0.0007043020450510085, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5786668658256531 }, { "rollout": 19, "rollout/raw_reward_unfiltered": 0.73828125, "rollout/response_len/mean": 572.74609375, "rollout/truncated_ratio": 0.16015625, "rollout/prefix_cache_hit_rate": 0.24107471475892528, "perf/rollout_time": 135.7566454410553, "perf/tokens_per_gpu_per_sec": 1080.0428923655365, "train/train_rollout_logprob_abs_diff": 0.014841067604720592, "train/train_rollout_kl": 0.0006971181719563901, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4794418513774872 }, { "rollout": 20, "rollout/raw_reward_unfiltered": 0.7421875, "rollout/response_len/mean": 583.6171875, "rollout/truncated_ratio": 0.203125, "rollout/prefix_cache_hit_rate": 0.23790181691125087, "perf/rollout_time": 91.58616018295288, "perf/tokens_per_gpu_per_sec": 1631.3163441020563, "train/train_rollout_logprob_abs_diff": 0.01600435934960842, "train/train_rollout_kl": 0.0007378841401077807, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5571737289428711 }, { "rollout": 21, "rollout/raw_reward_unfiltered": 0.73046875, "rollout/response_len/mean": 611.2890625, "rollout/truncated_ratio": 0.234375, "rollout/prefix_cache_hit_rate": 0.22106095936042638, "perf/rollout_time": 94.31646227836609, "perf/tokens_per_gpu_per_sec": 1659.2013336774085, "train/train_rollout_logprob_abs_diff": 0.01563888043165207, "train/train_rollout_kl": 0.0007074312306940556, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.47890737652778625 }, { "rollout": 22, "rollout/raw_reward_unfiltered": 0.7265625, "rollout/response_len/mean": 561.94921875, "rollout/truncated_ratio": 0.19140625, "rollout/prefix_cache_hit_rate": 0.19490394877267878, "perf/rollout_time": 89.18104839324951, "perf/tokens_per_gpu_per_sec": 1613.1117831856448, "train/train_rollout_logprob_abs_diff": 0.015313385054469109, "train/train_rollout_kl": 0.0007027019164524972, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.46052980422973633 }, { "rollout": 23, "rollout/raw_reward_unfiltered": 0.68359375, "rollout/response_len/mean": 555.12109375, "rollout/truncated_ratio": 0.19921875, "rollout/prefix_cache_hit_rate": 0.24792296072507553, "perf/rollout_time": 134.1308536529541, "perf/tokens_per_gpu_per_sec": 1059.4952326754997, "train/train_rollout_logprob_abs_diff": 0.015244018286466599, "train/train_rollout_kl": 0.0007002730271779001, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5241518616676331 }, { "rollout": 24, "rollout/raw_reward_unfiltered": 0.71875, "rollout/response_len/mean": 532.74609375, "rollout/truncated_ratio": 0.15234375, "rollout/prefix_cache_hit_rate": 0.23730228471001757, "perf/rollout_time": 86.10702610015869, "perf/tokens_per_gpu_per_sec": 1583.8777179618407, "train/train_rollout_logprob_abs_diff": 0.014960800297558308, "train/train_rollout_kl": 0.0006811818457208574, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5792106986045837 }, { "rollout": 25, "rollout/raw_reward_unfiltered": 0.75, "rollout/response_len/mean": 523.24609375, "rollout/truncated_ratio": 0.109375, "rollout/prefix_cache_hit_rate": 0.19469105113636365, "perf/rollout_time": 85.27961444854736, "perf/tokens_per_gpu_per_sec": 1570.727082505961, "train/train_rollout_logprob_abs_diff": 0.014428290538489819, "train/train_rollout_kl": 0.0006810532649978995, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.44866228103637695 }, { "rollout": 26, "rollout/raw_reward_unfiltered": 0.76953125, "rollout/response_len/mean": 501.00390625, "rollout/truncated_ratio": 0.1171875, "rollout/prefix_cache_hit_rate": 0.18747839612858624, "perf/rollout_time": 83.09267711639404, "perf/tokens_per_gpu_per_sec": 1543.5415544542025, "train/train_rollout_logprob_abs_diff": 0.015202391892671585, "train/train_rollout_kl": 0.0006997066084295511, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.3507111668586731 }, { "rollout": 27, "rollout/raw_reward_unfiltered": 0.73046875, "rollout/response_len/mean": 498.75390625, "rollout/truncated_ratio": 0.1171875, "rollout/prefix_cache_hit_rate": 0.23745551601423487, "perf/rollout_time": 82.69900321960449, "perf/tokens_per_gpu_per_sec": 1543.9242920612633, "train/train_rollout_logprob_abs_diff": 0.014727236703038216, "train/train_rollout_kl": 0.0006731738103553653, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4640332758426666 }, { "rollout": 28, "rollout/raw_reward_unfiltered": 0.76171875, "rollout/response_len/mean": 512.6953125, "rollout/truncated_ratio": 0.171875, "rollout/prefix_cache_hit_rate": 0.268427636643572, "perf/rollout_time": 84.83979177474976, "perf/tokens_per_gpu_per_sec": 1547.0334998990766, "train/train_rollout_logprob_abs_diff": 0.014963085763156414, "train/train_rollout_kl": 0.0007013016729615629, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4654533863067627 }, { "rollout": 29, "rollout/raw_reward_unfiltered": 0.625, "rollout/response_len/mean": 573.7109375, "rollout/truncated_ratio": 0.1796875, "rollout/prefix_cache_hit_rate": 0.21649316851008457, "perf/rollout_time": 90.22678804397583, "perf/tokens_per_gpu_per_sec": 1627.787081685948, "train/train_rollout_logprob_abs_diff": 0.015209117904305458, "train/train_rollout_kl": 0.0006958203739486635, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5346885919570923 }, { "rollout": 30, "rollout/raw_reward_unfiltered": 0.828125, "rollout/response_len/mean": 488.5546875, "rollout/truncated_ratio": 0.0859375, "rollout/prefix_cache_hit_rate": 0.2563032999629218, "perf/rollout_time": 36.08993124961853, "perf/tokens_per_gpu_per_sec": 3465.5095110861976, "train/train_rollout_logprob_abs_diff": 0.014287078753113747, "train/train_rollout_kl": 0.0006543620256707072, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5159274339675903 }, { "rollout": 31, "rollout/raw_reward_unfiltered": 0.6796875, "rollout/response_len/mean": 571.5, "rollout/truncated_ratio": 0.1796875, "rollout/prefix_cache_hit_rate": 0.1707306338028169, "perf/rollout_time": 136.02541494369507, "perf/tokens_per_gpu_per_sec": 1075.5637103592703, "train/train_rollout_logprob_abs_diff": 0.01495932787656784, "train/train_rollout_kl": 0.000707201543264091, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5745245218276978 }, { "rollout": 32, "rollout/raw_reward_unfiltered": 0.765625, "rollout/response_len/mean": 437.546875, "rollout/truncated_ratio": 0.06640625, "rollout/prefix_cache_hit_rate": 0.26569928684627575, "perf/rollout_time": 30.931512594223022, "perf/tokens_per_gpu_per_sec": 3621.290735743719, "train/train_rollout_logprob_abs_diff": 0.014505987986922264, "train/train_rollout_kl": 0.0006580452318303287, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5322532653808594 }, { "rollout": 33, "rollout/raw_reward_unfiltered": 0.66015625, "rollout/response_len/mean": 522.359375, "rollout/truncated_ratio": 0.140625, "rollout/prefix_cache_hit_rate": 0.22784974093264249, "perf/rollout_time": 85.91178154945374, "perf/tokens_per_gpu_per_sec": 1556.526911539181, "train/train_rollout_logprob_abs_diff": 0.01471567153930664, "train/train_rollout_kl": 0.0006646676338277757, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.4955158829689026 }, { "rollout": 34, "rollout/raw_reward_unfiltered": 0.75, "rollout/response_len/mean": 463.34375, "rollout/truncated_ratio": 0.06640625, "rollout/prefix_cache_hit_rate": 0.2609411309062742, "perf/rollout_time": 79.20198774337769, "perf/tokens_per_gpu_per_sec": 1497.6391802731976, "train/train_rollout_logprob_abs_diff": 0.01468592043966055, "train/train_rollout_kl": 0.0006723575061187148, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5421310663223267 }, { "rollout": 35, "rollout/raw_reward_unfiltered": 0.625, "rollout/response_len/mean": 529.98828125, "rollout/truncated_ratio": 0.15234375, "rollout/prefix_cache_hit_rate": 0.21548994038748137, "perf/rollout_time": 131.16780042648315, "perf/tokens_per_gpu_per_sec": 1034.377336197264, "train/train_rollout_logprob_abs_diff": 0.015098071657121181, "train/train_rollout_kl": 0.0006893901154398918, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.44896212220191956 }, { "rollout": 36, "rollout/raw_reward_unfiltered": 0.73046875, "rollout/response_len/mean": 480.015625, "rollout/truncated_ratio": 0.12890625, "rollout/prefix_cache_hit_rate": 0.18837962962962962, "perf/rollout_time": 81.89393997192383, "perf/tokens_per_gpu_per_sec": 1500.5261688731673, "train/train_rollout_logprob_abs_diff": 0.015040203928947449, "train/train_rollout_kl": 0.0006948452210053802, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.48088541626930237 }, { "rollout": 37, "rollout/raw_reward_unfiltered": 0.640625, "rollout/response_len/mean": 564.8125, "rollout/truncated_ratio": 0.17578125, "rollout/prefix_cache_hit_rate": 0.2310713025061772, "perf/rollout_time": 89.68950343132019, "perf/tokens_per_gpu_per_sec": 1612.1395979265449, "train/train_rollout_logprob_abs_diff": 0.015200495719909668, "train/train_rollout_kl": 0.0007005641236901283, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.6161320805549622 }, { "rollout": 38, "rollout/raw_reward_unfiltered": 0.78515625, "rollout/response_len/mean": 479.0546875, "rollout/truncated_ratio": 0.1015625, "rollout/prefix_cache_hit_rate": 0.22853889278206027, "perf/rollout_time": 36.191401720047, "perf/tokens_per_gpu_per_sec": 3388.594919551537, "train/train_rollout_logprob_abs_diff": 0.014686555601656437, "train/train_rollout_kl": 0.0006950990064069629, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.507800817489624 }, { "rollout": 39, "rollout/raw_reward_unfiltered": 0.71875, "rollout/response_len/mean": 486.734375, "rollout/truncated_ratio": 0.12109375, "rollout/prefix_cache_hit_rate": 0.23994045470949116, "perf/rollout_time": 36.35929727554321, "perf/tokens_per_gpu_per_sec": 3427.0189287682924, "train/train_rollout_logprob_abs_diff": 0.015183914452791214, "train/train_rollout_kl": 0.0007255493546836078, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5058785676956177 }, { "rollout": 40, "rollout/raw_reward_unfiltered": 0.73828125, "rollout/response_len/mean": 509.0234375, "rollout/truncated_ratio": 0.14453125, "rollout/prefix_cache_hit_rate": 0.2145565927654609, "perf/rollout_time": 83.72647714614868, "perf/tokens_per_gpu_per_sec": 1556.3774380777718, "train/train_rollout_logprob_abs_diff": 0.014945278875529766, "train/train_rollout_kl": 0.0006669381982646883, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5018005967140198 }, { "rollout": 41, "rollout/raw_reward_unfiltered": 0.80078125, "rollout/response_len/mean": 413.1953125, "rollout/truncated_ratio": 0.046875, "rollout/prefix_cache_hit_rate": 0.24163778162911612, "perf/rollout_time": 74.7496280670166, "perf/tokens_per_gpu_per_sec": 1415.09734209199, "train/train_rollout_logprob_abs_diff": 0.014267390593886375, "train/train_rollout_kl": 0.0006629524868912995, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.6064432263374329 }, { "rollout": 42, "rollout/raw_reward_unfiltered": 0.80078125, "rollout/response_len/mean": 408.11328125, "rollout/truncated_ratio": 0.046875, "rollout/prefix_cache_hit_rate": 0.1862448060941828, "perf/rollout_time": 74.55256962776184, "perf/tokens_per_gpu_per_sec": 1401.3869746093221, "train/train_rollout_logprob_abs_diff": 0.013608667999505997, "train/train_rollout_kl": 0.0006271644961088896, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.596887469291687 }, { "rollout": 43, "rollout/raw_reward_unfiltered": 0.73046875, "rollout/response_len/mean": 459.60546875, "rollout/truncated_ratio": 0.109375, "rollout/prefix_cache_hit_rate": 0.23399255054984036, "perf/rollout_time": 79.57858467102051, "perf/tokens_per_gpu_per_sec": 1478.5259185797877, "train/train_rollout_logprob_abs_diff": 0.015310846269130707, "train/train_rollout_kl": 0.0007228241302073002, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5716349482536316 }, { "rollout": 44, "rollout/raw_reward_unfiltered": 0.80859375, "rollout/response_len/mean": 420.01171875, "rollout/truncated_ratio": 0.046875, "rollout/prefix_cache_hit_rate": 0.2453078078078078, "perf/rollout_time": 29.959043502807617, "perf/tokens_per_gpu_per_sec": 3588.9997619558003, "train/train_rollout_logprob_abs_diff": 0.014354592189192772, "train/train_rollout_kl": 0.0006785090081393719, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5050225853919983 }, { "rollout": 45, "rollout/raw_reward_unfiltered": 0.79296875, "rollout/response_len/mean": 447.3984375, "rollout/truncated_ratio": 0.09765625, "rollout/prefix_cache_hit_rate": 0.2503731343283582, "perf/rollout_time": 32.99483370780945, "perf/tokens_per_gpu_per_sec": 3471.270715114751, "train/train_rollout_logprob_abs_diff": 0.014755471609532833, "train/train_rollout_kl": 0.0006759556708857417, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5377292633056641 }, { "rollout": 46, "rollout/raw_reward_unfiltered": 0.765625, "rollout/response_len/mean": 474.7421875, "rollout/truncated_ratio": 0.078125, "rollout/prefix_cache_hit_rate": 0.22692242547425473, "perf/rollout_time": 80.39321446418762, "perf/tokens_per_gpu_per_sec": 1511.7445024435385, "train/train_rollout_logprob_abs_diff": 0.014494070783257484, "train/train_rollout_kl": 0.0006792147760279477, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.9016517400741577 }, { "rollout": 47, "rollout/raw_reward_unfiltered": 0.6875, "rollout/response_len/mean": 513.5546875, "rollout/truncated_ratio": 0.1328125, "rollout/prefix_cache_hit_rate": 0.23639426399447133, "perf/rollout_time": 84.79509902000427, "perf/tokens_per_gpu_per_sec": 1550.4433808018139, "train/train_rollout_logprob_abs_diff": 0.014792761765420437, "train/train_rollout_kl": 0.0006991099799051881, "train/ppo_kl": 0.0, "train/pg_clipfrac": 0.0, "train/grad_norm": 0.5963178873062134 } ], "timers_s": { "actor_train": { "median": 163.05, "max": 218.8, "n": 48 }, "finalize_and_resume_engines": { "median": 0.0, "max": 0.0, "n": 48 }, "log_probs": { "median": 66.3, "max": 116.8, "n": 48 }, "sleep": { "median": 0.7, "max": 4.1, "n": 48 }, "train": { "median": 227.95, "max": 290.9, "n": 48 }, "train_wait": { "median": 95.05, "max": 161.8, "n": 48 }, "update_weights": { "median": 0.6, "max": 0.6, "n": 48 }, "update_weights_implementation": { "median": 0.2, "max": 0.2, "n": 48 }, "wake_up": { "median": 0.8, "max": 0.9, "n": 48 } } } }