diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -2,9 +2,9 @@
   "best_global_step": null,
   "best_metric": null,
   "best_model_checkpoint": null,
-  "epoch": 1.443213296398892,
+  "epoch": 1.6925373134328359,
   "eval_steps": 500,
-  "global_step": 130,
+  "global_step": 140,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
@@ -15,22 +15,22 @@
       "clip_ratio/low_mean": 0.0,
       "clip_ratio/low_min": 0.0,
       "clip_ratio/region_mean": 0.0,
-      "completions/clipped_ratio": 0.0234375,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 14.6328125,
+      "completions/clipped_ratio": 0.0625,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 16.09375,
       "completions/min_length": 2.0,
-      "epoch": 0.0110803324099723,
-      "grad_norm": 1.5475432762019699,
+      "epoch": 0.011940298507462687,
+      "grad_norm": 3.1591623828063176,
       "kl": 0.0,
       "learning_rate": 2.2222222222222223e-05,
-      "loss": 5.587935447692871e-09,
-      "memory(GiB)": 77.28,
-      "reward": 0.25,
-      "reward_std": 0.7772719860076904,
-      "rewards/Response_no_think_reward/mean": 0.25,
-      "rewards/Response_no_think_reward/std": 1.4795188903808594,
+      "loss": -3.725290298461914e-09,
+      "memory(GiB)": 77.11,
+      "reward": 0.1484375,
+      "reward_std": 0.7272911071777344,
+      "rewards/Response_no_think_reward_1/mean": 0.1484375,
+      "rewards/Response_no_think_reward_1/std": 0.9646495580673218,
       "step": 1,
-      "train_speed(iter/s)": 0.001043
+      "train_speed(iter/s)": 0.001668
     },
     {
       "clip_ratio/high_max": 0.0,
@@ -38,2450 +38,2640 @@
       "clip_ratio/low_mean": 0.0,
       "clip_ratio/low_min": 0.0,
       "clip_ratio/region_mean": 0.0,
-      "epoch": 0.0221606648199446,
-      "grad_norm": 1.552548658770951,
+      "epoch": 0.023880597014925373,
+      "grad_norm": 3.150729002521691,
       "kl": 0.0,
       "learning_rate": 4.4444444444444447e-05,
-      "loss": 5.587935447692871e-09,
-      "memory(GiB)": 78.06,
+      "loss": -3.725290298461914e-09,
+      "memory(GiB)": 77.11,
       "step": 2,
-      "train_speed(iter/s)": 0.001737
+      "train_speed(iter/s)": 0.002544
     },
     {
       "clip_ratio/high_max": 0.0,
       "clip_ratio/high_mean": 0.0,
-      "clip_ratio/low_mean": 0.0,
-      "clip_ratio/low_min": 0.0,
-      "clip_ratio/region_mean": 0.0,
-      "completions/clipped_ratio": 0.03125,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 17.8828125,
-      "completions/min_length": 3.0,
-      "epoch": 0.0332409972299169,
-      "grad_norm": 1.291839688026328,
-      "kl": 0.0012106498143111821,
+      "clip_ratio/low_mean": 0.0028647197177633643,
+      "clip_ratio/low_min": 0.0028647197177633643,
+      "clip_ratio/region_mean": 0.0028647197177633643,
+      "completions/clipped_ratio": 0.0390625,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 14.59375,
+      "completions/min_length": 2.0,
+      "epoch": 0.03582089552238806,
+      "grad_norm": 2.067277005959865,
+      "kl": 0.0022258755707298405,
       "learning_rate": 6.666666666666667e-05,
-      "loss": 0.0004578572406899184,
-      "memory(GiB)": 78.06,
-      "reward": 0.25,
-      "reward_std": 0.8250656127929688,
-      "rewards/Response_no_think_reward/mean": 0.25,
-      "rewards/Response_no_think_reward/std": 1.4634658098220825,
+      "loss": -0.0004270445788279176,
+      "memory(GiB)": 77.3,
+      "reward": -0.3828125,
+      "reward_std": 0.707726001739502,
+      "rewards/Response_no_think_reward_1/mean": -0.3828125,
+      "rewards/Response_no_think_reward_1/std": 0.923201322555542,
       "step": 3,
-      "train_speed(iter/s)": 0.001451
+      "train_speed(iter/s)": 0.002313
     },
     {
-      "clip_ratio/high_max": 0.004197790374746546,
-      "clip_ratio/high_mean": 0.004197790374746546,
-      "clip_ratio/low_mean": 0.012020835361909121,
-      "clip_ratio/low_min": 0.012020835361909121,
-      "clip_ratio/region_mean": 0.01621862585307099,
-      "epoch": 0.0443213296398892,
-      "grad_norm": 1.0301802967661304,
-      "kl": 0.005412253500253428,
+      "clip_ratio/high_max": 0.003551179339410737,
+      "clip_ratio/high_mean": 0.003551179339410737,
+      "clip_ratio/low_mean": 0.025148934859316796,
+      "clip_ratio/low_min": 0.025148934859316796,
+      "clip_ratio/region_mean": 0.028700114460662007,
+      "epoch": 0.04776119402985075,
+      "grad_norm": 1.6109973009110337,
+      "kl": 0.007886729676101822,
       "learning_rate": 8.888888888888889e-05,
-      "loss": -0.005840146914124489,
-      "memory(GiB)": 78.06,
+      "loss": -0.010280434973537922,
+      "memory(GiB)": 77.3,
       "step": 4,
-      "train_speed(iter/s)": 0.001769
+      "train_speed(iter/s)": 0.002696
     },
     {
       "clip_ratio/high_max": 0.0,
       "clip_ratio/high_mean": 0.0,
-      "clip_ratio/low_mean": 0.002237339736893773,
-      "clip_ratio/low_min": 0.002237339736893773,
-      "clip_ratio/region_mean": 0.002237339736893773,
-      "completions/clipped_ratio": 0.0078125,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 12.5546875,
+      "clip_ratio/low_mean": 0.0019339247373864055,
+      "clip_ratio/low_min": 0.0019339247373864055,
+      "clip_ratio/region_mean": 0.0019339247373864055,
+      "completions/clipped_ratio": 0.0390625,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 13.5390625,
       "completions/min_length": 3.0,
-      "epoch": 0.055401662049861494,
-      "grad_norm": 2.027395284218205,
-      "kl": 0.014184385421685874,
+      "epoch": 0.05970149253731343,
+      "grad_norm": 2.893722550507481,
+      "kl": 0.010198775078606559,
       "learning_rate": 0.00011111111111111112,
-      "loss": 0.0008896891959011555,
-      "memory(GiB)": 78.06,
-      "reward": 0.8515625,
-      "reward_std": 0.5118520259857178,
-      "rewards/Response_no_think_reward/mean": 0.8515625,
-      "rewards/Response_no_think_reward/std": 1.4147136211395264,
+      "loss": 0.0004082494997419417,
+      "memory(GiB)": 77.52,
+      "reward": 0.125,
+      "reward_std": 0.493943989276886,
+      "rewards/Response_no_think_reward_1/mean": 0.125,
+      "rewards/Response_no_think_reward_1/std": 0.9881184101104736,
       "step": 5,
-      "train_speed(iter/s)": 0.00164
+      "train_speed(iter/s)": 0.002517
     },
     {
-      "clip_ratio/high_max": 0.013844632252585143,
-      "clip_ratio/high_mean": 0.013844632252585143,
-      "clip_ratio/low_mean": 0.043920744908973575,
-      "clip_ratio/low_min": 0.043920744908973575,
-      "clip_ratio/region_mean": 0.05776537861675024,
-      "epoch": 0.0664819944598338,
-      "grad_norm": 6.212168397356187,
-      "kl": 1.7603500080003869,
+      "clip_ratio/high_max": 0.028469400480389595,
+      "clip_ratio/high_mean": 0.028469400480389595,
+      "clip_ratio/low_mean": 0.04948427155613899,
+      "clip_ratio/low_min": 0.04948427155613899,
+      "clip_ratio/region_mean": 0.07795367110520601,
+      "epoch": 0.07164179104477612,
+      "grad_norm": 1.5585579234945925,
+      "kl": 0.6935355372261256,
       "learning_rate": 0.00013333333333333334,
-      "loss": 0.01842591166496277,
-      "memory(GiB)": 79.98,
+      "loss": -0.005084950476884842,
+      "memory(GiB)": 79.48,
       "step": 6,
-      "train_speed(iter/s)": 0.00185
+      "train_speed(iter/s)": 0.002759
     },
     {
       "clip_ratio/high_max": 0.0,
       "clip_ratio/high_mean": 0.0,
-      "clip_ratio/low_mean": 0.001959657238330692,
-      "clip_ratio/low_min": 0.001959657238330692,
-      "clip_ratio/region_mean": 0.001959657238330692,
-      "completions/clipped_ratio": 0.0,
-      "completions/max_length": 70.0,
-      "completions/mean_length": 10.359375,
-      "completions/min_length": 2.0,
-      "epoch": 0.07756232686980609,
-      "grad_norm": 3.537674599924329,
-      "kl": 0.33545287084416486,
+      "clip_ratio/low_mean": 0.0010160960373468697,
+      "clip_ratio/low_min": 0.0010160960373468697,
+      "clip_ratio/region_mean": 0.0010160960373468697,
+      "completions/clipped_ratio": 0.0078125,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 11.9296875,
+      "completions/min_length": 3.0,
+      "epoch": 0.08358208955223881,
+      "grad_norm": 1.7579177577957044,
+      "kl": 0.07463712914614007,
       "learning_rate": 0.00015555555555555556,
-      "loss": 0.0018048422643914819,
-      "memory(GiB)": 79.98,
-      "reward": 0.53125,
-      "reward_std": 0.6121620535850525,
-      "rewards/Response_no_think_reward/mean": 0.53125,
-      "rewards/Response_no_think_reward/std": 1.3685873746871948,
+      "loss": 0.0008710746187716722,
+      "memory(GiB)": 79.48,
+      "reward": -0.0390625,
+      "reward_std": 0.6340709924697876,
+      "rewards/Response_no_think_reward_1/mean": -0.0390625,
+      "rewards/Response_no_think_reward_1/std": 0.9992307424545288,
       "step": 7,
-      "train_speed(iter/s)": 0.001773
+      "train_speed(iter/s)": 0.002622
     },
     {
-      "clip_ratio/high_max": 0.014042179333046079,
-      "clip_ratio/high_mean": 0.014042179333046079,
-      "clip_ratio/low_mean": 0.03925089433323592,
-      "clip_ratio/low_min": 0.03925089433323592,
-      "clip_ratio/region_mean": 0.053293074015527964,
-      "epoch": 0.0886426592797784,
-      "grad_norm": 2.64201486377639,
-      "kl": 1.295830228133127,
+      "clip_ratio/high_max": 0.02714990673121065,
+      "clip_ratio/high_mean": 0.02714990673121065,
+      "clip_ratio/low_mean": 0.03809016165905632,
+      "clip_ratio/low_min": 0.03809016165905632,
+      "clip_ratio/region_mean": 0.06524006789550185,
+      "epoch": 0.0955223880597015,
+      "grad_norm": 1.2730547516178623,
+      "kl": 0.21840606682235375,
       "learning_rate": 0.00017777777777777779,
-      "loss": 0.011679837480187416,
-      "memory(GiB)": 79.98,
+      "loss": -0.013987341895699501,
+      "memory(GiB)": 79.48,
       "step": 8,
-      "train_speed(iter/s)": 0.001932
+      "train_speed(iter/s)": 0.002802
     },
     {
-      "clip_ratio/high_max": 0.0011001251987181604,
-      "clip_ratio/high_mean": 0.0011001251987181604,
-      "clip_ratio/low_mean": 0.008499634364852682,
-      "clip_ratio/low_min": 0.008499634364852682,
-      "clip_ratio/region_mean": 0.009599759563570842,
-      "completions/clipped_ratio": 0.0,
-      "completions/max_length": 62.0,
-      "completions/mean_length": 11.4375,
-      "completions/min_length": 2.0,
-      "epoch": 0.0997229916897507,
-      "grad_norm": 4.1929387792795065,
-      "kl": 2.5014250652166083,
+      "clip_ratio/high_max": 0.0009191176504828036,
+      "clip_ratio/high_mean": 0.0009191176504828036,
+      "clip_ratio/low_mean": 0.0018382353009656072,
+      "clip_ratio/low_min": 0.0018382353009656072,
+      "clip_ratio/region_mean": 0.0027573530096560717,
+      "completions/clipped_ratio": 0.0078125,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 10.3828125,
+      "completions/min_length": 3.0,
+      "epoch": 0.10746268656716418,
+      "grad_norm": 3.3295900087180645,
+      "kl": 0.35752939723897725,
       "learning_rate": 0.0002,
-      "loss": 0.016260012984275818,
-      "memory(GiB)": 79.98,
-      "reward": 0.2734375,
-      "reward_std": 0.8412302732467651,
-      "rewards/Response_no_think_reward/mean": 0.2734375,
-      "rewards/Response_no_think_reward/std": 1.384474277496338,
+      "loss": 0.003791232593357563,
+      "memory(GiB)": 79.48,
+      "reward": 0.328125,
+      "reward_std": 0.5329009890556335,
+      "rewards/Response_no_think_reward_1/mean": 0.328125,
+      "rewards/Response_no_think_reward_1/std": 0.9483460783958435,
       "step": 9,
-      "train_speed(iter/s)": 0.001887
-    },
-    {
-      "clip_ratio/high_max": 0.02779325417941436,
-      "clip_ratio/high_mean": 0.02779325417941436,
-      "clip_ratio/low_mean": 0.07855538238072768,
-      "clip_ratio/low_min": 0.07855538238072768,
-      "clip_ratio/region_mean": 0.10634863609448075,
-      "epoch": 0.11080332409972299,
-      "grad_norm": 990.7609804346584,
-      "kl": 1.6530181597918272,
-      "learning_rate": 0.00019998312416333227,
-      "loss": 1.336751103401184,
-      "memory(GiB)": 79.98,
+      "train_speed(iter/s)": 0.002657
+    },
+    {
+      "clip_ratio/high_max": 0.03471234819153324,
+      "clip_ratio/high_mean": 0.03471234819153324,
+      "clip_ratio/low_mean": 0.045793990429956466,
+      "clip_ratio/low_min": 0.045793990429956466,
+      "clip_ratio/region_mean": 0.08050633978564292,
+      "epoch": 0.11940298507462686,
+      "grad_norm": 3.037041883710677,
+      "kl": 1.3611202164320275,
+      "learning_rate": 0.00019997998037428526,
+      "loss": -0.0012767184525728226,
+      "memory(GiB)": 79.48,
       "step": 10,
-      "train_speed(iter/s)": 0.002015
+      "train_speed(iter/s)": 0.002798
     },
     {
-      "clip_ratio/high_max": 0.001354054023977369,
-      "clip_ratio/high_mean": 0.001354054023977369,
-      "clip_ratio/low_mean": 0.004174399145995267,
-      "clip_ratio/low_min": 0.004174399145995267,
-      "clip_ratio/region_mean": 0.005528453169972636,
-      "completions/clipped_ratio": 0.0546875,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 26.453125,
-      "completions/min_length": 3.0,
-      "epoch": 0.12188365650969529,
-      "grad_norm": 1.115304569915306,
-      "kl": 0.3547552889212966,
-      "learning_rate": 0.00019993250234920636,
-      "loss": 0.005897670052945614,
-      "memory(GiB)": 79.98,
-      "reward": 0.5234375,
-      "reward_std": 0.876558780670166,
-      "rewards/Response_no_think_reward/mean": 0.5234375,
-      "rewards/Response_no_think_reward/std": 1.4251114130020142,
+      "clip_ratio/high_max": 0.0022767678019590676,
+      "clip_ratio/high_mean": 0.0022767678019590676,
+      "clip_ratio/low_mean": 0.004576534847728908,
+      "clip_ratio/low_min": 0.004576534847728908,
+      "clip_ratio/region_mean": 0.006853302649687976,
+      "completions/clipped_ratio": 0.015625,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 11.3828125,
+      "completions/min_length": 3.0,
+      "epoch": 0.13134328358208955,
+      "grad_norm": 1.1988213220999966,
+      "kl": 0.176690819360374,
+      "learning_rate": 0.00019991992951284932,
+      "loss": 0.0002530772762838751,
+      "memory(GiB)": 79.48,
+      "reward": 0.2890625,
+      "reward_std": 0.36220550537109375,
+      "rewards/Response_no_think_reward_1/mean": 0.2890625,
+      "rewards/Response_no_think_reward_1/std": 0.9487027525901794,
       "step": 11,
-      "train_speed(iter/s)": 0.001964
-    },
-    {
-      "clip_ratio/high_max": 0.02347446102066897,
-      "clip_ratio/high_mean": 0.02347446102066897,
-      "clip_ratio/low_mean": 0.0428259114123648,
-      "clip_ratio/low_min": 0.0428259114123648,
-      "clip_ratio/region_mean": 0.06630037224385887,
-      "epoch": 0.1329639889196676,
-      "grad_norm": 2.183295109715178,
-      "kl": 0.6218942860141397,
-      "learning_rate": 0.00019984815164333163,
-      "loss": 0.007074224762618542,
-      "memory(GiB)": 79.98,
+      "train_speed(iter/s)": 0.002687
+    },
+    {
+      "clip_ratio/high_max": 0.010013980441726744,
+      "clip_ratio/high_mean": 0.010013980441726744,
+      "clip_ratio/low_mean": 0.0431372388266027,
+      "clip_ratio/low_min": 0.0431372388266027,
+      "clip_ratio/region_mean": 0.053151219501160085,
+      "epoch": 0.14328358208955225,
+      "grad_norm": 4.4251813995519464,
+      "kl": 3.680493631065474,
+      "learning_rate": 0.00019981987145960755,
+      "loss": 0.022724991664290428,
+      "memory(GiB)": 79.48,
       "step": 12,
-      "train_speed(iter/s)": 0.00207
+      "train_speed(iter/s)": 0.002804
     },
     {
-      "clip_ratio/high_max": 0.0,
-      "clip_ratio/high_mean": 0.0,
-      "clip_ratio/low_mean": 0.001206563669256866,
-      "clip_ratio/low_min": 0.001206563669256866,
-      "clip_ratio/region_mean": 0.001206563669256866,
-      "completions/clipped_ratio": 0.0,
-      "completions/max_length": 83.0,
-      "completions/mean_length": 14.0234375,
-      "completions/min_length": 4.0,
-      "epoch": 0.1440443213296399,
-      "grad_norm": 1.0424369292135751,
-      "kl": 0.40942037590866676,
-      "learning_rate": 0.00019973010051548275,
-      "loss": 0.004908258095383644,
-      "memory(GiB)": 79.98,
-      "reward": 0.5625,
-      "reward_std": 0.7038782835006714,
-      "rewards/Response_no_think_reward/mean": 0.5625,
-      "rewards/Response_no_think_reward/std": 1.2720495462417603,
+      "clip_ratio/high_max": 0.000480769231216982,
+      "clip_ratio/high_mean": 0.000480769231216982,
+      "clip_ratio/low_mean": 0.0022044407669454813,
+      "clip_ratio/low_min": 0.0022044407669454813,
+      "clip_ratio/region_mean": 0.0026852099981624633,
+      "completions/clipped_ratio": 0.0078125,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 12.0703125,
+      "completions/min_length": 3.0,
+      "epoch": 0.15522388059701492,
+      "grad_norm": 2.0136556426936285,
+      "kl": 0.14468206313904375,
+      "learning_rate": 0.00019967984627705548,
+      "loss": -0.00039299592026509345,
+      "memory(GiB)": 79.48,
+      "reward": 0.0625,
+      "reward_std": 0.6480826735496521,
+      "rewards/Response_no_think_reward_1/mean": 0.0625,
+      "rewards/Response_no_think_reward_1/std": 0.9861242175102234,
       "step": 13,
-      "train_speed(iter/s)": 0.002042
-    },
-    {
-      "clip_ratio/high_max": 0.017414433998055756,
-      "clip_ratio/high_mean": 0.017414433998055756,
-      "clip_ratio/low_mean": 0.028266766399610788,
-      "clip_ratio/low_min": 0.028266766399610788,
-      "clip_ratio/region_mean": 0.04568120092153549,
-      "epoch": 0.15512465373961218,
-      "grad_norm": 0.650161789807741,
-      "kl": 0.4890678570009186,
-      "learning_rate": 0.00019957838880989078,
-      "loss": -0.0075666941702365875,
-      "memory(GiB)": 79.98,
+      "train_speed(iter/s)": 0.00269
+    },
+    {
+      "clip_ratio/high_max": 0.016389093652833253,
+      "clip_ratio/high_mean": 0.016389093652833253,
+      "clip_ratio/low_mean": 0.03954102098941803,
+      "clip_ratio/low_min": 0.03954102098941803,
+      "clip_ratio/region_mean": 0.05593011493328959,
+      "epoch": 0.16716417910447762,
+      "grad_norm": 0.7268697285990734,
+      "kl": 0.15212981263175607,
+      "learning_rate": 0.00019949991003022808,
+      "loss": -0.01738908141851425,
+      "memory(GiB)": 79.48,
       "step": 14,
-      "train_speed(iter/s)": 0.002135
+      "train_speed(iter/s)": 0.002789
     },
     {
-      "clip_ratio/high_max": 0.0018629207770572975,
-      "clip_ratio/high_mean": 0.0018629207770572975,
-      "clip_ratio/low_mean": 0.0069178942940197885,
-      "clip_ratio/low_min": 0.0069178942940197885,
-      "clip_ratio/region_mean": 0.00878081505652517,
-      "completions/clipped_ratio": 0.0078125,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 17.1953125,
-      "completions/min_length": 4.0,
-      "epoch": 0.16620498614958448,
-      "grad_norm": 2.2362713456466605,
-      "kl": 0.23374600778333843,
-      "learning_rate": 0.00019939306773179497,
-      "loss": 0.0009349192259833217,
-      "memory(GiB)": 79.98,
-      "reward": 0.6875,
-      "reward_std": 0.8529120683670044,
-      "rewards/Response_no_think_reward/mean": 0.6875,
-      "rewards/Response_no_think_reward/std": 1.3265905380249023,
+      "clip_ratio/high_max": 0.0005787037080153823,
+      "clip_ratio/high_mean": 0.0005787037080153823,
+      "clip_ratio/low_mean": 0.004868808202445507,
+      "clip_ratio/low_min": 0.004868808202445507,
+      "clip_ratio/region_mean": 0.005447512026876211,
+      "completions/clipped_ratio": 0.015625,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 9.8671875,
+      "completions/min_length": 3.0,
+      "epoch": 0.1791044776119403,
+      "grad_norm": 1.881956635614757,
+      "kl": 0.24015408230479807,
+      "learning_rate": 0.0001992801347642515,
+      "loss": 0.0037227380089461803,
+      "memory(GiB)": 79.48,
+      "reward": 0.0625,
+      "reward_std": 0.15197435021400452,
+      "rewards/Response_no_think_reward_1/mean": 0.0625,
+      "rewards/Response_no_think_reward_1/std": 0.9700231552124023,
       "step": 15,
-      "train_speed(iter/s)": 0.002091
-    },
-    {
-      "clip_ratio/high_max": 0.03511151310522109,
-      "clip_ratio/high_mean": 0.03511151310522109,
-      "clip_ratio/low_mean": 0.028677020454779267,
-      "clip_ratio/low_min": 0.028677020454779267,
-      "clip_ratio/region_mean": 0.06378853344358504,
-      "epoch": 0.1772853185595568,
-      "grad_norm": 3.3548685452904934,
-      "kl": 0.25148704896855634,
-      "learning_rate": 0.00019917419983016025,
-      "loss": 0.017475975677371025,
-      "memory(GiB)": 79.98,
+      "train_speed(iter/s)": 0.002716
+    },
+    {
+      "clip_ratio/high_max": 0.005905566504225135,
+      "clip_ratio/high_mean": 0.005905566504225135,
+      "clip_ratio/low_mean": 0.03072124859318137,
+      "clip_ratio/low_min": 0.03072124859318137,
+      "clip_ratio/region_mean": 0.036626815563067794,
+      "epoch": 0.191044776119403,
+      "grad_norm": 9.037445474392891,
+      "kl": 4.632834411575459,
+      "learning_rate": 0.00019902060847549718,
+      "loss": 0.06257862597703934,
+      "memory(GiB)": 79.48,
       "step": 16,
-      "train_speed(iter/s)": 0.002171
+      "train_speed(iter/s)": 0.002802
     },
     {
-      "clip_ratio/high_max": 0.000469924823846668,
-      "clip_ratio/high_mean": 0.000469924823846668,
-      "clip_ratio/low_mean": 0.0014688223309349269,
-      "clip_ratio/low_min": 0.0014688223309349269,
-      "clip_ratio/region_mean": 0.0019387471547815949,
-      "completions/clipped_ratio": 0.0078125,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 18.2265625,
-      "completions/min_length": 6.0,
-      "epoch": 0.1883656509695291,
-      "grad_norm": 0.6996032916694407,
-      "kl": 0.138063008276049,
-      "learning_rate": 0.00019892185897656578,
-      "loss": 0.0021217623725533485,
-      "memory(GiB)": 79.98,
-      "reward": 0.5,
-      "reward_std": 0.5098158717155457,
-      "rewards/Response_no_think_reward/mean": 0.5,
-      "rewards/Response_no_think_reward/std": 1.4086347818374634,
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.004252503567840904,
+      "clip_ratio/low_min": 0.004252503567840904,
+      "clip_ratio/region_mean": 0.004252503567840904,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 46.0,
+      "completions/mean_length": 7.0,
+      "completions/min_length": 3.0,
+      "epoch": 0.20298507462686566,
+      "grad_norm": 5.483351751991195,
+      "kl": 3.0633848210127326,
+      "learning_rate": 0.0001987214350763483,
+      "loss": 0.026951301842927933,
+      "memory(GiB)": 79.48,
+      "reward": 0.1640625,
+      "reward_std": 0.45976677536964417,
+      "rewards/Response_no_think_reward_1/mean": 0.1640625,
+      "rewards/Response_no_think_reward_1/std": 0.9116684198379517,
       "step": 17,
-      "train_speed(iter/s)": 0.002127
-    },
-    {
-      "clip_ratio/high_max": 0.01048101403284818,
-      "clip_ratio/high_mean": 0.01048101403284818,
-      "clip_ratio/low_mean": 0.028478411200921983,
-      "clip_ratio/low_min": 0.028478411200921983,
-      "clip_ratio/region_mean": 0.03895942587405443,
-      "epoch": 0.1994459833795014,
-      "grad_norm": 0.7437416506383379,
-      "kl": 0.2438321103884391,
-      "learning_rate": 0.00019863613034027224,
-      "loss": -0.007616878021508455,
-      "memory(GiB)": 79.98,
+      "train_speed(iter/s)": 0.002741
+    },
+    {
+      "clip_ratio/high_max": 0.02420679770875722,
+      "clip_ratio/high_mean": 0.02420679770875722,
+      "clip_ratio/low_mean": 0.029018129454925656,
+      "clip_ratio/low_min": 0.029018129454925656,
+      "clip_ratio/region_mean": 0.053224927629344165,
+      "epoch": 0.21492537313432836,
+      "grad_norm": 11.084807824277876,
+      "kl": 4.480629495417816,
+      "learning_rate": 0.00019838273435359444,
+      "loss": 0.03358945623040199,
+      "memory(GiB)": 79.48,
       "step": 18,
-      "train_speed(iter/s)": 0.002199
+      "train_speed(iter/s)": 0.002818
     },
     {
       "clip_ratio/high_max": 0.0,
       "clip_ratio/high_mean": 0.0,
-      "clip_ratio/low_mean": 0.0009733444603625685,
-      "clip_ratio/low_min": 0.0009733444603625685,
-      "clip_ratio/region_mean": 0.0009733444603625685,
+      "clip_ratio/low_mean": 0.009987752127926797,
+      "clip_ratio/low_min": 0.009987752127926797,
+      "clip_ratio/region_mean": 0.009987752127926797,
       "completions/clipped_ratio": 0.0078125,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 20.703125,
-      "completions/min_length": 6.0,
-      "epoch": 0.21052631578947367,
-      "grad_norm": 0.8303775863701255,
-      "kl": 0.09481111937202513,
-      "learning_rate": 0.0001983171103594755,
-      "loss": 0.0022672354243695736,
-      "memory(GiB)": 79.98,
-      "reward": -0.078125,
-      "reward_std": 0.6301807761192322,
-      "rewards/Response_no_think_reward/mean": -0.078125,
-      "rewards/Response_no_think_reward/std": 1.3666982650756836,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 8.6875,
+      "completions/min_length": 3.0,
+      "epoch": 0.22686567164179106,
+      "grad_norm": 2.774235490038128,
+      "kl": 1.3207979609724134,
+      "learning_rate": 0.00019800464192046955,
+      "loss": 0.012463792227208614,
+      "memory(GiB)": 79.48,
+      "reward": -0.0078125,
+      "reward_std": 0.3263596296310425,
+      "rewards/Response_no_think_reward_1/mean": -0.0078125,
+      "rewards/Response_no_think_reward_1/std": 0.9515514969825745,
       "step": 19,
-      "train_speed(iter/s)": 0.002152
-    },
-    {
-      "clip_ratio/high_max": 0.007191620621597394,
-      "clip_ratio/high_mean": 0.007191620621597394,
-      "clip_ratio/low_mean": 0.03550086636096239,
-      "clip_ratio/low_min": 0.03550086636096239,
-      "clip_ratio/region_mean": 0.0426924874773249,
-      "epoch": 0.22160664819944598,
-      "grad_norm": 1.7213808531720889,
-      "kl": 1.6638920252444223,
-      "learning_rate": 0.0001979649067087574,
-      "loss": 0.006861768197268248,
-      "memory(GiB)": 79.98,
+      "train_speed(iter/s)": 0.002746
+    },
+    {
+      "clip_ratio/high_max": 0.013878540194127709,
+      "clip_ratio/high_mean": 0.013878540194127709,
+      "clip_ratio/low_mean": 0.053771398146636784,
+      "clip_ratio/low_min": 0.053771398146636784,
+      "clip_ratio/region_mean": 0.06764993839897215,
+      "epoch": 0.23880597014925373,
+      "grad_norm": 18.987604867751966,
+      "kl": 9.466534710245469,
+      "learning_rate": 0.00019758730916235356,
+      "loss": 0.16672615706920624,
+      "memory(GiB)": 79.48,
       "step": 20,
-      "train_speed(iter/s)": 0.002217
+      "train_speed(iter/s)": 0.002807
     },
     {
-      "clip_ratio/high_max": 0.0004032258002553135,
-      "clip_ratio/high_mean": 0.0004032258002553135,
-      "clip_ratio/low_mean": 0.003614576125983149,
-      "clip_ratio/low_min": 0.003614576125983149,
-      "clip_ratio/region_mean": 0.004017801926238462,
-      "completions/clipped_ratio": 0.0078125,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 17.390625,
-      "completions/min_length": 6.0,
-      "epoch": 0.23268698060941828,
-      "grad_norm": 1.5808946015405074,
-      "kl": 0.1687323283404112,
-      "learning_rate": 0.00019757963826274357,
-      "loss": 0.003855248913168907,
-      "memory(GiB)": 79.98,
-      "reward": 0.5546875,
-      "reward_std": 0.8033354878425598,
-      "rewards/Response_no_think_reward/mean": 0.5546875,
-      "rewards/Response_no_think_reward/std": 1.3328590393066406,
+      "clip_ratio/high_max": 0.005019778618589044,
+      "clip_ratio/high_mean": 0.005019778618589044,
+      "clip_ratio/low_mean": 0.00046641789958812296,
+      "clip_ratio/low_min": 0.00046641789958812296,
+      "clip_ratio/region_mean": 0.005486196518177167,
+      "completions/clipped_ratio": 0.015625,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 8.234375,
+      "completions/min_length": 3.0,
+      "epoch": 0.2507462686567164,
+      "grad_norm": 3.1541164754847397,
+      "kl": 0.4539797595352866,
+      "learning_rate": 0.00019713090317615876,
+      "loss": 0.0023644084576517344,
+      "memory(GiB)": 79.48,
+      "reward": -0.1015625,
+      "reward_std": 0.3404619097709656,
+      "rewards/Response_no_think_reward_1/mean": -0.1015625,
+      "rewards/Response_no_think_reward_1/std": 0.9788298010826111,
       "step": 21,
-      "train_speed(iter/s)": 0.002189
-    },
-    {
-      "clip_ratio/high_max": 0.012200821249280125,
-      "clip_ratio/high_mean": 0.012200821249280125,
-      "clip_ratio/low_mean": 0.04007338697556406,
-      "clip_ratio/low_min": 0.04007338697556406,
-      "clip_ratio/region_mean": 0.05227420857409015,
-      "epoch": 0.24376731301939059,
-      "grad_norm": 0.8850881940917656,
-      "kl": 0.48105100472457707,
-      "learning_rate": 0.0001971614350559814,
-      "loss": -0.004205920733511448,
-      "memory(GiB)": 79.98,
+      "train_speed(iter/s)": 0.002748
+    },
+    {
+      "clip_ratio/high_max": 0.02849817555397749,
+      "clip_ratio/high_mean": 0.02849817555397749,
+      "clip_ratio/low_mean": 0.02262113688630052,
+      "clip_ratio/low_min": 0.02262113688630052,
+      "clip_ratio/region_mean": 0.051119312876835465,
+      "epoch": 0.2626865671641791,
+      "grad_norm": 3.225175633228704,
+      "kl": 0.27581185393501073,
+      "learning_rate": 0.00019663560670342558,
+      "loss": 0.0023057509679347277,
+      "memory(GiB)": 79.48,
       "step": 22,
-      "train_speed(iter/s)": 0.002248
+      "train_speed(iter/s)": 0.002811
     },
     {
-      "clip_ratio/high_max": 0.0008194574620574713,
-      "clip_ratio/high_mean": 0.0008194574620574713,
-      "clip_ratio/low_mean": 0.000811688310932368,
-      "clip_ratio/low_min": 0.000811688310932368,
-      "clip_ratio/region_mean": 0.0016311457729898393,
-      "completions/clipped_ratio": 0.0078125,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 15.28125,
-      "completions/min_length": 6.0,
-      "epoch": 0.2548476454293629,
-      "grad_norm": 0.6954813297644332,
-      "kl": 0.11993603070732206,
-      "learning_rate": 0.0001967104382390511,
-      "loss": 0.001324990182183683,
-      "memory(GiB)": 79.98,
-      "reward": 0.78125,
-      "reward_std": 0.5077463984489441,
-      "rewards/Response_no_think_reward/mean": 0.78125,
-      "rewards/Response_no_think_reward/std": 1.3970582485198975,
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0015432098880410194,
+      "clip_ratio/low_min": 0.0015432098880410194,
+      "clip_ratio/region_mean": 0.0015432098880410194,
+      "completions/clipped_ratio": 0.03125,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 9.1015625,
+      "completions/min_length": 3.0,
+      "epoch": 0.2746268656716418,
+      "grad_norm": 1.602042628934705,
+      "kl": 0.223259057267569,
+      "learning_rate": 0.00019610161805715397,
+      "loss": 0.0012852977961301804,
+      "memory(GiB)": 79.48,
+      "reward": 0.25,
+      "reward_std": 0.45234403014183044,
+      "rewards/Response_no_think_reward_1/mean": 0.25,
+      "rewards/Response_no_think_reward_1/std": 0.9474374651908875,
       "step": 23,
-      "train_speed(iter/s)": 0.002219
-    },
-    {
-      "clip_ratio/high_max": 0.014543175988364965,
-      "clip_ratio/high_mean": 0.014543175988364965,
-      "clip_ratio/low_mean": 0.027941336738877,
-      "clip_ratio/low_min": 0.027941336738877,
-      "clip_ratio/region_mean": 0.04248451231978834,
-      "epoch": 0.2659279778393352,
-      "grad_norm": 0.46891914008721675,
-      "kl": 0.11759324668673798,
-      "learning_rate": 0.00019622680003092503,
-      "loss": -0.009693928062915802,
-      "memory(GiB)": 79.98,
+      "train_speed(iter/s)": 0.002763
+    },
+    {
+      "clip_ratio/high_max": 0.02699369314359501,
+      "clip_ratio/high_mean": 0.02699369314359501,
+      "clip_ratio/low_mean": 0.024013921502046287,
+      "clip_ratio/low_min": 0.024013921502046287,
+      "clip_ratio/region_mean": 0.05100761540234089,
+      "epoch": 0.2865671641791045,
+      "grad_norm": 1.2163889618122583,
+      "kl": 0.48499861752497964,
+      "learning_rate": 0.00019552915104240065,
+      "loss": -0.007453325670212507,
+      "memory(GiB)": 79.48,
       "step": 24,
-      "train_speed(iter/s)": 0.002273
+      "train_speed(iter/s)": 0.002821
     },
     {
-      "clip_ratio/high_max": 0.0006530559621751308,
-      "clip_ratio/high_mean": 0.0006530559621751308,
-      "clip_ratio/low_mean": 0.0019866162620019168,
-      "clip_ratio/low_min": 0.0019866162620019168,
-      "clip_ratio/region_mean": 0.002639672253280878,
-      "completions/clipped_ratio": 0.0078125,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 12.9609375,
-      "completions/min_length": 4.0,
-      "epoch": 0.2770083102493075,
-      "grad_norm": 1.333182473379268,
-      "kl": 0.22027045290451497,
-      "learning_rate": 0.00019571068366759143,
-      "loss": 0.0026610023342072964,
-      "memory(GiB)": 79.98,
-      "reward": 0.4765625,
-      "reward_std": 0.8257243633270264,
-      "rewards/Response_no_think_reward/mean": 0.4765625,
-      "rewards/Response_no_think_reward/std": 1.3219220638275146,
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.001253594527952373,
+      "clip_ratio/low_min": 0.001253594527952373,
+      "clip_ratio/region_mean": 0.001253594527952373,
+      "completions/clipped_ratio": 0.0625,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 9.9375,
+      "completions/min_length": 3.0,
+      "epoch": 0.29850746268656714,
+      "grad_norm": 1.3884331155836789,
+      "kl": 0.5089308844180778,
+      "learning_rate": 0.00019491843487067306,
+      "loss": 0.00490172766149044,
+      "memory(GiB)": 79.48,
+      "reward": 0.2265625,
+      "reward_std": 0.46621453762054443,
+      "rewards/Response_no_think_reward_1/mean": 0.2265625,
+      "rewards/Response_no_think_reward_1/std": 0.9324832558631897,
       "step": 25,
-      "train_speed(iter/s)": 0.002248
-    },
-    {
-      "clip_ratio/high_max": 0.02311275008833036,
-      "clip_ratio/high_mean": 0.02311275008833036,
-      "clip_ratio/low_mean": 0.03803046100074425,
-      "clip_ratio/low_min": 0.03803046100074425,
-      "clip_ratio/region_mean": 0.06114321056520566,
-      "epoch": 0.2880886426592798,
-      "grad_norm": 0.7053648057917502,
-      "kl": 0.2949459235333052,
-      "learning_rate": 0.0001951622633469592,
-      "loss": -0.015641074627637863,
-      "memory(GiB)": 79.98,
+      "train_speed(iter/s)": 0.002775
+    },
+    {
+      "clip_ratio/high_max": 0.01939821511041373,
+      "clip_ratio/high_mean": 0.01939821511041373,
+      "clip_ratio/low_mean": 0.046715788485016674,
+      "clip_ratio/low_min": 0.046715788485016674,
+      "clip_ratio/region_mean": 0.06611400278052315,
+      "epoch": 0.31044776119402984,
+      "grad_norm": 7.610505518747904,
+      "kl": 2.1003341920441017,
+      "learning_rate": 0.00019426971406815463,
+      "loss": 0.024694038555026054,
+      "memory(GiB)": 79.48,
       "step": 26,
-      "train_speed(iter/s)": 0.002297
+      "train_speed(iter/s)": 0.002827
     },
     {
-      "clip_ratio/high_max": 0.0004111842135898769,
-      "clip_ratio/high_mean": 0.0004111842135898769,
-      "clip_ratio/low_mean": 0.0027908546617254615,
-      "clip_ratio/low_min": 0.0027908546617254615,
-      "clip_ratio/region_mean": 0.0032020388753153384,
-      "completions/clipped_ratio": 0.0,
-      "completions/max_length": 85.0,
-      "completions/mean_length": 16.875,
-      "completions/min_length": 4.0,
-      "epoch": 0.29916897506925205,
-      "grad_norm": 1.544635006921068,
-      "kl": 0.20233443519100547,
-      "learning_rate": 0.00019458172417006347,
-      "loss": 0.0030936466064304113,
-      "memory(GiB)": 79.98,
-      "reward": 0.6875,
-      "reward_std": 0.9086803197860718,
-      "rewards/Response_no_think_reward/mean": 0.6875,
-      "rewards/Response_no_think_reward/std": 1.3265905380249023,
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0030820727697573602,
+      "clip_ratio/low_min": 0.0030820727697573602,
+      "clip_ratio/region_mean": 0.0030820727697573602,
+      "completions/clipped_ratio": 0.03125,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 9.2265625,
+      "completions/min_length": 3.0,
+      "epoch": 0.32238805970149254,
+      "grad_norm": 2.593834753837433,
+      "kl": 0.6449481542804278,
+      "learning_rate": 0.00019358324837779863,
+      "loss": 0.007649811916053295,
+      "memory(GiB)": 79.48,
+      "reward": 0.2890625,
+      "reward_std": 0.6288036704063416,
+      "rewards/Response_no_think_reward_1/mean": 0.2890625,
+      "rewards/Response_no_think_reward_1/std": 0.870806872844696,
       "step": 27,
-      "train_speed(iter/s)": 0.002272
-    },
-    {
-      "clip_ratio/high_max": 0.03639351949095726,
-      "clip_ratio/high_mean": 0.03639351949095726,
-      "clip_ratio/low_mean": 0.024818695266731083,
-      "clip_ratio/low_min": 0.024818695266731083,
-      "clip_ratio/region_mean": 0.06121221440844238,
-      "epoch": 0.31024930747922436,
-      "grad_norm": 2.0012174050572105,
-      "kl": 0.24312824057415128,
-      "learning_rate": 0.00019396926207859084,
-      "loss": -0.008060472086071968,
-      "memory(GiB)": 79.98,
+      "train_speed(iter/s)": 0.00279
+    },
+    {
+      "clip_ratio/high_max": 0.02224817470414564,
+      "clip_ratio/high_mean": 0.02224817470414564,
+      "clip_ratio/low_mean": 0.03963817795738578,
+      "clip_ratio/low_min": 0.03963817795738578,
+      "clip_ratio/region_mean": 0.061886353767476976,
+      "epoch": 0.33432835820895523,
+      "grad_norm": 4.001134871741585,
+      "kl": 2.057775880966801,
+      "learning_rate": 0.00019285931265532871,
+      "loss": 0.024306027218699455,
+      "memory(GiB)": 79.48,
       "step": 28,
-      "train_speed(iter/s)": 0.002319
+      "train_speed(iter/s)": 0.002839
     },
     {
-      "clip_ratio/high_max": 0.0007382866751868278,
-      "clip_ratio/high_mean": 0.0007382866751868278,
-      "clip_ratio/low_mean": 0.0024169938114937395,
-      "clip_ratio/low_min": 0.0024169938114937395,
-      "clip_ratio/region_mean": 0.0031552804866805673,
-      "completions/clipped_ratio": 0.0234375,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 23.4140625,
-      "completions/min_length": 3.0,
-      "epoch": 0.32132963988919666,
-      "grad_norm": 1.0333627248580772,
-      "kl": 0.19911292963661253,
-      "learning_rate": 0.0001933250837887457,
-      "loss": 0.004103388637304306,
-      "memory(GiB)": 79.98,
-      "reward": 0.6640625,
-      "reward_std": 0.7995060086250305,
-      "rewards/Response_no_think_reward/mean": 0.6640625,
-      "rewards/Response_no_think_reward/std": 1.4323447942733765,
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0025941890198737383,
+      "clip_ratio/low_min": 0.0025941890198737383,
+      "clip_ratio/region_mean": 0.0025941890198737383,
+      "completions/clipped_ratio": 0.015625,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 7.6640625,
+      "completions/min_length": 3.0,
+      "epoch": 0.34626865671641793,
+      "grad_norm": 2.4322431330202017,
+      "kl": 0.9759200520347804,
+      "learning_rate": 0.0001920981967591891,
+      "loss": 0.009954690001904964,
+      "memory(GiB)": 79.48,
+      "reward": 0.234375,
+      "reward_std": 0.5239416360855103,
+      "rewards/Response_no_think_reward_1/mean": 0.234375,
+      "rewards/Response_no_think_reward_1/std": 0.8179518580436707,
       "step": 29,
-      "train_speed(iter/s)": 0.002285
-    },
-    {
-      "clip_ratio/high_max": 0.021052728639915586,
-      "clip_ratio/high_mean": 0.021052728639915586,
-      "clip_ratio/low_mean": 0.03886253567179665,
-      "clip_ratio/low_min": 0.03886253567179665,
-      "clip_ratio/region_mean": 0.05991526402067393,
-      "epoch": 0.33240997229916897,
-      "grad_norm": 0.4744156832935017,
-      "kl": 0.2653088476508856,
-      "learning_rate": 0.00019264940672148018,
-      "loss": -0.014724130742251873,
-      "memory(GiB)": 79.98,
+      "train_speed(iter/s)": 0.002811
+    },
+    {
+      "clip_ratio/high_max": 0.05181985488161445,
+      "clip_ratio/high_mean": 0.05181985488161445,
+      "clip_ratio/low_mean": 0.03574432339519262,
+      "clip_ratio/low_min": 0.03574432339519262,
+      "clip_ratio/region_mean": 0.0875641773454845,
+      "epoch": 0.3582089552238806,
+      "grad_norm": 2.469531358214721,
+      "kl": 0.3759418617701158,
+      "learning_rate": 0.00019130020543448704,
+      "loss": 0.030206725001335144,
+      "memory(GiB)": 79.48,
       "step": 30,
-      "train_speed(iter/s)": 0.002328
+      "train_speed(iter/s)": 0.002857
     },
     {
-      "clip_ratio/high_max": 0.0008336337341461331,
-      "clip_ratio/high_mean": 0.0008336337341461331,
-      "clip_ratio/low_mean": 0.0028033541166223586,
-      "clip_ratio/low_min": 0.0028033541166223586,
-      "clip_ratio/region_mean": 0.0036369878507684916,
-      "completions/clipped_ratio": 0.0078125,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 17.8125,
-      "completions/min_length": 3.0,
-      "epoch": 0.34349030470914127,
-      "grad_norm": 1.3971369328491496,
-      "kl": 0.3057649952825159,
-      "learning_rate": 0.0001919424589291108,
-      "loss": 0.0056169466115534306,
-      "memory(GiB)": 80.03,
-      "reward": 0.4296875,
-      "reward_std": 0.8740850687026978,
-      "rewards/Response_no_think_reward/mean": 0.4296875,
-      "rewards/Response_no_think_reward/std": 1.3555577993392944,
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0007102272938936949,
+      "clip_ratio/low_min": 0.0007102272938936949,
+      "clip_ratio/region_mean": 0.0007102272938936949,
+      "completions/clipped_ratio": 0.015625,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 7.4921875,
+      "completions/min_length": 3.0,
+      "epoch": 0.3701492537313433,
+      "grad_norm": 25.502961039749618,
+      "kl": 0.3221544102343614,
+      "learning_rate": 0.00019046565819097545,
+      "loss": 0.0061463117599487305,
+      "memory(GiB)": 79.48,
+      "reward": 0.265625,
+      "reward_std": 0.3810158967971802,
+      "rewards/Response_no_think_reward_1/mean": 0.265625,
+      "rewards/Response_no_think_reward_1/std": 0.8082680702209473,
       "step": 31,
-      "train_speed(iter/s)": 0.002304
-    },
-    {
-      "clip_ratio/high_max": 0.013716876972466707,
-      "clip_ratio/high_mean": 0.013716876972466707,
-      "clip_ratio/low_mean": 0.05669466912513599,
-      "clip_ratio/low_min": 0.05669466912513599,
-      "clip_ratio/region_mean": 0.07041154580656439,
-      "epoch": 0.3545706371191136,
-      "grad_norm": 0.7437646652660285,
-      "kl": 0.3698675720952451,
-      "learning_rate": 0.00019120447901834706,
-      "loss": -0.013283709064126015,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002819
+    },
+    {
+      "clip_ratio/high_max": 0.0004340277810115367,
+      "clip_ratio/high_mean": 0.0004340277810115367,
+      "clip_ratio/low_mean": 0.015644782513845712,
+      "clip_ratio/low_min": 0.015644782513845712,
+      "clip_ratio/region_mean": 0.01607881032396108,
+      "epoch": 0.382089552238806,
+      "grad_norm": 2.386529018247983,
+      "kl": 0.4242442559116171,
+      "learning_rate": 0.0001895948891751234,
+      "loss": 0.005282086320221424,
+      "memory(GiB)": 79.48,
       "step": 32,
-      "train_speed(iter/s)": 0.002345
+      "train_speed(iter/s)": 0.002862
     },
     {
-      "clip_ratio/high_max": 0.0013764221512246877,
-      "clip_ratio/high_mean": 0.0013764221512246877,
-      "clip_ratio/low_mean": 0.0013627433363581076,
-      "clip_ratio/low_min": 0.0013627433363581076,
-      "clip_ratio/region_mean": 0.0027391654875827953,
-      "completions/clipped_ratio": 0.0625,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 24.609375,
-      "completions/min_length": 4.0,
-      "epoch": 0.3656509695290859,
-      "grad_norm": 0.8795930368140337,
-      "kl": 0.22658177139237523,
-      "learning_rate": 0.00019043571606975777,
-      "loss": 0.0019196830689907074,
-      "memory(GiB)": 80.03,
-      "reward": 0.4140625,
-      "reward_std": 0.8300054669380188,
-      "rewards/Response_no_think_reward/mean": 0.4140625,
-      "rewards/Response_no_think_reward/std": 1.258216142654419,
+      "clip_ratio/high_max": 0.0008561643771827221,
+      "clip_ratio/high_mean": 0.0008561643771827221,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0008561643771827221,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 36.0,
+      "completions/mean_length": 6.8984375,
+      "completions/min_length": 3.0,
+      "epoch": 0.3940298507462687,
+      "grad_norm": 2.313143797352374,
+      "kl": 1.4197996056172997,
+      "learning_rate": 0.00018868824703632657,
+      "loss": 0.011696252971887589,
+      "memory(GiB)": 79.48,
+      "reward": 0.046875,
+      "reward_std": 0.6391555070877075,
+      "rewards/Response_no_think_reward_1/mean": 0.046875,
+      "rewards/Response_no_think_reward_1/std": 0.8215538859367371,
       "step": 33,
-      "train_speed(iter/s)": 0.002323
-    },
-    {
-      "clip_ratio/high_max": 0.016442283987998962,
-      "clip_ratio/high_mean": 0.016442283987998962,
-      "clip_ratio/low_mean": 0.05633212422253564,
-      "clip_ratio/low_min": 0.05633212422253564,
-      "clip_ratio/region_mean": 0.07277440826874226,
-      "epoch": 0.3767313019390582,
-      "grad_norm": 0.5188690402432328,
-      "kl": 0.23683911142870784,
-      "learning_rate": 0.00018963642955370201,
-      "loss": -0.016352392733097076,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002823
+    },
+    {
+      "clip_ratio/high_max": 0.03243769891560078,
+      "clip_ratio/high_mean": 0.03243769891560078,
+      "clip_ratio/low_mean": 0.057673962553963065,
+      "clip_ratio/low_min": 0.057673962553963065,
+      "clip_ratio/region_mean": 0.09011166216805577,
+      "epoch": 0.4059701492537313,
+      "grad_norm": 1.7539402550269323,
+      "kl": 1.9242291511036456,
+      "learning_rate": 0.00018774609478731046,
+      "loss": -0.004083915147930384,
+      "memory(GiB)": 79.48,
       "step": 34,
-      "train_speed(iter/s)": 0.002362
+      "train_speed(iter/s)": 0.002863
     },
     {
-      "clip_ratio/high_max": 0.0025397460121894255,
-      "clip_ratio/high_mean": 0.0025397460121894255,
-      "clip_ratio/low_mean": 0.00378477135382127,
-      "clip_ratio/low_min": 0.00378477135382127,
-      "clip_ratio/region_mean": 0.006324517366010696,
-      "completions/clipped_ratio": 0.015625,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 23.375,
-      "completions/min_length": 3.0,
-      "epoch": 0.3878116343490305,
-      "grad_norm": 0.7751350107093148,
-      "kl": 0.21053988160565495,
-      "learning_rate": 0.00018880688924275378,
-      "loss": 0.001979985274374485,
-      "memory(GiB)": 80.03,
-      "reward": 0.25,
-      "reward_std": 0.8051205277442932,
-      "rewards/Response_no_think_reward/mean": 0.25,
-      "rewards/Response_no_think_reward/std": 1.3101662397384644,
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0010245901066809893,
+      "clip_ratio/low_min": 0.0010245901066809893,
+      "clip_ratio/region_mean": 0.0010245901066809893,
+      "completions/clipped_ratio": 0.0078125,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 7.7109375,
+      "completions/min_length": 3.0,
+      "epoch": 0.417910447761194,
+      "grad_norm": 2.4226244872635485,
+      "kl": 1.010538819245994,
+      "learning_rate": 0.00018676880965878291,
+      "loss": 0.015125149860978127,
+      "memory(GiB)": 79.48,
+      "reward": 0.0546875,
+      "reward_std": 0.5867211818695068,
+      "rewards/Response_no_think_reward_1/mean": 0.0546875,
+      "rewards/Response_no_think_reward_1/std": 0.8539726138114929,
       "step": 35,
-      "train_speed(iter/s)": 0.00234
-    },
-    {
-      "clip_ratio/high_max": 0.02383261898648925,
-      "clip_ratio/high_mean": 0.02383261898648925,
-      "clip_ratio/low_mean": 0.02296329999808222,
-      "clip_ratio/low_min": 0.02296329999808222,
-      "clip_ratio/region_mean": 0.046795917907729745,
-      "epoch": 0.3988919667590028,
-      "grad_norm": 0.6919043847160915,
-      "kl": 0.2092050458304584,
-      "learning_rate": 0.0001879473751206489,
-      "loss": -0.016633104532957077,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.00283
+    },
+    {
+      "clip_ratio/high_max": 0.030840664519928396,
+      "clip_ratio/high_mean": 0.030840664519928396,
+      "clip_ratio/low_mean": 0.06221208241186105,
+      "clip_ratio/low_min": 0.06221208241186105,
+      "clip_ratio/region_mean": 0.09305274731013924,
+      "epoch": 0.4298507462686567,
+      "grad_norm": 2.284561806665497,
+      "kl": 2.4607611764222383,
+      "learning_rate": 0.00018575678294839373,
+      "loss": 0.010896118357777596,
+      "memory(GiB)": 79.48,
       "step": 36,
-      "train_speed(iter/s)": 0.002376
+      "train_speed(iter/s)": 0.002867
     },
     {
-      "clip_ratio/high_max": 0.0,
-      "clip_ratio/high_mean": 0.0,
-      "clip_ratio/low_mean": 0.0003396739193703979,
-      "clip_ratio/low_min": 0.0003396739193703979,
-      "clip_ratio/region_mean": 0.0003396739193703979,
-      "completions/clipped_ratio": 0.0078125,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 16.8828125,
-      "completions/min_length": 3.0,
-      "epoch": 0.4099722991689751,
-      "grad_norm": 1.0548464052821087,
-      "kl": 0.20961805986007676,
-      "learning_rate": 0.00018705817728778624,
-      "loss": 0.003657686058431864,
-      "memory(GiB)": 80.03,
-      "reward": 0.953125,
-      "reward_std": 0.6187193393707275,
-      "rewards/Response_no_think_reward/mean": 0.953125,
-      "rewards/Response_no_think_reward/std": 1.1961840391159058,
+      "clip_ratio/high_max": 0.001179245300590992,
+      "clip_ratio/high_mean": 0.001179245300590992,
+      "clip_ratio/low_mean": 0.00924376118928194,
+      "clip_ratio/low_min": 0.00924376118928194,
+      "clip_ratio/region_mean": 0.010423006489872932,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 27.0,
+      "completions/mean_length": 5.328125,
+      "completions/min_length": 2.0,
+      "epoch": 0.4417910447761194,
+      "grad_norm": 3.3426494147901047,
+      "kl": 0.3604982797114644,
+      "learning_rate": 0.000184710419864062,
+      "loss": 0.0029986475128680468,
+      "memory(GiB)": 79.48,
+      "reward": 0.328125,
+      "reward_std": 0.34082794189453125,
+      "rewards/Response_no_think_reward_1/mean": 0.328125,
+      "rewards/Response_no_think_reward_1/std": 0.8794179558753967,
       "step": 37,
-      "train_speed(iter/s)": 0.002358
-    },
-    {
-      "clip_ratio/high_max": 0.02387295541120693,
-      "clip_ratio/high_mean": 0.02387295541120693,
-      "clip_ratio/low_mean": 0.030282753868959844,
-      "clip_ratio/low_min": 0.030282753868959844,
-      "clip_ratio/region_mean": 0.05415570852346718,
-      "epoch": 0.42105263157894735,
-      "grad_norm": 0.49371052672691246,
-      "kl": 0.24616074899677187,
-      "learning_rate": 0.00018613959586331362,
-      "loss": -0.012075964361429214,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002821
+    },
+    {
+      "clip_ratio/high_max": 0.011844758177176118,
+      "clip_ratio/high_mean": 0.011844758177176118,
+      "clip_ratio/low_mean": 0.05284481216222048,
+      "clip_ratio/low_min": 0.05284481216222048,
+      "clip_ratio/region_mean": 0.06468957080505788,
+      "epoch": 0.4537313432835821,
+      "grad_norm": 0.9022263920887504,
+      "kl": 0.5205309305310948,
+      "learning_rate": 0.00018363013936173393,
+      "loss": -0.011295529082417488,
+      "memory(GiB)": 79.48,
       "step": 38,
-      "train_speed(iter/s)": 0.002393
+      "train_speed(iter/s)": 0.002857
     },
     {
-      "clip_ratio/high_max": 0.0013275333330966532,
-      "clip_ratio/high_mean": 0.0013275333330966532,
-      "clip_ratio/low_mean": 0.0013904034567531198,
-      "clip_ratio/low_min": 0.0013904034567531198,
-      "clip_ratio/region_mean": 0.002717936789849773,
-      "completions/clipped_ratio": 0.03125,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 21.6015625,
-      "completions/min_length": 4.0,
-      "epoch": 0.43213296398891965,
-      "grad_norm": 0.972842541028031,
-      "kl": 0.16011726018041372,
-      "learning_rate": 0.00018519194088383273,
-      "loss": 0.001440724590793252,
-      "memory(GiB)": 80.03,
-      "reward": 0.625,
-      "reward_std": 0.6670520305633545,
-      "rewards/Response_no_think_reward/mean": 0.625,
-      "rewards/Response_no_think_reward/std": 1.3457428216934204,
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.0078125,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 5.328125,
+      "completions/min_length": 3.0,
+      "epoch": 0.46567164179104475,
+      "grad_norm": 1.5271115661229067,
+      "kl": 0.27687166831583454,
+      "learning_rate": 0.00018251637397763597,
+      "loss": 0.0014096113154664636,
+      "memory(GiB)": 79.48,
+      "reward": 0.046875,
+      "reward_std": 0.3280481696128845,
+      "rewards/Response_no_think_reward_1/mean": 0.046875,
+      "rewards/Response_no_think_reward_1/std": 0.9379100203514099,
       "step": 39,
-      "train_speed(iter/s)": 0.002371
-    },
-    {
-      "clip_ratio/high_max": 0.03226059180451557,
-      "clip_ratio/high_mean": 0.03226059180451557,
-      "clip_ratio/low_mean": 0.03314233338460326,
-      "clip_ratio/low_min": 0.03314233338460326,
-      "clip_ratio/region_mean": 0.0654029252473265,
-      "epoch": 0.44321329639889195,
-      "grad_norm": 0.5253408277855696,
-      "kl": 0.18682625680230558,
-      "learning_rate": 0.00018421553219875658,
-      "loss": -0.013099671341478825,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002785
+    },
+    {
+      "clip_ratio/high_max": 0.04636762477457523,
+      "clip_ratio/high_mean": 0.04636762477457523,
+      "clip_ratio/low_mean": 0.02858343638945371,
+      "clip_ratio/low_min": 0.02858343638945371,
+      "clip_ratio/region_mean": 0.07495106221176684,
+      "epoch": 0.47761194029850745,
+      "grad_norm": 0.6061436422525854,
+      "kl": 0.7858691609289963,
+      "learning_rate": 0.00018136956965509064,
+      "loss": -0.008444712497293949,
+      "memory(GiB)": 79.48,
       "step": 40,
-      "train_speed(iter/s)": 0.002404
+      "train_speed(iter/s)": 0.002819
     },
     {
-      "clip_ratio/high_max": 0.00029620854184031487,
-      "clip_ratio/high_mean": 0.00029620854184031487,
-      "clip_ratio/low_mean": 0.0009311849280493334,
-      "clip_ratio/low_min": 0.0009311849280493334,
-      "clip_ratio/region_mean": 0.0012273934698896483,
-      "completions/clipped_ratio": 0.0,
-      "completions/max_length": 88.0,
-      "completions/mean_length": 17.5625,
-      "completions/min_length": 5.0,
-      "epoch": 0.45429362880886426,
-      "grad_norm": 1.2919745010410586,
-      "kl": 0.2366366102360189,
-      "learning_rate": 0.00018321069936235503,
-      "loss": 0.001985038397833705,
-      "memory(GiB)": 80.03,
-      "reward": 0.859375,
-      "reward_std": 0.6107450723648071,
-      "rewards/Response_no_think_reward/mean": 0.859375,
-      "rewards/Response_no_think_reward/std": 1.4071491956710815,
+      "clip_ratio/high_max": 0.001953125,
+      "clip_ratio/high_mean": 0.001953125,
+      "clip_ratio/low_mean": 0.0004006410308647901,
+      "clip_ratio/low_min": 0.0004006410308647901,
+      "clip_ratio/region_mean": 0.00235376603086479,
+      "completions/clipped_ratio": 0.0390625,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 8.296875,
+      "completions/min_length": 3.0,
+      "epoch": 0.48955223880597015,
+      "grad_norm": 2.15651488822414,
+      "kl": 0.5289879650626972,
+      "learning_rate": 0.000180190185565964,
+      "loss": 0.006278940010815859,
+      "memory(GiB)": 79.48,
+      "reward": 0.140625,
+      "reward_std": 0.3606899678707123,
+      "rewards/Response_no_think_reward_1/mean": 0.140625,
+      "rewards/Response_no_think_reward_1/std": 0.9284173250198364,
       "step": 41,
-      "train_speed(iter/s)": 0.002379
-    },
-    {
-      "clip_ratio/high_max": 0.01783788768807426,
-      "clip_ratio/high_mean": 0.01783788768807426,
-      "clip_ratio/low_mean": 0.044289187353570014,
-      "clip_ratio/low_min": 0.044289187353570014,
-      "clip_ratio/region_mean": 0.06212707597296685,
-      "epoch": 0.46537396121883656,
-      "grad_norm": 1.7719321167278534,
-      "kl": 0.6719344789162278,
-      "learning_rate": 0.0001821777815225245,
-      "loss": -0.008167732506990433,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002732
+    },
+    {
+      "clip_ratio/high_max": 0.019255130202509463,
+      "clip_ratio/high_mean": 0.019255130202509463,
+      "clip_ratio/low_mean": 0.049764272407628596,
+      "clip_ratio/low_min": 0.049764272407628596,
+      "clip_ratio/region_mean": 0.06901940272655338,
+      "epoch": 0.5014925373134328,
+      "grad_norm": 7.678211490755348,
+      "kl": 4.704732097394299,
+      "learning_rate": 0.00017897869392681685,
+      "loss": 0.036949772387742996,
+      "memory(GiB)": 79.48,
       "step": 42,
-      "train_speed(iter/s)": 0.00241
+      "train_speed(iter/s)": 0.002764
     },
     {
-      "clip_ratio/high_max": 0.00036231885314919055,
-      "clip_ratio/high_mean": 0.00036231885314919055,
-      "clip_ratio/low_mean": 0.002238474931800738,
-      "clip_ratio/low_min": 0.002238474931800738,
-      "clip_ratio/region_mean": 0.0026007937849499285,
-      "completions/clipped_ratio": 0.0,
-      "completions/max_length": 97.0,
-      "completions/mean_length": 15.875,
-      "completions/min_length": 6.0,
-      "epoch": 0.47645429362880887,
-      "grad_norm": 1.1682612472028975,
-      "kl": 0.2416230053640902,
-      "learning_rate": 0.00018111712730632022,
-      "loss": 0.0018095734994858503,
-      "memory(GiB)": 80.03,
-      "reward": 0.0859375,
-      "reward_std": 0.39637458324432373,
-      "rewards/Response_no_think_reward/mean": 0.0859375,
-      "rewards/Response_no_think_reward/std": 0.9962713122367859,
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0018887363257817924,
+      "clip_ratio/low_min": 0.0018887363257817924,
+      "clip_ratio/region_mean": 0.0018887363257817924,
+      "completions/clipped_ratio": 0.03125,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 10.0390625,
+      "completions/min_length": 3.0,
+      "epoch": 0.5134328358208955,
+      "grad_norm": 1.5758982248595066,
+      "kl": 0.34190677378865075,
+      "learning_rate": 0.00017773557980983262,
+      "loss": 0.004829109646379948,
+      "memory(GiB)": 79.48,
+      "reward": 0.3671875,
+      "reward_std": 0.355153352022171,
+      "rewards/Response_no_think_reward_1/mean": 0.3671875,
+      "rewards/Response_no_think_reward_1/std": 0.8406137228012085,
       "step": 43,
-      "train_speed(iter/s)": 0.002397
-    },
-    {
-      "clip_ratio/high_max": 0.007687599485507235,
-      "clip_ratio/high_mean": 0.007687599485507235,
-      "clip_ratio/low_mean": 0.042899149731965736,
-      "clip_ratio/low_min": 0.042899149731965736,
-      "clip_ratio/region_mean": 0.05058674863539636,
-      "epoch": 0.48753462603878117,
-      "grad_norm": 17.524605028595435,
-      "kl": 15.062655651359819,
-      "learning_rate": 0.00018002909470228842,
-      "loss": 0.09481670707464218,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002649
+    },
+    {
+      "clip_ratio/high_max": 0.010433136369101703,
+      "clip_ratio/high_mean": 0.010433136369101703,
+      "clip_ratio/low_mean": 0.0495308703975752,
+      "clip_ratio/low_min": 0.0495308703975752,
+      "clip_ratio/region_mean": 0.05996400595176965,
+      "epoch": 0.5253731343283582,
+      "grad_norm": 0.9699999569587556,
+      "kl": 0.5136687369667925,
+      "learning_rate": 0.00017646134094859815,
+      "loss": -0.008213551715016365,
+      "memory(GiB)": 79.48,
       "step": 44,
-      "train_speed(iter/s)": 0.002427
+      "train_speed(iter/s)": 0.00268
     },
     {
-      "clip_ratio/high_max": 0.0012922932510264218,
-      "clip_ratio/high_mean": 0.0012922932510264218,
-      "clip_ratio/low_mean": 0.002554390055593103,
-      "clip_ratio/low_min": 0.002554390055593103,
-      "clip_ratio/region_mean": 0.003846683306619525,
-      "completions/clipped_ratio": 0.0,
-      "completions/max_length": 32.0,
-      "completions/mean_length": 8.546875,
-      "completions/min_length": 4.0,
-      "epoch": 0.4986149584487535,
-      "grad_norm": 1.474323820635427,
-      "kl": 0.7951482257340103,
-      "learning_rate": 0.00017891405093963938,
-      "loss": 0.008442065678536892,
-      "memory(GiB)": 80.03,
-      "reward": 0.171875,
-      "reward_std": 0.540536105632782,
-      "rewards/Response_no_think_reward/mean": 0.171875,
-      "rewards/Response_no_think_reward/std": 0.9648089408874512,
+      "clip_ratio/high_max": 0.0009920635493472219,
+      "clip_ratio/high_mean": 0.0009920635493472219,
+      "clip_ratio/low_mean": 0.0006793478387407959,
+      "clip_ratio/low_min": 0.0006793478387407959,
+      "clip_ratio/region_mean": 0.0016714113880880177,
+      "completions/clipped_ratio": 0.0234375,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 6.53125,
+      "completions/min_length": 3.0,
+      "epoch": 0.5373134328358209,
+      "grad_norm": 2.6187879748810294,
+      "kl": 0.4996573789976537,
+      "learning_rate": 0.00017515648753881492,
+      "loss": 0.00461646867915988,
+      "memory(GiB)": 79.48,
+      "reward": 0.390625,
+      "reward_std": 0.4600672125816345,
+      "rewards/Response_no_think_reward_1/mean": 0.390625,
+      "rewards/Response_no_think_reward_1/std": 0.8251401782035828,
       "step": 45,
-      "train_speed(iter/s)": 0.002416
-    },
-    {
-      "clip_ratio/high_max": 0.01782548933988437,
-      "clip_ratio/high_mean": 0.01782548933988437,
-      "clip_ratio/low_mean": 0.04517949424916878,
-      "clip_ratio/low_min": 0.04517949424916878,
-      "clip_ratio/region_mean": 0.06300498393829912,
-      "epoch": 0.5096952908587258,
-      "grad_norm": 23.18489815639863,
-      "kl": 20.399557466851547,
-      "learning_rate": 0.0001777723723643014,
-      "loss": 0.15100935101509094,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002583
+    },
+    {
+      "clip_ratio/high_max": 0.023812146624550223,
+      "clip_ratio/high_mean": 0.023812146624550223,
+      "clip_ratio/low_mean": 0.030458416498731822,
+      "clip_ratio/low_min": 0.030458416498731822,
+      "clip_ratio/region_mean": 0.054270562424790114,
+      "epoch": 0.5492537313432836,
+      "grad_norm": 1.046619303634399,
+      "kl": 1.1794932310585864,
+      "learning_rate": 0.00017382154203402048,
+      "loss": -0.007490159012377262,
+      "memory(GiB)": 79.48,
       "step": 46,
-      "train_speed(iter/s)": 0.002445
+      "train_speed(iter/s)": 0.002612
     },
     {
-      "clip_ratio/high_max": 0.0001338329748250544,
-      "clip_ratio/high_mean": 0.0001338329748250544,
-      "clip_ratio/low_mean": 0.0020056332577951252,
-      "clip_ratio/low_min": 0.0020056332577951252,
-      "clip_ratio/region_mean": 0.0021394662326201797,
-      "completions/clipped_ratio": 0.015625,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 11.8203125,
-      "completions/min_length": 4.0,
-      "epoch": 0.5207756232686981,
-      "grad_norm": 1.9101810703272282,
-      "kl": 0.8325624349527061,
-      "learning_rate": 0.0001766044443118978,
-      "loss": 0.01317879930138588,
-      "memory(GiB)": 80.03,
-      "reward": 0.21875,
-      "reward_std": 0.5618736743927002,
-      "rewards/Response_no_think_reward/mean": 0.21875,
-      "rewards/Response_no_think_reward/std": 0.9301384091377258,
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0014244693447835743,
+      "clip_ratio/low_min": 0.0014244693447835743,
+      "clip_ratio/region_mean": 0.0014244693447835743,
+      "completions/clipped_ratio": 0.078125,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 11.59375,
+      "completions/min_length": 3.0,
+      "epoch": 0.5611940298507463,
+      "grad_norm": 1.8613572071063658,
+      "kl": 0.6177221334655769,
+      "learning_rate": 0.0001724570389364019,
+      "loss": 0.004859148990362883,
+      "memory(GiB)": 79.48,
+      "reward": 0.0625,
+      "reward_std": 0.47469794750213623,
+      "rewards/Response_no_think_reward_1/mean": 0.0625,
+      "rewards/Response_no_think_reward_1/std": 0.8761943578720093,
       "step": 47,
-      "train_speed(iter/s)": 0.002435
-    },
-    {
-      "clip_ratio/high_max": 0.01570215745596215,
-      "clip_ratio/high_mean": 0.01570215745596215,
-      "clip_ratio/low_mean": 0.01864259922876954,
-      "clip_ratio/low_min": 0.01864259922876954,
-      "clip_ratio/region_mean": 0.03434475651010871,
-      "epoch": 0.5318559556786704,
-      "grad_norm": 5.211743253538076,
-      "kl": 0.4838231955654919,
-      "learning_rate": 0.00017541066097768963,
-      "loss": 0.03044246882200241,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002518
+    },
+    {
+      "clip_ratio/high_max": 0.009966709709260613,
+      "clip_ratio/high_mean": 0.009966709709260613,
+      "clip_ratio/low_mean": 0.11994545813649893,
+      "clip_ratio/low_min": 0.11994545813649893,
+      "clip_ratio/region_mean": 0.12991216825321317,
+      "epoch": 0.573134328358209,
+      "grad_norm": 0.9444759818819449,
+      "kl": 0.9089975492097437,
+      "learning_rate": 0.00017106352458278522,
+      "loss": -0.009144289419054985,
+      "memory(GiB)": 79.48,
       "step": 48,
-      "train_speed(iter/s)": 0.002462
+      "train_speed(iter/s)": 0.002547
     },
     {
-      "clip_ratio/high_max": 0.0,
-      "clip_ratio/high_mean": 0.0,
-      "clip_ratio/low_mean": 0.002905012297560461,
-      "clip_ratio/low_min": 0.002905012297560461,
-      "clip_ratio/region_mean": 0.002905012297560461,
-      "completions/clipped_ratio": 0.0390625,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 17.7265625,
-      "completions/min_length": 4.0,
-      "epoch": 0.5429362880886427,
-      "grad_norm": 0.9459153229739418,
-      "kl": 0.33009129445417784,
-      "learning_rate": 0.00017419142528352817,
-      "loss": 0.004404420033097267,
-      "memory(GiB)": 80.03,
-      "reward": -0.0859375,
-      "reward_std": 0.547653079032898,
-      "rewards/Response_no_think_reward/mean": -0.0859375,
-      "rewards/Response_no_think_reward/std": 0.9138250946998596,
+      "clip_ratio/high_max": 0.00021043770539108664,
+      "clip_ratio/high_mean": 0.00021043770539108664,
+      "clip_ratio/low_mean": 0.002260449342429638,
+      "clip_ratio/low_min": 0.002260449342429638,
+      "clip_ratio/region_mean": 0.0024708870478207245,
+      "completions/clipped_ratio": 0.1328125,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 12.921875,
+      "completions/min_length": 3.0,
+      "epoch": 0.5850746268656717,
+      "grad_norm": 1.7364401500647022,
+      "kl": 0.6272030578693375,
+      "learning_rate": 0.00016964155692588616,
+      "loss": 0.004394113551825285,
+      "memory(GiB)": 79.48,
+      "reward": 0.1484375,
+      "reward_std": 0.424231618642807,
+      "rewards/Response_no_think_reward_1/mean": 0.1484375,
+      "rewards/Response_no_think_reward_1/std": 0.914363443851471,
       "step": 49,
-      "train_speed(iter/s)": 0.002452
-    },
-    {
-      "clip_ratio/high_max": 0.00849696435034275,
-      "clip_ratio/high_mean": 0.00849696435034275,
-      "clip_ratio/low_mean": 0.059440263896249235,
-      "clip_ratio/low_min": 0.059440263896249235,
-      "clip_ratio/region_mean": 0.06793722766451538,
-      "epoch": 0.554016620498615,
-      "grad_norm": 1.5192885268898135,
-      "kl": 0.5284001431518845,
-      "learning_rate": 0.0001729471487418621,
-      "loss": -0.012666964903473854,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002518
+    },
+    {
+      "clip_ratio/high_max": 0.03189032070804387,
+      "clip_ratio/high_mean": 0.03189032070804387,
+      "clip_ratio/low_mean": 0.04090093384729698,
+      "clip_ratio/low_min": 0.04090093384729698,
+      "clip_ratio/region_mean": 0.07279125548666343,
+      "epoch": 0.5970149253731343,
+      "grad_norm": 4.461755913109612,
+      "kl": 1.2612105239531957,
+      "learning_rate": 0.00016819170531091017,
+      "loss": 0.0014673060504719615,
+      "memory(GiB)": 79.48,
       "step": 50,
-      "train_speed(iter/s)": 0.002478
+      "train_speed(iter/s)": 0.002545
     },
     {
-      "clip_ratio/high_max": 0.0,
-      "clip_ratio/high_mean": 0.0,
-      "clip_ratio/low_mean": 0.0012755101779475808,
-      "clip_ratio/low_min": 0.0012755101779475808,
-      "clip_ratio/region_mean": 0.0012755101779475808,
-      "completions/clipped_ratio": 0.0078125,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 12.640625,
-      "completions/min_length": 3.0,
-      "epoch": 0.5650969529085873,
-      "grad_norm": 1.7932835802898264,
-      "kl": 0.582448753528297,
-      "learning_rate": 0.00017167825131684513,
-      "loss": 0.008432275615632534,
-      "memory(GiB)": 80.03,
-      "reward": 0.15625,
-      "reward_std": 0.7085258960723877,
-      "rewards/Response_no_think_reward/mean": 0.15625,
-      "rewards/Response_no_think_reward/std": 1.0228685140609741,
+      "clip_ratio/high_max": 0.0015625000232830644,
+      "clip_ratio/high_mean": 0.0015625000232830644,
+      "clip_ratio/low_mean": 0.0016522249497938901,
+      "clip_ratio/low_min": 0.0016522249497938901,
+      "clip_ratio/region_mean": 0.0032147249730769545,
+      "completions/clipped_ratio": 0.0859375,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 10.46875,
+      "completions/min_length": 3.0,
+      "epoch": 0.608955223880597,
+      "grad_norm": 2.519038758282151,
+      "kl": 0.549895275675226,
+      "learning_rate": 0.00016671455024759067,
+      "loss": 0.0033532448578625917,
+      "memory(GiB)": 79.48,
+      "reward": 0.1796875,
+      "reward_std": 0.39886242151260376,
+      "rewards/Response_no_think_reward_1/mean": 0.1796875,
+      "rewards/Response_no_think_reward_1/std": 0.9258628487586975,
       "step": 51,
-      "train_speed(iter/s)": 0.002465
-    },
-    {
-      "clip_ratio/high_max": 0.006597792147658765,
-      "clip_ratio/high_mean": 0.006597792147658765,
-      "clip_ratio/low_mean": 0.049801092012785375,
-      "clip_ratio/low_min": 0.049801092012785375,
-      "clip_ratio/region_mean": 0.05639888462610543,
-      "epoch": 0.5761772853185596,
-      "grad_norm": 0.8422307837338449,
-      "kl": 0.7035694038495421,
-      "learning_rate": 0.00017038516128259115,
-      "loss": -0.011608053930103779,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002473
+    },
+    {
+      "clip_ratio/high_max": 0.03367150644771755,
+      "clip_ratio/high_mean": 0.03367150644771755,
+      "clip_ratio/low_mean": 0.0576377569232136,
+      "clip_ratio/low_min": 0.0576377569232136,
+      "clip_ratio/region_mean": 0.09130926383659244,
+      "epoch": 0.6208955223880597,
+      "grad_norm": 0.9238461637378925,
+      "kl": 0.672293990442995,
+      "learning_rate": 0.00016521068317775753,
+      "loss": -0.019059473648667336,
+      "memory(GiB)": 79.48,
       "step": 52,
-      "train_speed(iter/s)": 0.00249
+      "train_speed(iter/s)": 0.002499
     },
     {
-      "clip_ratio/high_max": 0.0,
-      "clip_ratio/high_mean": 0.0,
-      "clip_ratio/low_mean": 0.0005558648990700021,
-      "clip_ratio/low_min": 0.0005558648990700021,
-      "clip_ratio/region_mean": 0.0005558648990700021,
-      "completions/clipped_ratio": 0.015625,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 12.5,
-      "completions/min_length": 4.0,
-      "epoch": 0.5872576177285319,
-      "grad_norm": 1.0211580850819229,
-      "kl": 0.7616109761074767,
-      "learning_rate": 0.00016906831507862443,
-      "loss": 0.005634145811200142,
-      "memory(GiB)": 80.03,
-      "reward": 0.1640625,
-      "reward_std": 0.70704185962677,
-      "rewards/Response_no_think_reward/mean": 0.1640625,
-      "rewards/Response_no_think_reward/std": 0.9702450633049011,
+      "clip_ratio/high_max": 0.0013558201026171446,
+      "clip_ratio/high_mean": 0.0013558201026171446,
+      "clip_ratio/low_mean": 0.010322635032935068,
+      "clip_ratio/low_min": 0.010322635032935068,
+      "clip_ratio/region_mean": 0.011678455251967534,
+      "completions/clipped_ratio": 0.1796875,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 14.0390625,
+      "completions/min_length": 3.0,
+      "epoch": 0.6328358208955224,
+      "grad_norm": 6.547271710761981,
+      "kl": 1.0316203383699758,
+      "learning_rate": 0.00016368070623852792,
+      "loss": 0.006902866996824741,
+      "memory(GiB)": 79.48,
+      "reward": 0.1328125,
+      "reward_std": 0.561459481716156,
+      "rewards/Response_no_think_reward_1/mean": 0.1328125,
+      "rewards/Response_no_think_reward_1/std": 0.9081528782844543,
       "step": 53,
-      "train_speed(iter/s)": 0.002477
-    },
-    {
-      "clip_ratio/high_max": 0.02744574609096162,
-      "clip_ratio/high_mean": 0.02744574609096162,
-      "clip_ratio/low_mean": 0.04221567645436153,
-      "clip_ratio/low_min": 0.04221567645436153,
-      "clip_ratio/region_mean": 0.06966142146848142,
-      "epoch": 0.5983379501385041,
-      "grad_norm": 0.5808426269905869,
-      "kl": 0.801087921798171,
-      "learning_rate": 0.00016772815716257412,
-      "loss": -0.01366308145225048,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002484
+    },
+    {
+      "clip_ratio/high_max": 0.0014829304564045742,
+      "clip_ratio/high_mean": 0.0014829304564045742,
+      "clip_ratio/low_mean": 0.1896184734068811,
+      "clip_ratio/low_min": 0.1896184734068811,
+      "clip_ratio/region_mean": 0.19110140530392528,
+      "epoch": 0.6447761194029851,
+      "grad_norm": 3.3472874723788295,
+      "kl": 2.011909826847841,
+      "learning_rate": 0.00016212523202121544,
+      "loss": 0.00577022647485137,
+      "memory(GiB)": 79.48,
       "step": 54,
-      "train_speed(iter/s)": 0.002502
+      "train_speed(iter/s)": 0.002509
     },
     {
-      "clip_ratio/high_max": 0.0003881987649947405,
-      "clip_ratio/high_mean": 0.0003881987649947405,
-      "clip_ratio/low_mean": 0.001932911021867767,
-      "clip_ratio/low_min": 0.001932911021867767,
-      "clip_ratio/region_mean": 0.0023211097868625075,
-      "completions/clipped_ratio": 0.0390625,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 18.9921875,
-      "completions/min_length": 6.0,
-      "epoch": 0.6094182825484764,
-      "grad_norm": 1.1265096543789663,
-      "kl": 0.5226203491911292,
-      "learning_rate": 0.00016636513986016213,
-      "loss": 0.007094700820744038,
-      "memory(GiB)": 80.03,
-      "reward": 0.1875,
-      "reward_std": 0.7182328104972839,
-      "rewards/Response_no_think_reward/mean": 0.1875,
-      "rewards/Response_no_think_reward/std": 1.0480577945709229,
+      "clip_ratio/high_max": 0.0017361111240461469,
+      "clip_ratio/high_mean": 0.0017361111240461469,
+      "clip_ratio/low_mean": 0.004226236138492823,
+      "clip_ratio/low_min": 0.004226236138492823,
+      "clip_ratio/region_mean": 0.0059623472625389695,
+      "completions/clipped_ratio": 0.0859375,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 10.125,
+      "completions/min_length": 3.0,
+      "epoch": 0.6567164179104478,
+      "grad_norm": 2.1721724253324193,
+      "kl": 1.4456337625160813,
+      "learning_rate": 0.00016054488332605283,
+      "loss": 0.007076151203364134,
+      "memory(GiB)": 79.48,
+      "reward": 0.3046875,
+      "reward_std": 0.4615631103515625,
+      "rewards/Response_no_think_reward_1/mean": 0.3046875,
+      "rewards/Response_no_think_reward_1/std": 0.8654214143753052,
       "step": 55,
-      "train_speed(iter/s)": 0.00249
-    },
-    {
-      "clip_ratio/high_max": 0.011908911721548066,
-      "clip_ratio/high_mean": 0.011908911721548066,
-      "clip_ratio/low_mean": 0.05817525731981732,
-      "clip_ratio/low_min": 0.05817525731981732,
-      "clip_ratio/region_mean": 0.07008416869211942,
-      "epoch": 0.6204986149584487,
-      "grad_norm": 0.5560920258062684,
-      "kl": 0.5747523186728358,
-      "learning_rate": 0.000164979723212536,
-      "loss": -0.014954826794564724,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002495
+    },
+    {
+      "clip_ratio/high_max": 0.07161095441551879,
+      "clip_ratio/high_mean": 0.07161095441551879,
+      "clip_ratio/low_mean": 0.0391811344306916,
+      "clip_ratio/low_min": 0.0391811344306916,
+      "clip_ratio/region_mean": 0.11079208552837372,
+      "epoch": 0.6686567164179105,
+      "grad_norm": 1.6216131312943962,
+      "kl": 0.9301650315364896,
+      "learning_rate": 0.00015894029291282758,
+      "loss": -0.018307652324438095,
+      "memory(GiB)": 79.48,
       "step": 56,
-      "train_speed(iter/s)": 0.002513
+      "train_speed(iter/s)": 0.002519
     },
     {
-      "clip_ratio/high_max": 0.002692167239729315,
-      "clip_ratio/high_mean": 0.002692167239729315,
-      "clip_ratio/low_mean": 0.0006428283377317712,
-      "clip_ratio/low_min": 0.0006428283377317712,
-      "clip_ratio/region_mean": 0.003334995577461086,
-      "completions/clipped_ratio": 0.0390625,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 16.53125,
-      "completions/min_length": 6.0,
-      "epoch": 0.631578947368421,
-      "grad_norm": 1.5596809103877662,
-      "kl": 0.6855434570461512,
-      "learning_rate": 0.00016357237482099684,
-      "loss": 0.0062956069596111774,
-      "memory(GiB)": 80.03,
-      "reward": 0.53125,
-      "reward_std": 0.5907745361328125,
-      "rewards/Response_no_think_reward/mean": 0.53125,
-      "rewards/Response_no_think_reward/std": 1.0790598392486572,
+      "clip_ratio/high_max": 0.0028409091755747795,
+      "clip_ratio/high_mean": 0.0028409091755747795,
+      "clip_ratio/low_mean": 0.007847597065847367,
+      "clip_ratio/low_min": 0.007847597065847367,
+      "clip_ratio/region_mean": 0.010688506241422147,
+      "completions/clipped_ratio": 0.0703125,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 8.2109375,
+      "completions/min_length": 3.0,
+      "epoch": 0.6805970149253732,
+      "grad_norm": 4.528309911623671,
+      "kl": 1.2233935340000244,
+      "learning_rate": 0.00015731210324752972,
+      "loss": 0.015045193955302238,
+      "memory(GiB)": 79.48,
+      "reward": 0.171875,
+      "reward_std": 0.5126261711120605,
+      "rewards/Response_no_think_reward_1/mean": 0.171875,
+      "rewards/Response_no_think_reward_1/std": 0.9315922856330872,
       "step": 57,
-      "train_speed(iter/s)": 0.002501
-    },
-    {
-      "clip_ratio/high_max": 0.0316368987550959,
-      "clip_ratio/high_mean": 0.0316368987550959,
-      "clip_ratio/low_mean": 0.015557856269879267,
-      "clip_ratio/low_min": 0.015557856269879267,
-      "clip_ratio/region_mean": 0.04719475514139049,
-      "epoch": 0.6426592797783933,
-      "grad_norm": 0.950735686233936,
-      "kl": 0.5240823943167925,
-      "learning_rate": 0.00016214356968917648,
-      "loss": -0.015531505458056927,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002481
+    },
+    {
+      "clip_ratio/high_max": 0.022757820785045624,
+      "clip_ratio/high_mean": 0.022757820785045624,
+      "clip_ratio/low_mean": 0.14223288698121905,
+      "clip_ratio/low_min": 0.14223288698121905,
+      "clip_ratio/region_mean": 0.1649907068349421,
+      "epoch": 0.6925373134328359,
+      "grad_norm": 75.05881389891198,
+      "kl": 71.87919758663793,
+      "learning_rate": 0.00015566096624511307,
+      "loss": 0.9493598341941833,
+      "memory(GiB)": 79.48,
       "step": 58,
-      "train_speed(iter/s)": 0.002524
+      "train_speed(iter/s)": 0.002504
     },
     {
-      "clip_ratio/high_max": 0.00011467889999039471,
-      "clip_ratio/high_mean": 0.00011467889999039471,
-      "clip_ratio/low_mean": 0.0004376750875962898,
-      "clip_ratio/low_min": 0.0004376750875962898,
-      "clip_ratio/region_mean": 0.0005523539875866845,
-      "completions/clipped_ratio": 0.078125,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 23.9140625,
-      "completions/min_length": 6.0,
-      "epoch": 0.6537396121883656,
-      "grad_norm": 1.3162721482480142,
-      "kl": 0.5214177745606321,
-      "learning_rate": 0.00016069379006271566,
-      "loss": 0.004921327345073223,
-      "memory(GiB)": 80.03,
-      "reward": -0.0625,
-      "reward_std": 0.6303451657295227,
-      "rewards/Response_no_think_reward/mean": -0.0625,
-      "rewards/Response_no_think_reward/std": 1.0019665956497192,
+      "clip_ratio/high_max": 0.0005230125389061868,
+      "clip_ratio/high_mean": 0.0005230125389061868,
+      "clip_ratio/low_mean": 0.012553135456983,
+      "clip_ratio/low_min": 0.012553135456983,
+      "clip_ratio/region_mean": 0.013076147937681526,
+      "completions/clipped_ratio": 0.1171875,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 9.3828125,
+      "completions/min_length": 3.0,
+      "epoch": 0.7044776119402985,
+      "grad_norm": 2.061647426051912,
+      "kl": 3.4461456341377925,
+      "learning_rate": 0.00015398754300847343,
+      "loss": 0.024864066392183304,
+      "memory(GiB)": 79.48,
+      "reward": 0.3046875,
+      "reward_std": 0.35084033012390137,
+      "rewards/Response_no_think_reward_1/mean": 0.3046875,
+      "rewards/Response_no_think_reward_1/std": 0.9010806679725647,
       "step": 59,
-      "train_speed(iter/s)": 0.002511
-    },
-    {
-      "clip_ratio/high_max": 0.00810479320352897,
-      "clip_ratio/high_mean": 0.00810479320352897,
-      "clip_ratio/low_mean": 0.05729365168372169,
-      "clip_ratio/low_min": 0.05729365168372169,
-      "clip_ratio/region_mean": 0.06539844395592809,
-      "epoch": 0.6648199445983379,
-      "grad_norm": 0.5733548474442324,
-      "kl": 0.6363338100200053,
-      "learning_rate": 0.00015922352526649803,
-      "loss": -0.021113965660333633,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002474
+    },
+    {
+      "clip_ratio/high_max": 0.010577498760540038,
+      "clip_ratio/high_mean": 0.010577498760540038,
+      "clip_ratio/low_mean": 0.15099563542753458,
+      "clip_ratio/low_min": 0.15099563542753458,
+      "clip_ratio/region_mean": 0.16157313529402018,
+      "epoch": 0.7164179104477612,
+      "grad_norm": 6.318543519105419,
+      "kl": 9.211844031116925,
+      "learning_rate": 0.00015229250356374804,
+      "loss": 0.058415383100509644,
+      "memory(GiB)": 79.48,
       "step": 60,
-      "train_speed(iter/s)": 0.002532
+      "train_speed(iter/s)": 0.002497
     },
     {
-      "clip_ratio/high_max": 0.0005996339968987741,
-      "clip_ratio/high_mean": 0.0005996339968987741,
-      "clip_ratio/low_mean": 0.004901745676761493,
-      "clip_ratio/low_min": 0.004901745676761493,
-      "clip_ratio/region_mean": 0.00550137969548814,
-      "completions/clipped_ratio": 0.0859375,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 22.375,
-      "completions/min_length": 2.0,
-      "epoch": 0.6759002770083102,
-      "grad_norm": 1.887884786225259,
-      "kl": 2.07791917472332,
-      "learning_rate": 0.00015773327153949465,
-      "loss": 0.008165515027940273,
-      "memory(GiB)": 80.03,
-      "reward": 0.40625,
-      "reward_std": 0.5973243117332458,
-      "rewards/Response_no_think_reward/mean": 0.40625,
-      "rewards/Response_no_think_reward/std": 1.186787486076355,
+      "clip_ratio/high_max": 0.0023436461633536965,
+      "clip_ratio/high_mean": 0.0023436461633536965,
+      "clip_ratio/low_mean": 0.01256567076779902,
+      "clip_ratio/low_min": 0.01256567076779902,
+      "clip_ratio/region_mean": 0.014909316785633564,
+      "completions/clipped_ratio": 0.1640625,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 9.0546875,
+      "completions/min_length": 3.0,
+      "epoch": 0.7283582089552239,
+      "grad_norm": 8.650077130485371,
+      "kl": 12.875115153714432,
+      "learning_rate": 0.00015057652659204197,
+      "loss": 0.11098186671733856,
+      "memory(GiB)": 79.48,
+      "reward": -0.2109375,
+      "reward_std": 0.3248431980609894,
+      "rewards/Response_no_think_reward_1/mean": -0.2109375,
+      "rewards/Response_no_think_reward_1/std": 0.7060185074806213,
       "step": 61,
-      "train_speed(iter/s)": 0.002506
-    },
-    {
-      "clip_ratio/high_max": 0.017942053091246635,
-      "clip_ratio/high_mean": 0.017942053091246635,
-      "clip_ratio/low_mean": 0.06714771036058664,
-      "clip_ratio/low_min": 0.06714771036058664,
-      "clip_ratio/region_mean": 0.08508976292796433,
-      "epoch": 0.6869806094182825,
-      "grad_norm": 11.312330346219078,
-      "kl": 0.5888316835043952,
-      "learning_rate": 0.00015622353186727544,
-      "loss": 0.04619387909770012,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002485
+    },
+    {
+      "clip_ratio/high_max": 0.04812772118020803,
+      "clip_ratio/high_mean": 0.04812772118020803,
+      "clip_ratio/low_mean": 0.016272668377496302,
+      "clip_ratio/low_min": 0.016272668377496302,
+      "clip_ratio/region_mean": 0.06440038897562772,
+      "epoch": 0.7402985074626866,
+      "grad_norm": 6.330882663211948,
+      "kl": 4.831557432677073,
+      "learning_rate": 0.00014884029915768944,
+      "loss": 0.08532620966434479,
+      "memory(GiB)": 79.48,
       "step": 62,
-      "train_speed(iter/s)": 0.002527
+      "train_speed(iter/s)": 0.002507
     },
     {
-      "clip_ratio/high_max": 0.00020032051543239504,
-      "clip_ratio/high_mean": 0.00020032051543239504,
-      "clip_ratio/low_mean": 0.001202335930429399,
-      "clip_ratio/low_min": 0.001202335930429399,
-      "clip_ratio/region_mean": 0.001402656445861794,
-      "completions/clipped_ratio": 0.0546875,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 18.5859375,
-      "completions/min_length": 2.0,
-      "epoch": 0.6980609418282548,
-      "grad_norm": 1.9804020858052087,
-      "kl": 6.258792589243967,
-      "learning_rate": 0.00015469481581224272,
-      "loss": 0.014128579758107662,
-      "memory(GiB)": 80.03,
-      "reward": 0.78125,
-      "reward_std": 0.5936684608459473,
-      "rewards/Response_no_think_reward/mean": 0.78125,
-      "rewards/Response_no_think_reward/std": 1.235546350479126,
+      "clip_ratio/high_max": 0.0009259259095415473,
+      "clip_ratio/high_mean": 0.0009259259095415473,
+      "clip_ratio/low_mean": 0.005223244777880609,
+      "clip_ratio/low_min": 0.005223244777880609,
+      "clip_ratio/region_mean": 0.006149170687422156,
+      "completions/clipped_ratio": 0.0625,
+      "completions/max_length": 49.0,
+      "completions/mean_length": 6.8203125,
+      "completions/min_length": 3.0,
+      "epoch": 0.7522388059701492,
+      "grad_norm": 0.8455863083310526,
+      "kl": 3.406710424387711,
+      "learning_rate": 0.00014708451643315827,
+      "loss": 0.015459287911653519,
+      "memory(GiB)": 79.48,
+      "reward": 0.2890625,
+      "reward_std": 0.09522313624620438,
+      "rewards/Response_no_think_reward_1/mean": 0.2890625,
+      "rewards/Response_no_think_reward_1/std": 0.9319553971290588,
       "step": 63,
-      "train_speed(iter/s)": 0.002486
-    },
-    {
-      "clip_ratio/high_max": 0.020077986438991502,
-      "clip_ratio/high_mean": 0.020077986438991502,
-      "clip_ratio/low_mean": 0.12254823022522032,
-      "clip_ratio/low_min": 0.12254823022522032,
-      "clip_ratio/region_mean": 0.142626216635108,
-      "epoch": 0.7091412742382271,
-      "grad_norm": 1.5460664241155249,
-      "kl": 5.241092938755173,
-      "learning_rate": 0.0001531476393416456,
-      "loss": -0.0033248686231672764,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002478
+    },
+    {
+      "clip_ratio/high_max": 0.00038343557389453053,
+      "clip_ratio/high_mean": 0.00038343557389453053,
+      "clip_ratio/low_mean": 0.03692773682996631,
+      "clip_ratio/low_min": 0.03692773682996631,
+      "clip_ratio/region_mean": 0.037311173509806395,
+      "epoch": 0.764179104477612,
+      "grad_norm": 1.460493225059439,
+      "kl": 3.925420918938471,
+      "learning_rate": 0.00014530988142070803,
+      "loss": 0.015160211361944675,
+      "memory(GiB)": 79.48,
       "step": 64,
-      "train_speed(iter/s)": 0.002506
+      "train_speed(iter/s)": 0.002499
     },
     {
-      "clip_ratio/high_max": 0.0006774475477868691,
-      "clip_ratio/high_mean": 0.0006774475477868691,
-      "clip_ratio/low_mean": 0.0009178321633953601,
-      "clip_ratio/low_min": 0.0009178321633953601,
-      "clip_ratio/region_mean": 0.0015952797257341444,
-      "completions/clipped_ratio": 0.0390625,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 15.1484375,
-      "completions/min_length": 6.0,
-      "epoch": 0.7202216066481995,
-      "grad_norm": 1.528383760465238,
-      "kl": 1.888367731589824,
-      "learning_rate": 0.00015158252465343242,
-      "loss": 0.010861902497708797,
-      "memory(GiB)": 80.03,
-      "reward": 0.59375,
-      "reward_std": 0.5738716125488281,
-      "rewards/Response_no_think_reward/mean": 0.59375,
-      "rewards/Response_no_think_reward/std": 1.1462881565093994,
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.02216397225856781,
+      "clip_ratio/low_min": 0.02216397225856781,
+      "clip_ratio/region_mean": 0.02216397225856781,
+      "completions/clipped_ratio": 0.15625,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 8.1484375,
+      "completions/min_length": 3.0,
+      "epoch": 0.7761194029850746,
+      "grad_norm": 16.54465377888419,
+      "kl": 13.583033993607387,
+      "learning_rate": 0.00014351710467091336,
+      "loss": 0.09176836162805557,
+      "memory(GiB)": 79.48,
+      "reward": 0.0,
+      "reward_std": 0.44421693682670593,
+      "rewards/Response_no_think_reward_1/mean": 0.0,
+      "rewards/Response_no_think_reward_1/std": 0.8418202996253967,
       "step": 65,
-      "train_speed(iter/s)": 0.002444
-    },
-    {
-      "clip_ratio/high_max": 0.05604529404081404,
-      "clip_ratio/high_mean": 0.05604529404081404,
-      "clip_ratio/low_mean": 0.012011443439405411,
-      "clip_ratio/low_min": 0.012011443439405411,
-      "clip_ratio/region_mean": 0.06805673893541098,
-      "epoch": 0.7313019390581718,
-      "grad_norm": 0.9299498266912626,
-      "kl": 1.0373663480422692,
-      "learning_rate": 0.00015000000000000001,
-      "loss": -0.001186850480735302,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.00249
+    },
+    {
+      "clip_ratio/high_max": 0.016292952001094818,
+      "clip_ratio/high_mean": 0.016292952001094818,
+      "clip_ratio/low_mean": 0.02969469508389011,
+      "clip_ratio/low_min": 0.02969469508389011,
+      "clip_ratio/region_mean": 0.04598764737602323,
+      "epoch": 0.7880597014925373,
+      "grad_norm": 7.478581616879351,
+      "kl": 9.063416050747037,
+      "learning_rate": 0.0001417069039981647,
+      "loss": 0.08187955617904663,
+      "memory(GiB)": 79.48,
       "step": 66,
-      "train_speed(iter/s)": 0.002463
+      "train_speed(iter/s)": 0.00251
     },
     {
-      "clip_ratio/high_max": 0.006279462773818523,
-      "clip_ratio/high_mean": 0.006279462773818523,
-      "clip_ratio/low_mean": 0.006620214961003512,
-      "clip_ratio/low_min": 0.006620214961003512,
-      "clip_ratio/region_mean": 0.012899677676614374,
-      "completions/clipped_ratio": 0.0546875,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 17.5234375,
-      "completions/min_length": 2.0,
-      "epoch": 0.7423822714681441,
-      "grad_norm": 3.0195311219652377,
-      "kl": 1.4996049946639687,
-      "learning_rate": 0.0001484005995098999,
-      "loss": 0.0103817880153656,
-      "memory(GiB)": 80.03,
-      "reward": 0.2421875,
-      "reward_std": 0.5205168724060059,
-      "rewards/Response_no_think_reward/mean": 0.2421875,
-      "rewards/Response_no_think_reward/std": 1.2595843076705933,
+      "clip_ratio/high_max": 0.0003571428533177823,
+      "clip_ratio/high_mean": 0.0003571428533177823,
+      "clip_ratio/low_mean": 0.011069363099522889,
+      "clip_ratio/low_min": 0.011069363099522889,
+      "clip_ratio/region_mean": 0.011426505981944501,
+      "completions/clipped_ratio": 0.1171875,
+      "completions/max_length": 48.0,
+      "completions/mean_length": 6.75,
+      "completions/min_length": 3.0,
+      "epoch": 0.8,
+      "grad_norm": 2.789323653853436,
+      "kl": 6.344451879562257,
+      "learning_rate": 0.00013988000419326072,
+      "loss": 0.04125535860657692,
+      "memory(GiB)": 79.48,
+      "reward": 0.21875,
+      "reward_std": 0.40496253967285156,
+      "rewards/Response_no_think_reward_1/mean": 0.21875,
+      "rewards/Response_no_think_reward_1/std": 0.8778777122497559,
       "step": 67,
-      "train_speed(iter/s)": 0.002399
-    },
-    {
-      "clip_ratio/high_max": 0.030079254298470914,
-      "clip_ratio/high_mean": 0.030079254298470914,
-      "clip_ratio/low_mean": 0.10815927106887102,
-      "clip_ratio/low_min": 0.10815927106887102,
-      "clip_ratio/region_mean": 0.1382385252509266,
-      "epoch": 0.7534626038781164,
-      "grad_norm": 7.7164901906176375,
-      "kl": 4.463950714329258,
-      "learning_rate": 0.0001467848630075608,
-      "loss": 0.0073772999458014965,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.0025
+    },
+    {
+      "clip_ratio/high_max": 0.036289573181420565,
+      "clip_ratio/high_mean": 0.036289573181420565,
+      "clip_ratio/low_mean": 0.07330909674055874,
+      "clip_ratio/low_min": 0.07330909674055874,
+      "clip_ratio/region_mean": 0.10959866992197931,
+      "epoch": 0.8119402985074626,
+      "grad_norm": 3.351094606732551,
+      "kl": 7.85955839138478,
+      "learning_rate": 0.00013803713673320772,
+      "loss": 0.0294882133603096,
+      "memory(GiB)": 79.48,
       "step": 68,
-      "train_speed(iter/s)": 0.002418
+      "train_speed(iter/s)": 0.00252
     },
     {
-      "clip_ratio/high_max": 0.0009369817780680023,
-      "clip_ratio/high_mean": 0.0009369817780680023,
-      "clip_ratio/low_mean": 0.006284750299528241,
-      "clip_ratio/low_min": 0.006284750299528241,
-      "clip_ratio/region_mean": 0.007221732055768371,
-      "completions/clipped_ratio": 0.1484375,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 26.9296875,
-      "completions/min_length": 3.0,
-      "epoch": 0.7645429362880887,
-      "grad_norm": 2.034534758963297,
-      "kl": 2.321827916195616,
-      "learning_rate": 0.00014515333583108896,
-      "loss": 0.01671188324689865,
-      "memory(GiB)": 80.03,
-      "reward": 0.5390625,
-      "reward_std": 0.4874863028526306,
-      "rewards/Response_no_think_reward/mean": 0.5390625,
-      "rewards/Response_no_think_reward/std": 1.3095791339874268,
+      "clip_ratio/high_max": 0.0034414873807691038,
+      "clip_ratio/high_mean": 0.0034414873807691038,
+      "clip_ratio/low_mean": 0.017763152020052075,
+      "clip_ratio/low_min": 0.017763152020052075,
+      "clip_ratio/region_mean": 0.021204639226198196,
+      "completions/clipped_ratio": 0.1953125,
+      "completions/max_length": 50.0,
+      "completions/mean_length": 9.0390625,
+      "completions/min_length": 3.0,
+      "epoch": 0.8238805970149253,
+      "grad_norm": 2.3936208496329043,
+      "kl": 8.319812071829801,
+      "learning_rate": 0.00013617903948834155,
+      "loss": 0.0558692142367363,
+      "memory(GiB)": 79.48,
+      "reward": 0.1484375,
+      "reward_std": 0.5162962079048157,
+      "rewards/Response_no_think_reward_1/mean": 0.1484375,
+      "rewards/Response_no_think_reward_1/std": 0.7642591595649719,
       "step": 69,
-      "train_speed(iter/s)": 0.002373
-    },
-    {
-      "clip_ratio/high_max": 0.012032406637445092,
-      "clip_ratio/high_mean": 0.012032406637445092,
-      "clip_ratio/low_mean": 0.11593639780767262,
-      "clip_ratio/low_min": 0.11593639780767262,
-      "clip_ratio/region_mean": 0.1279688044451177,
-      "epoch": 0.775623268698061,
-      "grad_norm": 1.497770517232466,
-      "kl": 3.323778461664915,
-      "learning_rate": 0.00014350656864820733,
-      "loss": 0.0011269270908087492,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002511
+    },
+    {
+      "clip_ratio/high_max": 0.05922704888507724,
+      "clip_ratio/high_mean": 0.05922704888507724,
+      "clip_ratio/low_mean": 0.03970964252948761,
+      "clip_ratio/low_min": 0.03970964252948761,
+      "clip_ratio/region_mean": 0.09893669094890356,
+      "epoch": 0.835820895522388,
+      "grad_norm": 6.476273530366032,
+      "kl": 5.813117837678874,
+      "learning_rate": 0.00013430645642688988,
+      "loss": 0.04987351968884468,
+      "memory(GiB)": 79.48,
       "step": 70,
-      "train_speed(iter/s)": 0.002392
+      "train_speed(iter/s)": 0.00253
     },
     {
-      "clip_ratio/high_max": 0.000255623715929687,
-      "clip_ratio/high_mean": 0.000255623715929687,
-      "clip_ratio/low_mean": 0.005516766890650615,
-      "clip_ratio/low_min": 0.005516766890650615,
-      "clip_ratio/region_mean": 0.005772390664787963,
-      "completions/clipped_ratio": 0.09375,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 17.25,
-      "completions/min_length": 2.0,
-      "epoch": 0.7867036011080333,
-      "grad_norm": 2.3636642819394384,
-      "kl": 2.6716066980734468,
-      "learning_rate": 0.00014184511727039612,
-      "loss": 0.020374953746795654,
-      "memory(GiB)": 80.03,
-      "reward": 0.34375,
-      "reward_std": 0.5405020713806152,
-      "rewards/Response_no_think_reward/mean": 0.34375,
-      "rewards/Response_no_think_reward/std": 1.2323558330535889,
+      "clip_ratio/high_max": 0.002159281895728782,
+      "clip_ratio/high_mean": 0.002159281895728782,
+      "clip_ratio/low_mean": 0.010578428045846522,
+      "clip_ratio/low_min": 0.010578428045846522,
+      "clip_ratio/region_mean": 0.012737710087094456,
+      "completions/clipped_ratio": 0.1171875,
+      "completions/max_length": 44.0,
+      "completions/mean_length": 9.2421875,
+      "completions/min_length": 3.0,
+      "epoch": 0.8477611940298507,
+      "grad_norm": 3.9784798974103217,
+      "kl": 7.147874560090713,
+      "learning_rate": 0.00013242013731709287,
+      "loss": 0.06330341100692749,
+      "memory(GiB)": 79.48,
+      "reward": 0.015625,
+      "reward_std": 0.39743539690971375,
+      "rewards/Response_no_think_reward_1/mean": 0.015625,
+      "rewards/Response_no_think_reward_1/std": 0.6985291242599487,
       "step": 71,
-      "train_speed(iter/s)": 0.00236
-    },
-    {
-      "clip_ratio/high_max": 0.004156995622906834,
-      "clip_ratio/high_mean": 0.004156995622906834,
-      "clip_ratio/low_mean": 0.11334922257810831,
-      "clip_ratio/low_min": 0.11334922257810831,
-      "clip_ratio/region_mean": 0.11750621721148491,
-      "epoch": 0.7977839335180056,
-      "grad_norm": 1.5028809890146027,
-      "kl": 3.652272095438093,
-      "learning_rate": 0.00014016954246529696,
-      "loss": 0.00900588370859623,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002526
+    },
+    {
+      "clip_ratio/high_max": 0.014909500489011407,
+      "clip_ratio/high_mean": 0.014909500489011407,
+      "clip_ratio/low_mean": 0.05899865587707609,
+      "clip_ratio/low_min": 0.05899865587707609,
+      "clip_ratio/region_mean": 0.07390815508551896,
+      "epoch": 0.8597014925373134,
+      "grad_norm": 107.14584369632281,
+      "kl": 5.481248813099228,
+      "learning_rate": 0.00013052083742700173,
+      "loss": 0.5873211026191711,
+      "memory(GiB)": 79.48,
       "step": 72,
-      "train_speed(iter/s)": 0.002378
+      "train_speed(iter/s)": 0.002544
     },
     {
-      "clip_ratio/high_max": 0.0024562697508372366,
-      "clip_ratio/high_mean": 0.0024562697508372366,
-      "clip_ratio/low_mean": 0.00900937442202121,
-      "clip_ratio/low_min": 0.00900937442202121,
-      "clip_ratio/region_mean": 0.01146564440568909,
-      "completions/clipped_ratio": 0.1015625,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 20.9609375,
-      "completions/min_length": 6.0,
-      "epoch": 0.8088642659279779,
-      "grad_norm": 1.9262088024122541,
-      "kl": 2.068145776007441,
-      "learning_rate": 0.00013848040976744457,
-      "loss": 0.009554898366332054,
-      "memory(GiB)": 80.03,
-      "reward": 0.7734375,
-      "reward_std": 0.6535134315490723,
-      "rewards/Response_no_think_reward/mean": 0.7734375,
-      "rewards/Response_no_think_reward/std": 1.224518895149231,
+      "clip_ratio/high_max": 0.005898644914850593,
+      "clip_ratio/high_mean": 0.005898644914850593,
+      "clip_ratio/low_mean": 0.013190844911150634,
+      "clip_ratio/low_min": 0.013190844911150634,
+      "clip_ratio/region_mean": 0.01908948994241655,
+      "completions/clipped_ratio": 0.1484375,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 7.625,
+      "completions/min_length": 3.0,
+      "epoch": 0.8716417910447761,
+      "grad_norm": 5.671656969858763,
+      "kl": 5.887194114271551,
+      "learning_rate": 0.0001286093172220748,
+      "loss": 0.05823807045817375,
+      "memory(GiB)": 79.48,
+      "reward": 0.0078125,
+      "reward_std": 0.47193682193756104,
+      "rewards/Response_no_think_reward_1/mean": 0.0078125,
+      "rewards/Response_no_think_reward_1/std": 0.8648526072502136,
       "step": 73,
-      "train_speed(iter/s)": 0.002347
-    },
-    {
-      "clip_ratio/high_max": 0.05884167249314487,
-      "clip_ratio/high_mean": 0.05884167249314487,
-      "clip_ratio/low_mean": 0.014398490195162594,
-      "clip_ratio/low_min": 0.014398490195162594,
-      "clip_ratio/region_mean": 0.0732401623390615,
-      "epoch": 0.8199445983379502,
-      "grad_norm": 3.7794078536832525,
-      "kl": 1.1005137297324836,
-      "learning_rate": 0.00013677828928738934,
-      "loss": 0.027932219207286835,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002537
+    },
+    {
+      "clip_ratio/high_max": 0.05940199224278331,
+      "clip_ratio/high_mean": 0.05940199224278331,
+      "clip_ratio/low_mean": 0.07782801624853164,
+      "clip_ratio/low_min": 0.07782801624853164,
+      "clip_ratio/region_mean": 0.1372300088405609,
+      "epoch": 0.8835820895522388,
+      "grad_norm": 5.588550612212405,
+      "kl": 9.119772452046163,
+      "learning_rate": 0.00012668634206069304,
+      "loss": 0.05784386023879051,
+      "memory(GiB)": 79.48,
       "step": 74,
-      "train_speed(iter/s)": 0.002365
+      "train_speed(iter/s)": 0.002555
     },
     {
-      "clip_ratio/high_max": 0.0020199596765451133,
-      "clip_ratio/high_mean": 0.0020199596765451133,
-      "clip_ratio/low_mean": 0.009859619050985202,
-      "clip_ratio/low_min": 0.009859619050985202,
-      "clip_ratio/region_mean": 0.0118795787129784,
-      "completions/clipped_ratio": 0.09375,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 19.734375,
-      "completions/min_length": 2.0,
-      "epoch": 0.8310249307479224,
-      "grad_norm": 2.6931881467895598,
-      "kl": 1.9703011065721512,
-      "learning_rate": 0.00013506375551927547,
-      "loss": 0.010023066774010658,
-      "memory(GiB)": 80.03,
-      "reward": 0.4140625,
-      "reward_std": 0.6804871559143066,
-      "rewards/Response_no_think_reward/mean": 0.4140625,
-      "rewards/Response_no_think_reward/std": 1.2005729675292969,
+      "clip_ratio/high_max": 0.0005630630766972899,
+      "clip_ratio/high_mean": 0.0005630630766972899,
+      "clip_ratio/low_mean": 0.009406007709912956,
+      "clip_ratio/low_min": 0.009406007709912956,
+      "clip_ratio/region_mean": 0.009969070786610246,
+      "completions/clipped_ratio": 0.0546875,
+      "completions/max_length": 43.0,
+      "completions/mean_length": 7.3671875,
+      "completions/min_length": 3.0,
+      "epoch": 0.8955223880597015,
+      "grad_norm": 40.351102862526105,
+      "kl": 37.61209568884806,
+      "learning_rate": 0.00012475268188771627,
+      "loss": 0.25722047686576843,
+      "memory(GiB)": 79.48,
+      "reward": 0.515625,
+      "reward_std": 0.3077147901058197,
+      "rewards/Response_no_think_reward_1/mean": 0.515625,
+      "rewards/Response_no_think_reward_1/std": 0.5883966684341431,
       "step": 75,
-      "train_speed(iter/s)": 0.00233
-    },
-    {
-      "clip_ratio/high_max": 0.004052987133036368,
-      "clip_ratio/high_mean": 0.004052987133036368,
-      "clip_ratio/low_mean": 0.14585177681874484,
-      "clip_ratio/low_min": 0.14585177681874484,
-      "clip_ratio/region_mean": 0.14990476449020207,
-      "epoch": 0.8421052631578947,
-      "grad_norm": 1.4459331642351634,
-      "kl": 4.161040774546564,
-      "learning_rate": 0.00013333738714693956,
-      "loss": -0.005643587093800306,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.00255
+    },
+    {
+      "clip_ratio/high_max": 0.03623928187880665,
+      "clip_ratio/high_mean": 0.03623928187880665,
+      "clip_ratio/low_mean": 0.00604772218503058,
+      "clip_ratio/low_min": 0.00604772218503058,
+      "clip_ratio/region_mean": 0.04228700394742191,
+      "epoch": 0.9074626865671642,
+      "grad_norm": 17.208461552664357,
+      "kl": 6.087301138206385,
+      "learning_rate": 0.00012280911092620297,
+      "loss": 0.16766606271266937,
+      "memory(GiB)": 79.48,
       "step": 76,
-      "train_speed(iter/s)": 0.002347
+      "train_speed(iter/s)": 0.002568
     },
     {
-      "clip_ratio/high_max": 0.00046641789958812296,
-      "clip_ratio/high_mean": 0.00046641789958812296,
-      "clip_ratio/low_mean": 0.0038580247201025486,
-      "clip_ratio/low_min": 0.0038580247201025486,
-      "clip_ratio/region_mean": 0.0043244426196906716,
-      "completions/clipped_ratio": 0.0859375,
-      "completions/max_length": 99.0,
-      "completions/mean_length": 14.8828125,
-      "completions/min_length": 2.0,
-      "epoch": 0.853185595567867,
-      "grad_norm": 2.393092790871116,
-      "kl": 2.344057558570057,
-      "learning_rate": 0.00013159976684859527,
-      "loss": 0.014792806468904018,
-      "memory(GiB)": 80.03,
-      "reward": 0.8046875,
-      "reward_std": 0.515557050704956,
-      "rewards/Response_no_think_reward/mean": 0.8046875,
-      "rewards/Response_no_think_reward/std": 1.2862604856491089,
+      "clip_ratio/high_max": 0.0016891892300918698,
+      "clip_ratio/high_mean": 0.0016891892300918698,
+      "clip_ratio/low_mean": 0.012179655604995787,
+      "clip_ratio/low_min": 0.012179655604995787,
+      "clip_ratio/region_mean": 0.013868844718672335,
+      "completions/clipped_ratio": 0.125,
+      "completions/max_length": 39.0,
+      "completions/mean_length": 7.21875,
+      "completions/min_length": 3.0,
+      "epoch": 0.9194029850746268,
+      "grad_norm": 4.3979207862679734,
+      "kl": 7.126492358453106,
+      "learning_rate": 0.00012085640736741708,
+      "loss": 0.06262560933828354,
+      "memory(GiB)": 79.48,
+      "reward": 0.0546875,
+      "reward_std": 0.39322251081466675,
+      "rewards/Response_no_think_reward_1/mean": 0.0546875,
+      "rewards/Response_no_think_reward_1/std": 0.7561672329902649,
       "step": 77,
-      "train_speed(iter/s)": 0.002292
-    },
-    {
-      "clip_ratio/high_max": 0.005874009046237916,
-      "clip_ratio/high_mean": 0.005874009046237916,
-      "clip_ratio/low_mean": 0.0609364231931977,
-      "clip_ratio/low_min": 0.0609364231931977,
-      "clip_ratio/region_mean": 0.0668104327050969,
-      "epoch": 0.8642659279778393,
-      "grad_norm": 0.9235795825109938,
-      "kl": 3.164612793829292,
-      "learning_rate": 0.00012985148110016947,
-      "loss": 0.001517204917035997,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002562
+    },
+    {
+      "clip_ratio/high_max": 0.026138584362342954,
+      "clip_ratio/high_mean": 0.026138584362342954,
+      "clip_ratio/low_mean": 0.04480157361831516,
+      "clip_ratio/low_min": 0.04480157361831516,
+      "clip_ratio/region_mean": 0.07094015786424279,
+      "epoch": 0.9313432835820895,
+      "grad_norm": 2.916485581854171,
+      "kl": 4.029124836612027,
+      "learning_rate": 0.00011889535305924618,
+      "loss": 0.04432527720928192,
+      "memory(GiB)": 79.48,
       "step": 78,
-      "train_speed(iter/s)": 0.002309
+      "train_speed(iter/s)": 0.002579
     },
     {
-      "clip_ratio/high_max": 0.002277967141708359,
-      "clip_ratio/high_mean": 0.002277967141708359,
-      "clip_ratio/low_mean": 0.0037592062435578555,
-      "clip_ratio/low_min": 0.0037592062435578555,
-      "clip_ratio/region_mean": 0.006037173385266215,
-      "completions/clipped_ratio": 0.0625,
-      "completions/max_length": 87.0,
-      "completions/mean_length": 15.703125,
-      "completions/min_length": 2.0,
-      "epoch": 0.8753462603878116,
-      "grad_norm": 1.666555946903183,
-      "kl": 2.2231151023879647,
-      "learning_rate": 0.00012809311997735696,
-      "loss": 0.012878447771072388,
-      "memory(GiB)": 80.03,
-      "reward": 0.546875,
-      "reward_std": 0.7733994126319885,
-      "rewards/Response_no_think_reward/mean": 0.546875,
-      "rewards/Response_no_think_reward/std": 1.3032931089401245,
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0125213154242374,
+      "clip_ratio/low_min": 0.0125213154242374,
+      "clip_ratio/region_mean": 0.0125213154242374,
+      "completions/clipped_ratio": 0.1015625,
+      "completions/max_length": 41.0,
+      "completions/mean_length": 7.5703125,
+      "completions/min_length": 3.0,
+      "epoch": 0.9432835820895522,
+      "grad_norm": 5.401724081079944,
+      "kl": 2.715451994285104,
+      "learning_rate": 0.00011692673319315541,
+      "loss": 0.025846321135759354,
+      "memory(GiB)": 79.48,
+      "reward": 0.15625,
+      "reward_std": 0.32662156224250793,
+      "rewards/Response_no_think_reward_1/mean": 0.15625,
+      "rewards/Response_no_think_reward_1/std": 0.778019905090332,
       "step": 79,
-      "train_speed(iter/s)": 0.002271
-    },
-    {
-      "clip_ratio/high_max": 0.03329824731918052,
-      "clip_ratio/high_mean": 0.03329824731918052,
-      "clip_ratio/low_mean": 0.017765316180884838,
-      "clip_ratio/low_min": 0.017765316180884838,
-      "clip_ratio/region_mean": 0.05106356361648068,
-      "epoch": 0.8864265927977839,
-      "grad_norm": 1.0568349375834465,
-      "kl": 1.923786539278808,
-      "learning_rate": 0.00012632527695645993,
-      "loss": -0.0005231135291978717,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002572
+    },
+    {
+      "clip_ratio/high_max": 0.02864583395421505,
+      "clip_ratio/high_mean": 0.02864583395421505,
+      "clip_ratio/low_mean": 0.06598890712484717,
+      "clip_ratio/low_min": 0.06598890712484717,
+      "clip_ratio/region_mean": 0.09463474014773965,
+      "epoch": 0.9552238805970149,
+      "grad_norm": 1.7607357402294221,
+      "kl": 3.6259496022976236,
+      "learning_rate": 0.00011495133598980263,
+      "loss": 0.011118962429463863,
+      "memory(GiB)": 79.48,
       "step": 80,
-      "train_speed(iter/s)": 0.002287
+      "train_speed(iter/s)": 0.002589
     },
     {
-      "clip_ratio/high_max": 0.00021551724057644606,
-      "clip_ratio/high_mean": 0.00021551724057644606,
-      "clip_ratio/low_mean": 0.006801646784879267,
-      "clip_ratio/low_min": 0.006801646784879267,
-      "clip_ratio/region_mean": 0.007017164025455713,
-      "completions/clipped_ratio": 0.046875,
-      "completions/max_length": 90.0,
-      "completions/mean_length": 14.03125,
-      "completions/min_length": 2.0,
-      "epoch": 0.8975069252077562,
-      "grad_norm": 9.477113742654934,
-      "kl": 3.0111945159733295,
-      "learning_rate": 0.00012454854871407994,
-      "loss": 0.016959797590970993,
-      "memory(GiB)": 80.03,
-      "reward": 0.2578125,
-      "reward_std": 0.5107755661010742,
-      "rewards/Response_no_think_reward/mean": 0.2578125,
-      "rewards/Response_no_think_reward/std": 1.2874077558517456,
+      "clip_ratio/high_max": 0.0010000000474974513,
+      "clip_ratio/high_mean": 0.0010000000474974513,
+      "clip_ratio/low_mean": 0.009746301162522286,
+      "clip_ratio/low_min": 0.009746301162522286,
+      "clip_ratio/region_mean": 0.010746301151812077,
+      "completions/clipped_ratio": 0.1015625,
+      "completions/max_length": 45.0,
+      "completions/mean_length": 7.421875,
+      "completions/min_length": 3.0,
+      "epoch": 0.9671641791044776,
+      "grad_norm": 6.226807855082435,
+      "kl": 3.5353202793048695,
+      "learning_rate": 0.00011296995238344084,
+      "loss": 0.0407019704580307,
+      "memory(GiB)": 79.48,
+      "reward": 0.2109375,
+      "reward_std": 0.3596799373626709,
+      "rewards/Response_no_think_reward_1/mean": 0.2109375,
+      "rewards/Response_no_think_reward_1/std": 0.78019118309021,
       "step": 81,
-      "train_speed(iter/s)": 0.002263
-    },
-    {
-      "clip_ratio/high_max": 0.012387449765810743,
-      "clip_ratio/high_mean": 0.012387449765810743,
-      "clip_ratio/low_mean": 0.04806764563545585,
-      "clip_ratio/low_min": 0.04806764563545585,
-      "clip_ratio/region_mean": 0.06045509548857808,
-      "epoch": 0.9085872576177285,
-      "grad_norm": 1.1639747912015215,
-      "kl": 3.7243148013949394,
-      "learning_rate": 0.00012276353492572935,
-      "loss": 0.0026693246327340603,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002582
+    },
+    {
+      "clip_ratio/high_max": 0.025790343526750803,
+      "clip_ratio/high_mean": 0.025790343526750803,
+      "clip_ratio/low_mean": 0.027053400757722557,
+      "clip_ratio/low_min": 0.027053400757722557,
+      "clip_ratio/region_mean": 0.052843744633719325,
+      "epoch": 0.9791044776119403,
+      "grad_norm": 43.50825312796573,
+      "kl": 2.1143818981945515,
+      "learning_rate": 0.00011098337570523396,
+      "loss": 0.1263183057308197,
+      "memory(GiB)": 79.48,
       "step": 82,
-      "train_speed(iter/s)": 0.002279
+      "train_speed(iter/s)": 0.002599
     },
     {
-      "clip_ratio/high_max": 0.0004032258002553135,
-      "clip_ratio/high_mean": 0.0004032258002553135,
-      "clip_ratio/low_mean": 0.0061736139468848705,
-      "clip_ratio/low_min": 0.0061736139468848705,
-      "clip_ratio/region_mean": 0.006576839747140184,
-      "completions/clipped_ratio": 0.03125,
-      "completions/max_length": 99.0,
-      "completions/mean_length": 12.625,
-      "completions/min_length": 2.0,
-      "epoch": 0.9196675900277008,
-      "grad_norm": 7.345333737975268,
-      "kl": 1.6264500059187412,
-      "learning_rate": 0.00012097083806343103,
-      "loss": 0.02225027047097683,
-      "memory(GiB)": 80.03,
-      "reward": 0.6640625,
-      "reward_std": 0.4172249436378479,
-      "rewards/Response_no_think_reward/mean": 0.6640625,
-      "rewards/Response_no_think_reward/std": 1.1588573455810547,
+      "clip_ratio/high_max": 0.0045458003878593445,
+      "clip_ratio/high_mean": 0.0045458003878593445,
+      "clip_ratio/low_mean": 0.004786531790159643,
+      "clip_ratio/low_min": 0.004786531790159643,
+      "clip_ratio/region_mean": 0.009332332178018987,
+      "completions/clipped_ratio": 0.0625,
+      "completions/max_length": 23.0,
+      "completions/mean_length": 5.6796875,
+      "completions/min_length": 3.0,
+      "epoch": 1.0119402985074626,
+      "grad_norm": 5.294876949688284,
+      "kl": 1.5498897642164593,
+      "learning_rate": 0.00010899240136561299,
+      "loss": 0.011598482728004456,
+      "memory(GiB)": 79.48,
+      "reward": 0.4453125,
+      "reward_std": 0.1938612163066864,
+      "rewards/Response_no_think_reward_1/mean": 0.4453125,
+      "rewards/Response_no_think_reward_1/std": 0.7187981605529785,
       "step": 83,
-      "train_speed(iter/s)": 0.00227
-    },
-    {
-      "clip_ratio/high_max": 0.0035271961241960526,
-      "clip_ratio/high_mean": 0.0035271961241960526,
-      "clip_ratio/low_mean": 0.02831025089835748,
-      "clip_ratio/low_min": 0.02831025089835748,
-      "clip_ratio/region_mean": 0.03183744702255353,
-      "epoch": 0.9307479224376731,
-      "grad_norm": 84.3494363093827,
-      "kl": 10.616167868487537,
-      "learning_rate": 0.00011917106319237386,
-      "loss": 0.17172452807426453,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002589
+    },
+    {
+      "clip_ratio/high_max": 0.025593542493879795,
+      "clip_ratio/high_mean": 0.025593542493879795,
+      "clip_ratio/low_mean": 0.025054577039554715,
+      "clip_ratio/low_min": 0.025054577039554715,
+      "clip_ratio/region_mean": 0.05064811953343451,
+      "epoch": 1.0238805970149254,
+      "grad_norm": 2.6737027093734143,
+      "kl": 1.398324867649535,
+      "learning_rate": 0.00010699782653579973,
+      "loss": -0.00029344583163037896,
+      "memory(GiB)": 79.48,
       "step": 84,
-      "train_speed(iter/s)": 0.002286
-    },
-    {
-      "clip_ratio/high_max": 0.0011701860348694026,
-      "clip_ratio/high_mean": 0.0011701860348694026,
-      "clip_ratio/low_mean": 0.014238029951229692,
-      "clip_ratio/low_min": 0.014238029951229692,
-      "clip_ratio/region_mean": 0.015408215986099094,
-      "completions/clipped_ratio": 0.140625,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 19.7578125,
-      "completions/min_length": 4.0,
-      "epoch": 0.9418282548476454,
-      "grad_norm": 1.9907484000781221,
-      "kl": 2.995624510163907,
-      "learning_rate": 0.00011736481776669306,
-      "loss": 0.01750401221215725,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002605
+    },
+    {
+      "clip_ratio/high_max": 0.0017993993242271245,
+      "clip_ratio/high_mean": 0.0017993993242271245,
+      "clip_ratio/low_mean": 0.01037665200419724,
+      "clip_ratio/low_min": 0.01037665200419724,
+      "clip_ratio/region_mean": 0.012176051270216703,
+      "completions/clipped_ratio": 0.171875,
+      "completions/max_length": 42.0,
+      "completions/mean_length": 6.125,
+      "completions/min_length": 3.0,
+      "epoch": 1.035820895522388,
+      "grad_norm": 2.5145149964606297,
+      "kl": 3.1435449587497715,
+      "learning_rate": 0.00010500044982862519,
+      "loss": 0.024344690144062042,
+      "memory(GiB)": 79.48,
       "reward": 0.4765625,
-      "reward_std": 0.4854952394962311,
-      "rewards/Response_no_think_reward/mean": 0.4765625,
-      "rewards/Response_no_think_reward/std": 1.1080580949783325,
+      "reward_std": 0.2772369980812073,
+      "rewards/Response_no_think_reward_1/mean": 0.4765625,
+      "rewards/Response_no_think_reward_1/std": 0.8414914011955261,
       "step": 85,
-      "train_speed(iter/s)": 0.002274
-    },
-    {
-      "clip_ratio/high_max": 0.008182557401596569,
-      "clip_ratio/high_mean": 0.008182557401596569,
-      "clip_ratio/low_mean": 0.11124554229900241,
-      "clip_ratio/low_min": 0.11124554229900241,
-      "clip_ratio/region_mean": 0.11942810016626026,
-      "epoch": 0.9529085872576177,
-      "grad_norm": 1.9564641583381899,
-      "kl": 4.675610944104847,
-      "learning_rate": 0.00011555271142444433,
-      "loss": 0.009754904545843601,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002595
+    },
+    {
+      "clip_ratio/high_max": 0.021866617840714753,
+      "clip_ratio/high_mean": 0.021866617840714753,
+      "clip_ratio/low_mean": 0.07049822807312012,
+      "clip_ratio/low_min": 0.07049822807312012,
+      "clip_ratio/region_mean": 0.09236484486609697,
+      "epoch": 1.0477611940298508,
+      "grad_norm": 2.607332219212116,
+      "kl": 4.641115245953188,
+      "learning_rate": 0.00010300107097877114,
+      "loss": 0.010013583116233349,
+      "memory(GiB)": 79.48,
       "step": 86,
-      "train_speed(iter/s)": 0.002289
+      "train_speed(iter/s)": 0.00261
     },
     {
-      "clip_ratio/high_max": 0.0004139889351790771,
-      "clip_ratio/high_mean": 0.0004139889351790771,
-      "clip_ratio/low_mean": 0.009557914454489946,
-      "clip_ratio/low_min": 0.009557914454489946,
-      "clip_ratio/region_mean": 0.0099719034624286,
-      "completions/clipped_ratio": 0.0859375,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 17.90625,
-      "completions/min_length": 2.0,
-      "epoch": 0.96398891966759,
-      "grad_norm": 1.805689670049186,
-      "kl": 3.4809365491382778,
-      "learning_rate": 0.00011373535578184082,
-      "loss": 0.0213579460978508,
-      "memory(GiB)": 80.03,
-      "reward": 0.5859375,
-      "reward_std": 0.5313136577606201,
-      "rewards/Response_no_think_reward/mean": 0.5859375,
-      "rewards/Response_no_think_reward/std": 1.1939964294433594,
+      "clip_ratio/high_max": 0.0018382353009656072,
+      "clip_ratio/high_mean": 0.0018382353009656072,
+      "clip_ratio/low_mean": 0.0032051282469183207,
+      "clip_ratio/low_min": 0.0032051282469183207,
+      "clip_ratio/region_mean": 0.005043363547883928,
+      "completions/clipped_ratio": 0.03125,
+      "completions/max_length": 20.0,
+      "completions/mean_length": 5.4375,
+      "completions/min_length": 3.0,
+      "epoch": 1.0597014925373134,
+      "grad_norm": 8.56914849705318,
+      "kl": 0.7865612751156732,
+      "learning_rate": 0.00010100049052256235,
+      "loss": 0.013117787428200245,
+      "memory(GiB)": 79.48,
+      "reward": 0.53125,
+      "reward_std": 0.1860944628715515,
+      "rewards/Response_no_think_reward_1/mean": 0.53125,
+      "rewards/Response_no_think_reward_1/std": 0.7202088236808777,
       "step": 87,
-      "train_speed(iter/s)": 0.002277
-    },
-    {
-      "clip_ratio/high_max": 0.009385127894347534,
-      "clip_ratio/high_mean": 0.009385127894347534,
-      "clip_ratio/low_mean": 0.08647578035015613,
-      "clip_ratio/low_min": 0.08647578035015613,
-      "clip_ratio/region_mean": 0.095860909903422,
-      "epoch": 0.9750692520775623,
-      "grad_norm": 2.411018383110019,
-      "kl": 4.761912747140741,
-      "learning_rate": 0.00011191336422682237,
-      "loss": 0.01895134523510933,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002597
+    },
+    {
+      "clip_ratio/high_max": 0.009521116502583027,
+      "clip_ratio/high_mean": 0.009521116502583027,
+      "clip_ratio/low_mean": 0.0203368536895141,
+      "clip_ratio/low_min": 0.0203368536895141,
+      "clip_ratio/region_mean": 0.029857969493605196,
+      "epoch": 1.0716417910447762,
+      "grad_norm": 3.3791848142288035,
+      "kl": 0.6961851019823371,
+      "learning_rate": 9.899950947743767e-05,
+      "loss": 0.00659452797845006,
+      "memory(GiB)": 79.48,
       "step": 88,
-      "train_speed(iter/s)": 0.002291
+      "train_speed(iter/s)": 0.002612
     },
     {
-      "clip_ratio/high_max": 0.0031819915457163006,
-      "clip_ratio/high_mean": 0.0031819915457163006,
-      "clip_ratio/low_mean": 0.012216789647936821,
-      "clip_ratio/low_min": 0.012216789647936821,
-      "clip_ratio/region_mean": 0.015398780989926308,
-      "completions/clipped_ratio": 0.0625,
-      "completions/max_length": 64.0,
-      "completions/mean_length": 11.28125,
-      "completions/min_length": 2.0,
-      "epoch": 0.9861495844875346,
-      "grad_norm": 2.2626010793409574,
-      "kl": 6.020935451146215,
-      "learning_rate": 0.00011008735171202684,
-      "loss": 0.031282562762498856,
-      "memory(GiB)": 80.03,
-      "reward": 0.2890625,
-      "reward_std": 0.847247302532196,
-      "rewards/Response_no_think_reward/mean": 0.2890625,
-      "rewards/Response_no_think_reward/std": 1.1915208101272583,
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.004996612435206771,
+      "clip_ratio/low_min": 0.004996612435206771,
+      "clip_ratio/region_mean": 0.004996612435206771,
+      "completions/clipped_ratio": 0.1171875,
+      "completions/max_length": 31.0,
+      "completions/mean_length": 5.2265625,
+      "completions/min_length": 3.0,
+      "epoch": 1.0835820895522388,
+      "grad_norm": 3.6519414582158323,
+      "kl": 4.849585925694555,
+      "learning_rate": 9.699892902122886e-05,
+      "loss": 0.06144208833575249,
+      "memory(GiB)": 79.48,
+      "reward": 0.1640625,
+      "reward_std": 0.3352247476577759,
+      "rewards/Response_no_think_reward_1/mean": 0.1640625,
+      "rewards/Response_no_think_reward_1/std": 0.8303053379058838,
       "step": 89,
-      "train_speed(iter/s)": 0.002254
-    },
-    {
-      "clip_ratio/high_max": 0.024421937006991357,
-      "clip_ratio/high_mean": 0.024421937006991357,
-      "clip_ratio/low_mean": 0.0521851975354366,
-      "clip_ratio/low_min": 0.0521851975354366,
-      "clip_ratio/region_mean": 0.07660713430959731,
-      "epoch": 0.997229916897507,
-      "grad_norm": 1.8302536272850396,
-      "kl": 6.024846433196217,
-      "learning_rate": 0.00010825793454723325,
-      "loss": 0.017346249893307686,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002606
+    },
+    {
+      "clip_ratio/high_max": 0.017538156593218446,
+      "clip_ratio/high_mean": 0.017538156593218446,
+      "clip_ratio/low_mean": 0.03697483101859689,
+      "clip_ratio/low_min": 0.03697483101859689,
+      "clip_ratio/region_mean": 0.054512988310307264,
+      "epoch": 1.0955223880597016,
+      "grad_norm": 13.104523239395307,
+      "kl": 2.37760816863738,
+      "learning_rate": 9.499955017137484e-05,
+      "loss": 0.10894370079040527,
+      "memory(GiB)": 79.48,
       "step": 90,
-      "train_speed(iter/s)": 0.002268
+      "train_speed(iter/s)": 0.002621
     },
     {
-      "clip_ratio/high_max": 0.0011776478204410523,
-      "clip_ratio/high_mean": 0.0011776478204410523,
-      "clip_ratio/low_mean": 0.0035608798498287797,
-      "clip_ratio/low_min": 0.0035608798498287797,
-      "clip_ratio/region_mean": 0.004738527670269832,
-      "completions/clipped_ratio": 0.125,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 24.046875,
-      "completions/min_length": 3.0,
-      "epoch": 1.0110803324099722,
-      "grad_norm": 2.3698869054826233,
-      "kl": 5.333947827733937,
-      "learning_rate": 0.00010642573019134703,
-      "loss": 0.028505954891443253,
-      "memory(GiB)": 80.03,
-      "reward": 0.84375,
-      "reward_std": 0.6655995845794678,
-      "rewards/Response_no_think_reward/mean": 0.84375,
-      "rewards/Response_no_think_reward/std": 1.2065274715423584,
+      "clip_ratio/high_max": 0.004422514699399471,
+      "clip_ratio/high_mean": 0.004422514699399471,
+      "clip_ratio/low_mean": 0.007080804323777556,
+      "clip_ratio/low_min": 0.007080804323777556,
+      "clip_ratio/region_mean": 0.01150331913959235,
+      "completions/clipped_ratio": 0.046875,
+      "completions/max_length": 7.0,
+      "completions/mean_length": 4.6640625,
+      "completions/min_length": 3.0,
+      "epoch": 1.1074626865671642,
+      "grad_norm": 3.292956583617785,
+      "kl": 3.1563512758002616,
+      "learning_rate": 9.30021734642003e-05,
+      "loss": 0.03713127598166466,
+      "memory(GiB)": 79.48,
+      "reward": 0.421875,
+      "reward_std": 0.2630079686641693,
+      "rewards/Response_no_think_reward_1/mean": 0.421875,
+      "rewards/Response_no_think_reward_1/std": 0.7382611632347107,
       "step": 91,
-      "train_speed(iter/s)": 0.002232
-    },
-    {
-      "clip_ratio/high_max": 0.02221274602925405,
-      "clip_ratio/high_mean": 0.02221274602925405,
-      "clip_ratio/low_mean": 0.04682085904642008,
-      "clip_ratio/low_min": 0.04682085904642008,
-      "clip_ratio/region_mean": 0.0690336050465703,
-      "epoch": 1.0221606648199446,
-      "grad_norm": 1.377879165623118,
-      "kl": 6.767704317186144,
-      "learning_rate": 0.00010459135704399718,
-      "loss": 0.01659482903778553,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002603
+    },
+    {
+      "clip_ratio/high_max": 0.012390351388603449,
+      "clip_ratio/high_mean": 0.012390351388603449,
+      "clip_ratio/low_mean": 0.01934311306104064,
+      "clip_ratio/low_min": 0.01934311306104064,
+      "clip_ratio/region_mean": 0.03173346375115216,
+      "epoch": 1.1194029850746268,
+      "grad_norm": 6.523254504794387,
+      "kl": 2.134530827694107,
+      "learning_rate": 9.100759863438702e-05,
+      "loss": 0.04548652470111847,
+      "memory(GiB)": 79.48,
       "step": 92,
-      "train_speed(iter/s)": 0.002246
+      "train_speed(iter/s)": 0.002618
     },
     {
-      "clip_ratio/high_max": 0.0007944914977997541,
-      "clip_ratio/high_mean": 0.0007944914977997541,
-      "clip_ratio/low_mean": 0.005214237666223198,
-      "clip_ratio/low_min": 0.005214237666223198,
-      "clip_ratio/region_mean": 0.006008729164022952,
-      "completions/clipped_ratio": 0.046875,
-      "completions/max_length": 79.0,
-      "completions/mean_length": 11.1953125,
-      "completions/min_length": 2.0,
-      "epoch": 1.0332409972299168,
-      "grad_norm": 1.128328936062302,
-      "kl": 2.6099998716963455,
-      "learning_rate": 0.00010275543423681621,
-      "loss": 0.01443527452647686,
-      "memory(GiB)": 80.03,
-      "reward": 0.5703125,
-      "reward_std": 0.34169840812683105,
-      "rewards/Response_no_think_reward/mean": 0.5703125,
-      "rewards/Response_no_think_reward/std": 1.26543128490448,
+      "clip_ratio/high_max": 0.0018115942366421223,
+      "clip_ratio/high_mean": 0.0018115942366421223,
+      "clip_ratio/low_mean": 0.010487846680916846,
+      "clip_ratio/low_min": 0.010487846680916846,
+      "clip_ratio/region_mean": 0.012299440917558968,
+      "completions/clipped_ratio": 0.078125,
+      "completions/max_length": 17.0,
+      "completions/mean_length": 4.7109375,
+      "completions/min_length": 3.0,
+      "epoch": 1.1313432835820896,
+      "grad_norm": 2.8653500050121345,
+      "kl": 3.8929527901345864,
+      "learning_rate": 8.901662429476607e-05,
+      "loss": 0.04490555822849274,
+      "memory(GiB)": 79.48,
+      "reward": 0.1796875,
+      "reward_std": 0.25065141916275024,
+      "rewards/Response_no_think_reward_1/mean": 0.1796875,
+      "rewards/Response_no_think_reward_1/std": 0.9173188805580139,
       "step": 93,
-      "train_speed(iter/s)": 0.00223
-    },
-    {
-      "clip_ratio/high_max": 0.024202606233302504,
-      "clip_ratio/high_mean": 0.024202606233302504,
-      "clip_ratio/low_mean": 0.017650849069468677,
-      "clip_ratio/low_min": 0.017650849069468677,
-      "clip_ratio/region_mean": 0.04185345536097884,
-      "epoch": 1.0443213296398892,
-      "grad_norm": 0.5979565041663649,
-      "kl": 1.9247902451315895,
-      "learning_rate": 0.00010091858142447265,
-      "loss": 0.005481676664203405,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.00261
+    },
+    {
+      "clip_ratio/high_max": 0.019497282337397337,
+      "clip_ratio/high_mean": 0.019497282337397337,
+      "clip_ratio/low_mean": 0.05975891789421439,
+      "clip_ratio/low_min": 0.05975891789421439,
+      "clip_ratio/region_mean": 0.07925620023161173,
+      "epoch": 1.1432835820895522,
+      "grad_norm": 8.612775013595648,
+      "kl": 4.982872966560535,
+      "learning_rate": 8.703004761655917e-05,
+      "loss": 0.07919233292341232,
+      "memory(GiB)": 79.48,
       "step": 94,
-      "train_speed(iter/s)": 0.002243
-    },
-    {
-      "clip_ratio/high_max": 0.002601411077193916,
-      "clip_ratio/high_mean": 0.002601411077193916,
-      "clip_ratio/low_mean": 0.005088170932140201,
-      "clip_ratio/low_min": 0.005088170932140201,
-      "clip_ratio/region_mean": 0.007689581951126456,
-      "completions/clipped_ratio": 0.09375,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 18.25,
-      "completions/min_length": 4.0,
-      "epoch": 1.0554016620498614,
-      "grad_norm": 2.0648308983451917,
-      "kl": 1.713204285595566,
-      "learning_rate": 9.908141857552737e-05,
-      "loss": 0.015094820410013199,
-      "memory(GiB)": 80.03,
-      "reward": 0.8046875,
-      "reward_std": 0.47176384925842285,
-      "rewards/Response_no_think_reward/mean": 0.8046875,
-      "rewards/Response_no_think_reward/std": 1.1708977222442627,
+      "train_speed(iter/s)": 0.002625
+    },
+    {
+      "clip_ratio/high_max": 0.0027173913549631834,
+      "clip_ratio/high_mean": 0.0027173913549631834,
+      "clip_ratio/low_mean": 0.005094834603369236,
+      "clip_ratio/low_min": 0.005094834603369236,
+      "clip_ratio/region_mean": 0.007812225958332419,
+      "completions/clipped_ratio": 0.0859375,
+      "completions/max_length": 10.0,
+      "completions/mean_length": 4.9375,
+      "completions/min_length": 3.0,
+      "epoch": 1.155223880597015,
+      "grad_norm": 10.776677331970285,
+      "kl": 9.849615207003808,
+      "learning_rate": 8.504866401019737e-05,
+      "loss": 0.1374131143093109,
+      "memory(GiB)": 79.48,
+      "reward": 0.3203125,
+      "reward_std": 0.14966703951358795,
+      "rewards/Response_no_think_reward_1/mean": 0.3203125,
+      "rewards/Response_no_think_reward_1/std": 0.6266219019889832,
       "step": 95,
-      "train_speed(iter/s)": 0.002221
-    },
-    {
-      "clip_ratio/high_max": 0.01881888063508086,
-      "clip_ratio/high_mean": 0.01881888063508086,
-      "clip_ratio/low_mean": 0.026617751631420106,
-      "clip_ratio/low_min": 0.026617751631420106,
-      "clip_ratio/region_mean": 0.0454366315389052,
-      "epoch": 1.0664819944598338,
-      "grad_norm": 1.5544793796965792,
-      "kl": 1.6157679219031706,
-      "learning_rate": 9.724456576318381e-05,
-      "loss": 0.012739625759422779,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002622
+    },
+    {
+      "clip_ratio/high_max": 0.019021738320589066,
+      "clip_ratio/high_mean": 0.019021738320589066,
+      "clip_ratio/low_mean": 0.005626177066005766,
+      "clip_ratio/low_min": 0.005626177066005766,
+      "clip_ratio/region_mean": 0.024647915386594832,
+      "epoch": 1.1671641791044776,
+      "grad_norm": 3.0583039624376402,
+      "kl": 4.468168576029711,
+      "learning_rate": 8.307326680684461e-05,
+      "loss": 0.07699939608573914,
+      "memory(GiB)": 79.48,
       "step": 96,
-      "train_speed(iter/s)": 0.002234
+      "train_speed(iter/s)": 0.002636
     },
     {
-      "clip_ratio/high_max": 0.0020559211261570454,
-      "clip_ratio/high_mean": 0.0020559211261570454,
-      "clip_ratio/low_mean": 0.0043267360888421535,
-      "clip_ratio/low_min": 0.0043267360888421535,
-      "clip_ratio/region_mean": 0.006382657098583877,
-      "completions/clipped_ratio": 0.015625,
-      "completions/max_length": 41.0,
-      "completions/mean_length": 11.046875,
-      "completions/min_length": 3.0,
-      "epoch": 1.077562326869806,
-      "grad_norm": 2.5531547462941115,
-      "kl": 1.4817758100107312,
-      "learning_rate": 9.540864295600283e-05,
-      "loss": 0.016522858291864395,
-      "memory(GiB)": 80.03,
-      "reward": 0.9453125,
-      "reward_std": 0.4848037362098694,
-      "rewards/Response_no_think_reward/mean": 0.9453125,
-      "rewards/Response_no_think_reward/std": 1.1454023122787476,
+      "clip_ratio/high_max": 0.007378174108453095,
+      "clip_ratio/high_mean": 0.007378174108453095,
+      "clip_ratio/low_mean": 0.00924376118928194,
+      "clip_ratio/low_min": 0.00924376118928194,
+      "clip_ratio/region_mean": 0.016621935297735035,
+      "completions/clipped_ratio": 0.0546875,
+      "completions/max_length": 12.0,
+      "completions/mean_length": 4.9453125,
+      "completions/min_length": 3.0,
+      "epoch": 1.1791044776119404,
+      "grad_norm": 9.444776640622303,
+      "kl": 3.863095655280631,
+      "learning_rate": 8.110464694075383e-05,
+      "loss": 0.02421477437019348,
+      "memory(GiB)": 79.48,
+      "reward": 0.4375,
+      "reward_std": 0.25460314750671387,
+      "rewards/Response_no_think_reward_1/mean": 0.4375,
+      "rewards/Response_no_think_reward_1/std": 0.7290377616882324,
       "step": 97,
-      "train_speed(iter/s)": 0.00222
-    },
-    {
-      "clip_ratio/high_max": 0.01358306163456291,
-      "clip_ratio/high_mean": 0.01358306163456291,
-      "clip_ratio/low_mean": 0.019928007503040135,
-      "clip_ratio/low_min": 0.019928007503040135,
-      "clip_ratio/region_mean": 0.03351106960326433,
-      "epoch": 1.0886426592797784,
-      "grad_norm": 1.0799722223140724,
-      "kl": 1.3886900492943823,
-      "learning_rate": 9.357426980865301e-05,
-      "loss": -0.0007159767556004226,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002626
+    },
+    {
+      "clip_ratio/high_max": 0.02187895681709051,
+      "clip_ratio/high_mean": 0.02187895681709051,
+      "clip_ratio/low_mean": 0.005211503244936466,
+      "clip_ratio/low_min": 0.005211503244936466,
+      "clip_ratio/region_mean": 0.027090460062026978,
+      "epoch": 1.191044776119403,
+      "grad_norm": 4.753171281552747,
+      "kl": 3.498804785136599,
+      "learning_rate": 7.914359263258295e-05,
+      "loss": 0.02348589338362217,
+      "memory(GiB)": 79.48,
       "step": 98,
-      "train_speed(iter/s)": 0.002234
+      "train_speed(iter/s)": 0.00264
     },
     {
-      "clip_ratio/high_max": 0.001544758939417079,
-      "clip_ratio/high_mean": 0.001544758939417079,
-      "clip_ratio/low_mean": 0.0037943847200949676,
-      "clip_ratio/low_min": 0.0037943847200949676,
-      "clip_ratio/region_mean": 0.005339143652236089,
+      "clip_ratio/high_max": 0.0014204545877873898,
+      "clip_ratio/high_mean": 0.0014204545877873898,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0014204545877873898,
       "completions/clipped_ratio": 0.0546875,
-      "completions/max_length": 101.0,
-      "completions/mean_length": 18.546875,
-      "completions/min_length": 3.0,
-      "epoch": 1.0997229916897506,
-      "grad_norm": 0.943375664401446,
-      "kl": 1.7392279328778386,
-      "learning_rate": 9.174206545276677e-05,
-      "loss": 0.014223725534975529,
-      "memory(GiB)": 80.03,
-      "reward": 1.1015625,
-      "reward_std": 0.39179152250289917,
-      "rewards/Response_no_think_reward/mean": 1.1015625,
-      "rewards/Response_no_think_reward/std": 1.2221051454544067,
+      "completions/max_length": 11.0,
+      "completions/mean_length": 5.5859375,
+      "completions/min_length": 3.0,
+      "epoch": 1.2029850746268655,
+      "grad_norm": 4.853951435141314,
+      "kl": 7.24955918840169,
+      "learning_rate": 7.719088907379706e-05,
+      "loss": 0.08564330637454987,
+      "memory(GiB)": 79.48,
+      "reward": 0.359375,
+      "reward_std": 0.1706565022468567,
+      "rewards/Response_no_think_reward_1/mean": 0.359375,
+      "rewards/Response_no_think_reward_1/std": 0.6608795523643494,
       "step": 99,
-      "train_speed(iter/s)": 0.002219
-    },
-    {
-      "clip_ratio/high_max": 0.01209607784403488,
-      "clip_ratio/high_mean": 0.01209607784403488,
-      "clip_ratio/low_mean": 0.011367922488716431,
-      "clip_ratio/low_min": 0.011367922488716431,
-      "clip_ratio/region_mean": 0.023464000318199396,
-      "epoch": 1.110803324099723,
-      "grad_norm": 0.8253425068817393,
-      "kl": 1.7898913251701742,
-      "learning_rate": 8.991264828797319e-05,
-      "loss": 0.006411677226424217,
-      "memory(GiB)": 80.03,
+      "train_speed(iter/s)": 0.002636
+    },
+    {
+      "clip_ratio/high_max": 0.01420454541221261,
+      "clip_ratio/high_mean": 0.01420454541221261,
+      "clip_ratio/low_mean": 0.0028409091755747795,
+      "clip_ratio/low_min": 0.0028409091755747795,
+      "clip_ratio/region_mean": 0.017045455053448677,
+      "epoch": 1.2149253731343284,
+      "grad_norm": 2.2986744180745324,
+      "kl": 4.886410776793127,
+      "learning_rate": 7.524731811228374e-05,
+      "loss": 0.06466365605592728,
+      "memory(GiB)": 79.48,
       "step": 100,
-      "train_speed(iter/s)": 0.002232
+      "train_speed(iter/s)": 0.00265
     },
     {
-      "clip_ratio/high_max": 0.0038593837525695562,
-      "clip_ratio/high_mean": 0.0038593837525695562,
-      "clip_ratio/low_mean": 0.007897521747509018,
-      "clip_ratio/low_min": 0.007897521747509018,
-      "clip_ratio/region_mean": 0.011756905500078574,
-      "completions/clipped_ratio": 0.140625,
-      "completions/max_length": 51.0,
-      "completions/mean_length": 18.0390625,
-      "completions/min_length": 2.0,
-      "epoch": 1.1218836565096952,
-      "grad_norm": 2.1091313211648033,
-      "kl": 4.433779507409781,
-      "learning_rate": 8.808663577317764e-05,
-      "loss": 0.026584401726722717,
-      "memory(GiB)": 77.36,
-      "reward": 0.234375,
-      "reward_std": 0.353938490152359,
-      "rewards/Response_no_think_reward_1/mean": 0.234375,
-      "rewards/Response_no_think_reward_1/std": 0.8275223970413208,
+      "clip_ratio/high_max": 0.00620507646817714,
+      "clip_ratio/high_mean": 0.00620507646817714,
+      "clip_ratio/low_mean": 0.008002021699212492,
+      "clip_ratio/low_min": 0.008002021699212492,
+      "clip_ratio/region_mean": 0.014207098283804953,
+      "completions/clipped_ratio": 0.0859375,
+      "completions/max_length": 19.0,
+      "completions/mean_length": 6.2890625,
+      "completions/min_length": 3.0,
+      "epoch": 1.2268656716417912,
+      "grad_norm": 35.84498974880178,
+      "kl": 13.474940237792907,
+      "learning_rate": 7.331365793930698e-05,
+      "loss": 0.23400571942329407,
+      "memory(GiB)": 79.48,
+      "reward": 0.3671875,
+      "reward_std": 0.1965562105178833,
+      "rewards/Response_no_think_reward_1/mean": 0.3671875,
+      "rewards/Response_no_think_reward_1/std": 0.7410472631454468,
       "step": 101,
-      "train_speed(iter/s)": 0.112917
-    },
-    {
-      "clip_ratio/high_max": 0.0257808348396793,
-      "clip_ratio/high_mean": 0.0257808348396793,
-      "clip_ratio/low_mean": 0.013182859780499712,
-      "clip_ratio/low_min": 0.013182859780499712,
-      "clip_ratio/region_mean": 0.038963694794801995,
-      "epoch": 1.1329639889196677,
-      "grad_norm": 0.8069647208167382,
-      "kl": 2.7774715912528336,
-      "learning_rate": 8.626464421815919e-05,
-      "loss": 0.008937789127230644,
-      "memory(GiB)": 77.36,
+      "train_speed(iter/s)": 0.002642
+    },
+    {
+      "clip_ratio/high_max": 0.011945601785555482,
+      "clip_ratio/high_mean": 0.011945601785555482,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.011945601785555482,
+      "epoch": 1.2388059701492538,
+      "grad_norm": 27.884607230951485,
+      "kl": 1.4154490869450456,
+      "learning_rate": 7.139068277792523e-05,
+      "loss": 0.2148125171661377,
+      "memory(GiB)": 79.48,
       "step": 102,
-      "train_speed(iter/s)": 0.09385
+      "train_speed(iter/s)": 0.002655
     },
     {
-      "clip_ratio/high_max": 0.00028669723542407155,
-      "clip_ratio/high_mean": 0.00028669723542407155,
-      "clip_ratio/low_mean": 0.0008802816737443209,
-      "clip_ratio/low_min": 0.0008802816737443209,
-      "clip_ratio/region_mean": 0.0011669789091683924,
-      "completions/clipped_ratio": 0.109375,
-      "completions/max_length": 51.0,
-      "completions/mean_length": 18.140625,
-      "completions/min_length": 3.0,
-      "epoch": 1.1440443213296398,
-      "grad_norm": 3.4014714828505634,
-      "kl": 1.8993340344168246,
-      "learning_rate": 8.444728857555572e-05,
-      "loss": 0.029605647549033165,
-      "memory(GiB)": 77.47,
-      "reward": 0.390625,
-      "reward_std": 0.1173202246427536,
-      "rewards/Response_no_think_reward_1/mean": 0.390625,
-      "rewards/Response_no_think_reward_1/std": 0.8440096974372864,
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.007009345572441816,
+      "clip_ratio/low_min": 0.007009345572441816,
+      "clip_ratio/region_mean": 0.007009345572441816,
+      "completions/clipped_ratio": 0.0546875,
+      "completions/max_length": 40.0,
+      "completions/mean_length": 5.640625,
+      "completions/min_length": 3.0,
+      "epoch": 1.2507462686567163,
+      "grad_norm": 2.3687718583378876,
+      "kl": 1.394652240909636,
+      "learning_rate": 6.94791625729983e-05,
+      "loss": 0.018386229872703552,
+      "memory(GiB)": 79.48,
+      "reward": 0.6328125,
+      "reward_std": 0.20033246278762817,
+      "rewards/Response_no_think_reward_1/mean": 0.6328125,
+      "rewards/Response_no_think_reward_1/std": 0.6258360743522644,
       "step": 103,
-      "train_speed(iter/s)": 0.053798
-    },
-    {
-      "clip_ratio/high_max": 0.0013847328373230994,
-      "clip_ratio/high_mean": 0.0013847328373230994,
-      "clip_ratio/low_mean": 0.004156515002250671,
-      "clip_ratio/low_min": 0.004156515002250671,
-      "clip_ratio/region_mean": 0.00554124778136611,
-      "epoch": 1.1551246537396123,
-      "grad_norm": 0.6574286317590833,
-      "kl": 0.8905834904871881,
-      "learning_rate": 8.263518223330697e-05,
-      "loss": 0.007299074437469244,
-      "memory(GiB)": 77.47,
+      "train_speed(iter/s)": 0.002645
+    },
+    {
+      "clip_ratio/high_max": 0.031393789453431964,
+      "clip_ratio/high_mean": 0.031393789453431964,
+      "clip_ratio/low_mean": 0.01482532313093543,
+      "clip_ratio/low_min": 0.01482532313093543,
+      "clip_ratio/region_mean": 0.04621911211870611,
+      "epoch": 1.2626865671641792,
+      "grad_norm": 1.3586707954485446,
+      "kl": 1.0225394079461694,
+      "learning_rate": 6.757986268290712e-05,
+      "loss": 0.011677918955683708,
+      "memory(GiB)": 79.48,
       "step": 104,
-      "train_speed(iter/s)": 0.049358
+      "train_speed(iter/s)": 0.002658
     },
     {
-      "clip_ratio/high_max": 0.00036549707874655724,
-      "clip_ratio/high_mean": 0.00036549707874655724,
-      "clip_ratio/low_mean": 0.003542276710504666,
-      "clip_ratio/low_min": 0.003542276710504666,
-      "clip_ratio/region_mean": 0.003907773789251223,
-      "completions/clipped_ratio": 0.0546875,
-      "completions/max_length": 51.0,
-      "completions/mean_length": 12.1015625,
-      "completions/min_length": 2.0,
-      "epoch": 1.1662049861495845,
-      "grad_norm": 1.418255621232663,
-      "kl": 1.0295969531871378,
-      "learning_rate": 8.082893680762619e-05,
-      "loss": 0.008070322684943676,
-      "memory(GiB)": 77.47,
-      "reward": 0.5625,
-      "reward_std": 0.283821702003479,
-      "rewards/Response_no_think_reward_1/mean": 0.5625,
-      "rewards/Response_no_think_reward_1/std": 0.6728714108467102,
+      "clip_ratio/high_max": 0.00795477326028049,
+      "clip_ratio/high_mean": 0.00795477326028049,
+      "clip_ratio/low_mean": 0.007858287775889039,
+      "clip_ratio/low_min": 0.007858287775889039,
+      "clip_ratio/region_mean": 0.015813061269000173,
+      "completions/clipped_ratio": 0.0625,
+      "completions/max_length": 32.0,
+      "completions/mean_length": 5.828125,
+      "completions/min_length": 3.0,
+      "epoch": 1.2746268656716417,
+      "grad_norm": 16.43559312212207,
+      "kl": 1.8490474935271664,
+      "learning_rate": 6.569354357311014e-05,
+      "loss": 0.03543264418840408,
+      "memory(GiB)": 79.48,
+      "reward": 0.3671875,
+      "reward_std": 0.31494253873825073,
+      "rewards/Response_no_think_reward_1/mean": 0.3671875,
+      "rewards/Response_no_think_reward_1/std": 0.7923958897590637,
       "step": 105,
-      "train_speed(iter/s)": 0.035507
-    },
-    {
-      "clip_ratio/high_max": 0.015719514689408243,
-      "clip_ratio/high_mean": 0.015719514689408243,
-      "clip_ratio/low_mean": 0.010580109432339668,
-      "clip_ratio/low_min": 0.010580109432339668,
-      "clip_ratio/region_mean": 0.02629962412174791,
-      "epoch": 1.1772853185595569,
-      "grad_norm": 0.9124403050854863,
-      "kl": 0.933376073371619,
-      "learning_rate": 7.902916193656898e-05,
-      "loss": -0.002763347467407584,
-      "memory(GiB)": 79.43,
+      "train_speed(iter/s)": 0.002652
+    },
+    {
+      "clip_ratio/high_max": 0.011914517963305116,
+      "clip_ratio/high_mean": 0.011914517963305116,
+      "clip_ratio/low_mean": 0.03227056178729981,
+      "clip_ratio/low_min": 0.03227056178729981,
+      "clip_ratio/region_mean": 0.04418508114758879,
+      "epoch": 1.2865671641791045,
+      "grad_norm": 2.5408508655554636,
+      "kl": 1.8903868702007571,
+      "learning_rate": 6.382096051165847e-05,
+      "loss": 0.014127654023468494,
+      "memory(GiB)": 79.48,
       "step": 106,
-      "train_speed(iter/s)": 0.033633
+      "train_speed(iter/s)": 0.002665
     },
     {
-      "clip_ratio/high_max": 0.0023960003745742142,
-      "clip_ratio/high_mean": 0.0023960003745742142,
-      "clip_ratio/low_mean": 0.002585217938758433,
-      "clip_ratio/low_min": 0.002585217938758433,
-      "clip_ratio/region_mean": 0.004981218371540308,
-      "completions/clipped_ratio": 0.0625,
-      "completions/max_length": 51.0,
-      "completions/mean_length": 13.6328125,
-      "completions/min_length": 2.0,
-      "epoch": 1.188365650969529,
-      "grad_norm": 2.0082146476980807,
-      "kl": 1.3074679019264295,
-      "learning_rate": 7.72364650742707e-05,
-      "loss": 0.011914699338376522,
-      "memory(GiB)": 79.43,
-      "reward": 0.328125,
-      "reward_std": 0.18394747376441956,
-      "rewards/Response_no_think_reward_1/mean": 0.328125,
-      "rewards/Response_no_think_reward_1/std": 0.7110973596572876,
+      "clip_ratio/high_max": 0.0030637255404144526,
+      "clip_ratio/high_mean": 0.0030637255404144526,
+      "clip_ratio/low_mean": 0.015822806861251593,
+      "clip_ratio/low_min": 0.015822806861251593,
+      "clip_ratio/region_mean": 0.018886532401666045,
+      "completions/clipped_ratio": 0.109375,
+      "completions/max_length": 47.0,
+      "completions/mean_length": 6.7109375,
+      "completions/min_length": 3.0,
+      "epoch": 1.2985074626865671,
+      "grad_norm": 5.1535076282650945,
+      "kl": 2.8769574181642383,
+      "learning_rate": 6.19628632667923e-05,
+      "loss": 0.038440439850091934,
+      "memory(GiB)": 79.48,
+      "reward": -0.0625,
+      "reward_std": 0.4184814691543579,
+      "rewards/Response_no_think_reward_1/mean": -0.0625,
+      "rewards/Response_no_think_reward_1/std": 0.8394786715507507,
       "step": 107,
-      "train_speed(iter/s)": 0.027655
-    },
-    {
-      "clip_ratio/high_max": 0.00717889575753361,
-      "clip_ratio/high_mean": 0.00717889575753361,
-      "clip_ratio/low_mean": 0.008150914101861417,
-      "clip_ratio/low_min": 0.008150914101861417,
-      "clip_ratio/region_mean": 0.015329810208640993,
-      "epoch": 1.1994459833795015,
-      "grad_norm": 0.5916941576021421,
-      "kl": 1.0833495813399168,
-      "learning_rate": 7.54514512859201e-05,
-      "loss": 0.0037035662680864334,
-      "memory(GiB)": 79.43,
+      "train_speed(iter/s)": 0.002655
+    },
+    {
+      "clip_ratio/high_max": 0.011060049437219277,
+      "clip_ratio/high_mean": 0.011060049437219277,
+      "clip_ratio/low_mean": 0.05394765589153394,
+      "clip_ratio/low_min": 0.05394765589153394,
+      "clip_ratio/region_mean": 0.06500770541606471,
+      "epoch": 1.31044776119403,
+      "grad_norm": 38.3461665329109,
+      "kl": 2.019717270624824,
+      "learning_rate": 6.011999580673931e-05,
+      "loss": 0.16980835795402527,
+      "memory(GiB)": 79.48,
       "step": 108,
-      "train_speed(iter/s)": 0.026591
+      "train_speed(iter/s)": 0.002668
     },
     {
-      "clip_ratio/high_max": 0.004251995822414756,
-      "clip_ratio/high_mean": 0.004251995822414756,
-      "clip_ratio/low_mean": 0.008190131688024849,
-      "clip_ratio/low_min": 0.008190131688024849,
-      "clip_ratio/region_mean": 0.012442127510439605,
-      "completions/clipped_ratio": 0.09375,
-      "completions/max_length": 51.0,
-      "completions/mean_length": 12.234375,
-      "completions/min_length": 4.0,
-      "epoch": 1.2105263157894737,
-      "grad_norm": 2.6396182071410137,
-      "kl": 2.141993957106024,
-      "learning_rate": 7.36747230435401e-05,
-      "loss": 0.011151588521897793,
-      "memory(GiB)": 79.43,
-      "reward": -0.2265625,
-      "reward_std": 0.3407740592956543,
-      "rewards/Response_no_think_reward_1/mean": -0.2265625,
-      "rewards/Response_no_think_reward_1/std": 0.6425201892852783,
+      "clip_ratio/high_max": 0.0016025641234591603,
+      "clip_ratio/high_mean": 0.0016025641234591603,
+      "clip_ratio/low_mean": 0.0032051282469183207,
+      "clip_ratio/low_min": 0.0032051282469183207,
+      "clip_ratio/region_mean": 0.004807692486792803,
+      "completions/clipped_ratio": 0.0078125,
+      "completions/max_length": 25.0,
+      "completions/mean_length": 5.703125,
+      "completions/min_length": 3.0,
+      "epoch": 1.3223880597014925,
+      "grad_norm": 1.9241770555383604,
+      "kl": 0.4095292093356875,
+      "learning_rate": 5.829309600183536e-05,
+      "loss": 0.003778851358219981,
+      "memory(GiB)": 79.48,
+      "reward": 0.453125,
+      "reward_std": 0.24464011192321777,
+      "rewards/Response_no_think_reward_1/mean": 0.453125,
+      "rewards/Response_no_think_reward_1/std": 0.8405039310455322,
       "step": 109,
-      "train_speed(iter/s)": 0.022574
-    },
-    {
-      "clip_ratio/high_max": 0.004035328107420355,
-      "clip_ratio/high_mean": 0.004035328107420355,
-      "clip_ratio/low_mean": 0.08847818686626852,
-      "clip_ratio/low_min": 0.08847818686626852,
-      "clip_ratio/region_mean": 0.0925135153811425,
-      "epoch": 1.221606648199446,
-      "grad_norm": 1.4171762109533736,
-      "kl": 3.1860878374427557,
-      "learning_rate": 7.190688002264308e-05,
-      "loss": 0.0037668771110475063,
-      "memory(GiB)": 79.43,
+      "train_speed(iter/s)": 0.00266
+    },
+    {
+      "clip_ratio/high_max": 0.016826923470944166,
+      "clip_ratio/high_mean": 0.016826923470944166,
+      "clip_ratio/low_mean": 0.025240384973585606,
+      "clip_ratio/low_min": 0.025240384973585606,
+      "clip_ratio/region_mean": 0.042067307978868484,
+      "epoch": 1.3343283582089551,
+      "grad_norm": 4.00720293405374,
+      "kl": 0.3944609015534297,
+      "learning_rate": 5.648289532908666e-05,
+      "loss": 0.0013700753916054964,
+      "memory(GiB)": 79.48,
       "step": 110,
-      "train_speed(iter/s)": 0.021907
+      "train_speed(iter/s)": 0.002673
     },
     {
       "clip_ratio/high_max": 0.0,
       "clip_ratio/high_mean": 0.0,
-      "clip_ratio/low_mean": 0.002568919211626053,
-      "clip_ratio/low_min": 0.002568919211626053,
-      "clip_ratio/region_mean": 0.002568919211626053,
-      "completions/clipped_ratio": 0.046875,
-      "completions/max_length": 48.0,
-      "completions/mean_length": 11.7734375,
-      "completions/min_length": 5.0,
-      "epoch": 1.2326869806094183,
-      "grad_norm": 1.02177770841713,
-      "kl": 1.6274185269139707,
-      "learning_rate": 7.014851889983057e-05,
-      "loss": 0.010478168725967407,
-      "memory(GiB)": 79.43,
-      "reward": 0.453125,
-      "reward_std": 0.3291260004043579,
-      "rewards/Response_no_think_reward_1/mean": 0.453125,
-      "rewards/Response_no_think_reward_1/std": 0.685730516910553,
+      "clip_ratio/low_mean": 0.012351020704954863,
+      "clip_ratio/low_min": 0.012351020704954863,
+      "clip_ratio/region_mean": 0.012351020704954863,
+      "completions/clipped_ratio": 0.2265625,
+      "completions/max_length": 23.0,
+      "completions/mean_length": 6.03125,
+      "completions/min_length": 3.0,
+      "epoch": 1.346268656716418,
+      "grad_norm": 5.342919033145562,
+      "kl": 6.366508552979212,
+      "learning_rate": 5.4690118579292015e-05,
+      "loss": 0.08252155035734177,
+      "memory(GiB)": 79.48,
+      "reward": -0.0703125,
+      "reward_std": 0.5281283259391785,
+      "rewards/Response_no_think_reward_1/mean": -0.0703125,
+      "rewards/Response_no_think_reward_1/std": 0.7954951524734497,
       "step": 111,
-      "train_speed(iter/s)": 0.019127
-    },
-    {
-      "clip_ratio/high_max": 0.013096909533487633,
-      "clip_ratio/high_mean": 0.013096909533487633,
-      "clip_ratio/low_mean": 0.016107605304569006,
-      "clip_ratio/low_min": 0.016107605304569006,
-      "clip_ratio/region_mean": 0.029204514692537487,
-      "epoch": 1.2437673130193905,
-      "grad_norm": 0.79999454502468,
-      "kl": 1.5544351362623274,
-      "learning_rate": 6.840023315140475e-05,
-      "loss": 0.0022247314918786287,
-      "memory(GiB)": 79.43,
+      "train_speed(iter/s)": 0.002667
+    },
+    {
+      "clip_ratio/high_max": 0.0056201552506536245,
+      "clip_ratio/high_mean": 0.0056201552506536245,
+      "clip_ratio/low_mean": 0.07672008802182972,
+      "clip_ratio/low_min": 0.07672008802182972,
+      "clip_ratio/region_mean": 0.08234024350531399,
+      "epoch": 1.3582089552238805,
+      "grad_norm": 4.841852348859136,
+      "kl": 7.482738017570227,
+      "learning_rate": 5.291548356684177e-05,
+      "loss": 0.06593235582113266,
+      "memory(GiB)": 79.48,
       "step": 112,
-      "train_speed(iter/s)": 0.018681
+      "train_speed(iter/s)": 0.00268
     },
     {
-      "clip_ratio/high_max": 0.0008340688509633765,
-      "clip_ratio/high_mean": 0.0008340688509633765,
-      "clip_ratio/low_mean": 0.007082634954713285,
-      "clip_ratio/low_min": 0.007082634954713285,
-      "clip_ratio/region_mean": 0.007916703820228577,
-      "completions/clipped_ratio": 0.1171875,
-      "completions/max_length": 51.0,
-      "completions/mean_length": 13.53125,
-      "completions/min_length": 3.0,
-      "epoch": 1.254847645429363,
-      "grad_norm": 1.6246389101003882,
-      "kl": 1.6449745513964444,
-      "learning_rate": 6.666261285306047e-05,
-      "loss": 0.014917208813130856,
-      "memory(GiB)": 79.43,
-      "reward": 0.1953125,
-      "reward_std": 0.2754020392894745,
-      "rewards/Response_no_think_reward_1/mean": 0.1953125,
-      "rewards/Response_no_think_reward_1/std": 0.869958758354187,
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.00446532026398927,
+      "clip_ratio/low_min": 0.00446532026398927,
+      "clip_ratio/region_mean": 0.00446532026398927,
+      "completions/clipped_ratio": 0.046875,
+      "completions/max_length": 10.0,
+      "completions/mean_length": 5.328125,
+      "completions/min_length": 3.0,
+      "epoch": 1.3701492537313433,
+      "grad_norm": 2.4231207405852158,
+      "kl": 2.7402262951763987,
+      "learning_rate": 5.115970084231059e-05,
+      "loss": 0.021689381450414658,
+      "memory(GiB)": 79.48,
+      "reward": 0.4453125,
+      "reward_std": 0.33350807428359985,
+      "rewards/Response_no_think_reward_1/mean": 0.4453125,
+      "rewards/Response_no_think_reward_1/std": 0.7817665934562683,
       "step": 113,
-      "train_speed(iter/s)": 0.016592
-    },
-    {
-      "clip_ratio/high_max": 0.006006928611896001,
-      "clip_ratio/high_mean": 0.006006928611896001,
-      "clip_ratio/low_mean": 0.039706501411274076,
-      "clip_ratio/low_min": 0.039706501411274076,
-      "clip_ratio/region_mean": 0.045713430270552635,
-      "epoch": 1.2659279778393353,
-      "grad_norm": 0.940807595373605,
-      "kl": 1.8106578167062253,
-      "learning_rate": 6.493624448072457e-05,
-      "loss": 0.006363555323332548,
-      "memory(GiB)": 79.43,
+      "train_speed(iter/s)": 0.002673
+    },
+    {
+      "clip_ratio/high_max": 0.02434239676222205,
+      "clip_ratio/high_mean": 0.02434239676222205,
+      "clip_ratio/low_mean": 0.0223974745022133,
+      "clip_ratio/low_min": 0.0223974745022133,
+      "clip_ratio/region_mean": 0.046739870798774064,
+      "epoch": 1.382089552238806,
+      "grad_norm": 1.0551090300064008,
+      "kl": 2.37436256357978,
+      "learning_rate": 4.9423473407958035e-05,
+      "loss": 0.008349667303264141,
+      "memory(GiB)": 79.48,
       "step": 114,
-      "train_speed(iter/s)": 0.016278
+      "train_speed(iter/s)": 0.002685
     },
     {
-      "clip_ratio/high_max": 0.0052188277477398515,
-      "clip_ratio/high_mean": 0.0052188277477398515,
-      "clip_ratio/low_mean": 0.004236701555782929,
-      "clip_ratio/low_min": 0.004236701555782929,
-      "clip_ratio/region_mean": 0.009455529390834272,
-      "completions/clipped_ratio": 0.1796875,
-      "completions/max_length": 51.0,
-      "completions/mean_length": 14.3671875,
-      "completions/min_length": 2.0,
-      "epoch": 1.2770083102493075,
-      "grad_norm": 2.8040352101197437,
-      "kl": 5.506896490347572,
-      "learning_rate": 6.322171071261071e-05,
-      "loss": 0.03229736536741257,
-      "memory(GiB)": 79.43,
-      "reward": 0.296875,
-      "reward_std": 0.43980443477630615,
-      "rewards/Response_no_think_reward_1/mean": 0.296875,
-      "rewards/Response_no_think_reward_1/std": 0.8905397057533264,
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.011147994082421064,
+      "clip_ratio/low_min": 0.011147994082421064,
+      "clip_ratio/region_mean": 0.011147994082421064,
+      "completions/clipped_ratio": 0.15625,
+      "completions/max_length": 11.0,
+      "completions/mean_length": 5.3515625,
+      "completions/min_length": 3.0,
+      "epoch": 1.3940298507462687,
+      "grad_norm": 35.484376517766954,
+      "kl": 24.603704601235222,
+      "learning_rate": 4.7707496436252e-05,
+      "loss": 0.3364083468914032,
+      "memory(GiB)": 79.48,
+      "reward": 0.1328125,
+      "reward_std": 0.28117600083351135,
+      "rewards/Response_no_think_reward_1/mean": 0.1328125,
+      "rewards/Response_no_think_reward_1/std": 0.8264437913894653,
       "step": 115,
-      "train_speed(iter/s)": 0.014636
-    },
-    {
-      "clip_ratio/high_max": 0.01615347486222163,
-      "clip_ratio/high_mean": 0.01615347486222163,
-      "clip_ratio/low_mean": 0.015428576618432999,
-      "clip_ratio/low_min": 0.015428576618432999,
-      "clip_ratio/region_mean": 0.031582050723955035,
-      "epoch": 1.2880886426592797,
-      "grad_norm": 1.8650893219420759,
-      "kl": 4.258469146443531,
-      "learning_rate": 6.151959023255545e-05,
-      "loss": 0.020270783454179764,
-      "memory(GiB)": 79.43,
+      "train_speed(iter/s)": 0.002678
+    },
+    {
+      "clip_ratio/high_max": 0.002314814832061529,
+      "clip_ratio/high_mean": 0.002314814832061529,
+      "clip_ratio/low_mean": 0.0054197743302211165,
+      "clip_ratio/low_min": 0.0054197743302211165,
+      "clip_ratio/region_mean": 0.007734589162282646,
+      "epoch": 1.4059701492537313,
+      "grad_norm": 7.954623939964381,
+      "kl": 12.087648045126116,
+      "learning_rate": 4.601245699152659e-05,
+      "loss": 0.2005119025707245,
+      "memory(GiB)": 79.48,
       "step": 116,
-      "train_speed(iter/s)": 0.014407
+      "train_speed(iter/s)": 0.00269
     },
     {
-      "clip_ratio/high_max": 0.005145640461705625,
-      "clip_ratio/high_mean": 0.005145640461705625,
-      "clip_ratio/low_mean": 0.008849590638419613,
-      "clip_ratio/low_min": 0.008849590638419613,
-      "clip_ratio/region_mean": 0.013995230983709916,
-      "completions/clipped_ratio": 0.1640625,
-      "completions/max_length": 51.0,
-      "completions/mean_length": 16.3125,
-      "completions/min_length": 2.0,
-      "epoch": 1.299168975069252,
-      "grad_norm": 2.4581301952944186,
-      "kl": 3.745300827547908,
-      "learning_rate": 5.983045753470308e-05,
-      "loss": 0.024165078997612,
-      "memory(GiB)": 79.43,
-      "reward": 0.453125,
-      "reward_std": 0.5826787948608398,
-      "rewards/Response_no_think_reward_1/mean": 0.453125,
-      "rewards/Response_no_think_reward_1/std": 0.8405039310455322,
+      "clip_ratio/high_max": 0.0005434782360680401,
+      "clip_ratio/high_mean": 0.0005434782360680401,
+      "clip_ratio/low_mean": 0.004596828715875745,
+      "clip_ratio/low_min": 0.004596828715875745,
+      "clip_ratio/region_mean": 0.005140306951943785,
+      "completions/clipped_ratio": 0.0703125,
+      "completions/max_length": 40.0,
+      "completions/mean_length": 5.8671875,
+      "completions/min_length": 3.0,
+      "epoch": 1.417910447761194,
+      "grad_norm": 3.80570877135741,
+      "kl": 3.121465804113541,
+      "learning_rate": 4.433903375488697e-05,
+      "loss": 0.044249728322029114,
+      "memory(GiB)": 79.48,
+      "reward": 0.3046875,
+      "reward_std": 0.38101160526275635,
+      "rewards/Response_no_think_reward_1/mean": 0.3046875,
+      "rewards/Response_no_think_reward_1/std": 0.8562746047973633,
       "step": 117,
-      "train_speed(iter/s)": 0.013208
-    },
-    {
-      "clip_ratio/high_max": 0.04523819196037948,
-      "clip_ratio/high_mean": 0.04523819196037948,
-      "clip_ratio/low_mean": 0.01510127488290891,
-      "clip_ratio/low_min": 0.01510127488290891,
-      "clip_ratio/region_mean": 0.060339466377627105,
-      "epoch": 1.3102493074792243,
-      "grad_norm": 2.131099568305027,
-      "kl": 2.847630614414811,
-      "learning_rate": 5.8154882729603876e-05,
-      "loss": 0.01335166022181511,
-      "memory(GiB)": 79.43,
+      "train_speed(iter/s)": 0.002684
+    },
+    {
+      "clip_ratio/high_max": 0.014024957199580967,
+      "clip_ratio/high_mean": 0.014024957199580967,
+      "clip_ratio/low_mean": 0.0131244600052014,
+      "clip_ratio/low_min": 0.0131244600052014,
+      "clip_ratio/region_mean": 0.02714941732119769,
+      "epoch": 1.4298507462686567,
+      "grad_norm": 3.857626589553114,
+      "kl": 2.146273965074215,
+      "learning_rate": 4.268789675247029e-05,
+      "loss": 0.03236865624785423,
+      "memory(GiB)": 79.48,
       "step": 118,
-      "train_speed(iter/s)": 0.013037
+      "train_speed(iter/s)": 0.002695
     },
     {
-      "clip_ratio/high_max": 0.002125054510543123,
-      "clip_ratio/high_mean": 0.002125054510543123,
-      "clip_ratio/low_mean": 0.0127813016588334,
-      "clip_ratio/low_min": 0.0127813016588334,
-      "clip_ratio/region_mean": 0.0149063560529612,
-      "completions/clipped_ratio": 0.203125,
-      "completions/max_length": 51.0,
-      "completions/mean_length": 19.59375,
-      "completions/min_length": 3.0,
-      "epoch": 1.3213296398891967,
-      "grad_norm": 2.690385409772627,
-      "kl": 2.0104650848079473,
-      "learning_rate": 5.64934313517927e-05,
-      "loss": 0.01620793715119362,
-      "memory(GiB)": 79.43,
-      "reward": 0.2578125,
-      "reward_std": 0.41504764556884766,
-      "rewards/Response_no_think_reward_1/mean": 0.2578125,
-      "rewards/Response_no_think_reward_1/std": 0.8625734448432922,
+      "clip_ratio/high_max": 0.0012886597542092204,
+      "clip_ratio/high_mean": 0.0012886597542092204,
+      "clip_ratio/low_mean": 0.0051683366764336824,
+      "clip_ratio/low_min": 0.0051683366764336824,
+      "clip_ratio/region_mean": 0.006456996430642903,
+      "completions/clipped_ratio": 0.0703125,
+      "completions/max_length": 22.0,
+      "completions/mean_length": 5.6171875,
+      "completions/min_length": 3.0,
+      "epoch": 1.4417910447761195,
+      "grad_norm": 3.2126971808505513,
+      "kl": 1.1720350683217475,
+      "learning_rate": 4.105970708717244e-05,
+      "loss": 0.018675971776247025,
+      "memory(GiB)": 79.48,
+      "reward": 0.5625,
+      "reward_std": 0.22125522792339325,
+      "rewards/Response_no_think_reward_1/mean": 0.5625,
+      "rewards/Response_no_think_reward_1/std": 0.6490453481674194,
       "step": 119,
-      "train_speed(iter/s)": 0.011541
-    },
-    {
-      "clip_ratio/high_max": 0.0026382115320302546,
-      "clip_ratio/high_mean": 0.0026382115320302546,
-      "clip_ratio/low_mean": 0.07074824426672421,
-      "clip_ratio/low_min": 0.07074824426672421,
-      "clip_ratio/region_mean": 0.07338645646814257,
-      "epoch": 1.332409972299169,
-      "grad_norm": 1.3611711034111147,
-      "kl": 2.366683575557545,
-      "learning_rate": 5.484666416891109e-05,
-      "loss": 0.00427972199395299,
-      "memory(GiB)": 79.43,
+      "train_speed(iter/s)": 0.002674
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.020605165394954383,
+      "clip_ratio/low_min": 0.020605165394954383,
+      "clip_ratio/region_mean": 0.020605165394954383,
+      "epoch": 1.4537313432835821,
+      "grad_norm": 1.102830732170957,
+      "kl": 1.583265769013451,
+      "learning_rate": 3.945511667394719e-05,
+      "loss": 0.003417772939428687,
+      "memory(GiB)": 79.48,
       "step": 120,
-      "train_speed(iter/s)": 0.011424
+      "train_speed(iter/s)": 0.002685
     },
     {
-      "clip_ratio/high_max": 0.00034340660204179585,
-      "clip_ratio/high_mean": 0.00034340660204179585,
-      "clip_ratio/low_mean": 0.0057576168910600245,
-      "clip_ratio/low_min": 0.0057576168910600245,
-      "clip_ratio/region_mean": 0.00610102349310182,
-      "completions/clipped_ratio": 0.109375,
-      "completions/max_length": 51.0,
-      "completions/mean_length": 14.1640625,
-      "completions/min_length": 3.0,
-      "epoch": 1.3434903047091413,
-      "grad_norm": 1.7730415769710828,
-      "kl": 2.253111455589533,
-      "learning_rate": 5.321513699243924e-05,
-      "loss": 0.015828199684619904,
-      "memory(GiB)": 79.43,
-      "reward": 0.2578125,
-      "reward_std": 0.2504267692565918,
-      "rewards/Response_no_think_reward_1/mean": 0.2578125,
-      "rewards/Response_no_think_reward_1/std": 0.8533961176872253,
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.007155258092097938,
+      "clip_ratio/low_min": 0.007155258092097938,
+      "clip_ratio/region_mean": 0.007155258092097938,
+      "completions/clipped_ratio": 0.046875,
+      "completions/max_length": 23.0,
+      "completions/mean_length": 5.6640625,
+      "completions/min_length": 3.0,
+      "epoch": 1.4656716417910447,
+      "grad_norm": 2.9093250272424034,
+      "kl": 1.8708921101060696,
+      "learning_rate": 3.787476797878459e-05,
+      "loss": 0.01941162720322609,
+      "memory(GiB)": 79.48,
+      "reward": 0.59375,
+      "reward_std": 0.3198433816432953,
+      "rewards/Response_no_think_reward_1/mean": 0.59375,
+      "rewards/Response_no_think_reward_1/std": 0.692337155342102,
       "step": 121,
-      "train_speed(iter/s)": 0.010746
-    },
-    {
-      "clip_ratio/high_max": 0.002522797236451879,
-      "clip_ratio/high_mean": 0.002522797236451879,
-      "clip_ratio/low_mean": 0.03892370511312038,
-      "clip_ratio/low_min": 0.03892370511312038,
-      "clip_ratio/region_mean": 0.041446502320468426,
-      "epoch": 1.3545706371191135,
-      "grad_norm": 0.9270755733265906,
-      "kl": 2.852388353087008,
-      "learning_rate": 5.159940049010015e-05,
-      "loss": 0.00514911487698555,
-      "memory(GiB)": 79.43,
+      "train_speed(iter/s)": 0.002659
+    },
+    {
+      "clip_ratio/high_max": 0.02566964237485081,
+      "clip_ratio/high_mean": 0.02566964237485081,
+      "clip_ratio/low_mean": 0.01106150820851326,
+      "clip_ratio/low_min": 0.01106150820851326,
+      "clip_ratio/region_mean": 0.03673115174751729,
+      "epoch": 1.4776119402985075,
+      "grad_norm": 1.6206572311785468,
+      "kl": 2.092067622463219,
+      "learning_rate": 3.631929376147207e-05,
+      "loss": 0.00795831624418497,
+      "memory(GiB)": 79.48,
       "step": 122,
-      "train_speed(iter/s)": 0.010652
+      "train_speed(iter/s)": 0.00267
     },
     {
-      "clip_ratio/high_max": 0.0008753835718380287,
-      "clip_ratio/high_mean": 0.0008753835718380287,
-      "clip_ratio/low_mean": 0.0012499999720603228,
-      "clip_ratio/low_min": 0.0012499999720603228,
-      "clip_ratio/region_mean": 0.0021253835438983515,
-      "completions/clipped_ratio": 0.1171875,
-      "completions/max_length": 51.0,
-      "completions/mean_length": 15.8515625,
-      "completions/min_length": 4.0,
-      "epoch": 1.365650969529086,
-      "grad_norm": 24.163527329310554,
-      "kl": 24.496757203305606,
-      "learning_rate": 5.000000000000002e-05,
-      "loss": 0.23566541075706482,
-      "memory(GiB)": 79.43,
-      "reward": 0.3984375,
-      "reward_std": 0.3253360986709595,
-      "rewards/Response_no_think_reward_1/mean": 0.3984375,
-      "rewards/Response_no_think_reward_1/std": 0.6912255883216858,
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.0078125,
+      "completions/max_length": 16.0,
+      "completions/mean_length": 5.140625,
+      "completions/min_length": 3.0,
+      "epoch": 1.48955223880597,
+      "grad_norm": 1.1898236332850007,
+      "kl": 0.5545606576924911,
+      "learning_rate": 3.47893168222425e-05,
+      "loss": 0.005211046896874905,
+      "memory(GiB)": 79.48,
+      "reward": 0.5078125,
+      "reward_std": 0.2012200653553009,
+      "rewards/Response_no_think_reward_1/mean": 0.5078125,
+      "rewards/Response_no_think_reward_1/std": 0.763292670249939,
       "step": 123,
-      "train_speed(iter/s)": 0.010181
-    },
-    {
-      "clip_ratio/high_max": 0.0009191176941385493,
-      "clip_ratio/high_mean": 0.0009191176941385493,
-      "clip_ratio/low_mean": 0.025084137567318976,
-      "clip_ratio/low_min": 0.025084137567318976,
-      "clip_ratio/region_mean": 0.026003255392424762,
-      "epoch": 1.3767313019390581,
-      "grad_norm": 4.61798364138243,
-      "kl": 5.482510100933723,
-      "learning_rate": 4.841747534656763e-05,
-      "loss": 0.03564080968499184,
-      "memory(GiB)": 79.43,
+      "train_speed(iter/s)": 0.002638
+    },
+    {
+      "clip_ratio/high_max": 0.015350877307355404,
+      "clip_ratio/high_mean": 0.015350877307355404,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.015350877307355404,
+      "epoch": 1.5014925373134327,
+      "grad_norm": 0.3409132689994619,
+      "kl": 0.5577427046882804,
+      "learning_rate": 3.328544975240932e-05,
+      "loss": -0.0002708420215640217,
+      "memory(GiB)": 79.48,
       "step": 124,
-      "train_speed(iter/s)": 0.010103
+      "train_speed(iter/s)": 0.002649
     },
     {
-      "clip_ratio/high_max": 0.0004921259824186563,
-      "clip_ratio/high_mean": 0.0004921259824186563,
-      "clip_ratio/low_mean": 0.0009282178361900151,
-      "clip_ratio/low_min": 0.0009282178361900151,
-      "clip_ratio/region_mean": 0.0014203438186086714,
-      "completions/clipped_ratio": 0.140625,
-      "completions/max_length": 51.0,
-      "completions/mean_length": 16.671875,
-      "completions/min_length": 3.0,
-      "epoch": 1.3878116343490305,
-      "grad_norm": 1.7230796893481761,
-      "kl": 1.7308420957997441,
-      "learning_rate": 4.685236065835443e-05,
-      "loss": 0.01634293608367443,
-      "memory(GiB)": 79.43,
-      "reward": 0.171875,
-      "reward_std": 0.13258251547813416,
-      "rewards/Response_no_think_reward_1/mean": 0.171875,
-      "rewards/Response_no_think_reward_1/std": 0.8971465826034546,
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.00601851858664304,
+      "clip_ratio/low_min": 0.00601851858664304,
+      "clip_ratio/region_mean": 0.00601851858664304,
+      "completions/clipped_ratio": 0.0390625,
+      "completions/max_length": 30.0,
+      "completions/mean_length": 5.8671875,
+      "completions/min_length": 3.0,
+      "epoch": 1.5134328358208955,
+      "grad_norm": 5.906541393453947,
+      "kl": 6.567164011168643,
+      "learning_rate": 3.180829468908986e-05,
+      "loss": 0.07308901846408844,
+      "memory(GiB)": 79.48,
+      "reward": 0.3671875,
+      "reward_std": 0.1412346363067627,
+      "rewards/Response_no_think_reward_1/mean": 0.3671875,
+      "rewards/Response_no_think_reward_1/std": 0.8406137228012085,
       "step": 125,
-      "train_speed(iter/s)": 0.009672
-    },
-    {
-      "clip_ratio/high_max": 0.0003094059356953949,
-      "clip_ratio/high_mean": 0.0003094059356953949,
-      "clip_ratio/low_mean": 0.02542525064200163,
-      "clip_ratio/low_min": 0.02542525064200163,
-      "clip_ratio/region_mean": 0.025734656490385532,
-      "epoch": 1.3988919667590027,
-      "grad_norm": 0.6685767852950313,
-      "kl": 2.058195663616061,
-      "learning_rate": 4.530518418775733e-05,
-      "loss": 0.008037411607801914,
-      "memory(GiB)": 79.43,
+      "train_speed(iter/s)": 0.002642
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.010185185354202986,
+      "clip_ratio/low_min": 0.010185185354202986,
+      "clip_ratio/region_mean": 0.010185185354202986,
+      "epoch": 1.5253731343283583,
+      "grad_norm": 1.637348051569756,
+      "kl": 3.2303040275146486,
+      "learning_rate": 3.035844307411384e-05,
+      "loss": 0.04529271274805069,
+      "memory(GiB)": 79.48,
       "step": 126,
-      "train_speed(iter/s)": 0.009605
+      "train_speed(iter/s)": 0.002652
     },
     {
-      "clip_ratio/high_max": 0.0035992932971566916,
-      "clip_ratio/high_mean": 0.0035992932971566916,
-      "clip_ratio/low_mean": 0.0026174120721407235,
-      "clip_ratio/low_min": 0.0026174120721407235,
-      "clip_ratio/region_mean": 0.006216705543920398,
-      "completions/clipped_ratio": 0.078125,
-      "completions/max_length": 51.0,
-      "completions/mean_length": 13.1015625,
-      "completions/min_length": 3.0,
-      "epoch": 1.4099722991689752,
-      "grad_norm": 0.898899557372063,
-      "kl": 1.8244512832025066,
-      "learning_rate": 4.3776468132724604e-05,
-      "loss": 0.013885595835745335,
-      "memory(GiB)": 79.43,
-      "reward": 0.703125,
-      "reward_std": 0.1841355264186859,
-      "rewards/Response_no_think_reward_1/mean": 0.703125,
-      "rewards/Response_no_think_reward_1/std": 0.552152156829834,
+      "clip_ratio/high_max": 0.00046296295477077365,
+      "clip_ratio/high_mean": 0.00046296295477077365,
+      "clip_ratio/low_mean": 0.004759002127684653,
+      "clip_ratio/low_min": 0.004759002127684653,
+      "clip_ratio/region_mean": 0.0052219650824554265,
+      "completions/clipped_ratio": 0.046875,
+      "completions/max_length": 29.0,
+      "completions/mean_length": 6.4453125,
+      "completions/min_length": 3.0,
+      "epoch": 1.537313432835821,
+      "grad_norm": 1.5814533572332585,
+      "kl": 1.9979073457579943,
+      "learning_rate": 2.8936475417214794e-05,
+      "loss": 0.022465700283646584,
+      "memory(GiB)": 79.48,
+      "reward": 0.3359375,
+      "reward_std": 0.17123225331306458,
+      "rewards/Response_no_think_reward_1/mean": 0.3359375,
+      "rewards/Response_no_think_reward_1/std": 0.7453514337539673,
       "step": 127,
-      "train_speed(iter/s)": 0.009212
-    },
-    {
-      "clip_ratio/high_max": 0.00787301326636225,
-      "clip_ratio/high_mean": 0.00787301326636225,
-      "clip_ratio/low_mean": 0.003833794384263456,
-      "clip_ratio/low_min": 0.003833794384263456,
-      "clip_ratio/region_mean": 0.011706807999871671,
-      "epoch": 1.4210526315789473,
-      "grad_norm": 0.6061431338619399,
-      "kl": 1.8278243900567759,
-      "learning_rate": 4.2266728460505375e-05,
-      "loss": 0.009479128755629063,
-      "memory(GiB)": 79.43,
+      "train_speed(iter/s)": 0.00264
+    },
+    {
+      "clip_ratio/high_max": 0.0014710274408571422,
+      "clip_ratio/high_mean": 0.0014710274408571422,
+      "clip_ratio/low_mean": 0.022759197046980262,
+      "clip_ratio/low_min": 0.022759197046980262,
+      "clip_ratio/region_mean": 0.024230224487837404,
+      "epoch": 1.5492537313432835,
+      "grad_norm": 0.9057768060895434,
+      "kl": 1.9242111706989817,
+      "learning_rate": 2.7542961063598104e-05,
+      "loss": 0.013444737531244755,
+      "memory(GiB)": 79.48,
       "step": 128,
-      "train_speed(iter/s)": 0.009157
+      "train_speed(iter/s)": 0.002651
     },
     {
-      "clip_ratio/high_max": 0.0016208597226068377,
-      "clip_ratio/high_mean": 0.0016208597226068377,
-      "clip_ratio/low_mean": 0.004291799967177212,
-      "clip_ratio/low_min": 0.004291799967177212,
-      "clip_ratio/region_mean": 0.00591265968978405,
-      "completions/clipped_ratio": 0.2421875,
-      "completions/max_length": 51.0,
-      "completions/mean_length": 13.8359375,
-      "completions/min_length": 3.0,
-      "epoch": 1.4321329639889195,
-      "grad_norm": 4.256081630918958,
-      "kl": 8.371784689603373,
-      "learning_rate": 4.077647473350201e-05,
-      "loss": 0.07109890133142471,
-      "memory(GiB)": 79.43,
-      "reward": 0.1640625,
-      "reward_std": 0.12844271957874298,
-      "rewards/Response_no_think_reward_1/mean": 0.1640625,
-      "rewards/Response_no_think_reward_1/std": 0.9029901623725891,
+      "clip_ratio/high_max": 0.0026041667442768812,
+      "clip_ratio/high_mean": 0.0026041667442768812,
+      "clip_ratio/low_mean": 0.010856111068278551,
+      "clip_ratio/low_min": 0.010856111068278551,
+      "clip_ratio/region_mean": 0.013460277812555432,
+      "completions/clipped_ratio": 0.03125,
+      "completions/max_length": 28.0,
+      "completions/mean_length": 5.6875,
+      "completions/min_length": 3.0,
+      "epoch": 1.5611940298507463,
+      "grad_norm": 2.279027191303148,
+      "kl": 1.264391661035006,
+      "learning_rate": 2.617845796597954e-05,
+      "loss": 0.005756520200520754,
+      "memory(GiB)": 79.48,
+      "reward": 0.3671875,
+      "reward_std": 0.15756267309188843,
+      "rewards/Response_no_think_reward_1/mean": 0.3671875,
+      "rewards/Response_no_think_reward_1/std": 0.7410472631454468,
       "step": 129,
-      "train_speed(iter/s)": 0.008828
-    },
-    {
-      "clip_ratio/high_max": 0.00940803368575871,
-      "clip_ratio/high_mean": 0.00940803368575871,
-      "clip_ratio/low_mean": 0.006534474319778383,
-      "clip_ratio/low_min": 0.006534474319778383,
-      "clip_ratio/region_mean": 0.01594250788912177,
-      "epoch": 1.443213296398892,
-      "grad_norm": 2.0975168570419678,
-      "kl": 6.735093111405149,
-      "learning_rate": 3.9306209937284346e-05,
-      "loss": 0.05932926759123802,
-      "memory(GiB)": 79.43,
+      "train_speed(iter/s)": 0.002635
+    },
+    {
+      "clip_ratio/high_max": 0.0026041667442768812,
+      "clip_ratio/high_mean": 0.0026041667442768812,
+      "clip_ratio/low_mean": 0.06580501515418291,
+      "clip_ratio/low_min": 0.06580501515418291,
+      "clip_ratio/region_mean": 0.06840918306261301,
+      "epoch": 1.573134328358209,
+      "grad_norm": 1.198012219937442,
+      "kl": 1.689065290265944,
+      "learning_rate": 2.484351246118507e-05,
+      "loss": -0.000346488319337368,
+      "memory(GiB)": 79.48,
       "step": 130,
-      "train_speed(iter/s)": 0.008781
+      "train_speed(iter/s)": 0.002646
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0059523810632526875,
+      "clip_ratio/low_min": 0.0059523810632526875,
+      "clip_ratio/region_mean": 0.0059523810632526875,
+      "completions/clipped_ratio": 0.0234375,
+      "completions/max_length": 25.0,
+      "completions/mean_length": 5.15625,
+      "completions/min_length": 3.0,
+      "epoch": 1.5850746268656717,
+      "grad_norm": 2.595184247375665,
+      "kl": 0.8083602132865053,
+      "learning_rate": 2.353865905140187e-05,
+      "loss": 0.009201270528137684,
+      "memory(GiB)": 79.48,
+      "reward": 0.4296875,
+      "reward_std": 0.13553348183631897,
+      "rewards/Response_no_think_reward_1/mean": 0.4296875,
+      "rewards/Response_no_think_reward_1/std": 0.7496308088302612,
+      "step": 131,
+      "train_speed(iter/s)": 0.002635
+    },
+    {
+      "clip_ratio/high_max": 0.0019841270986944437,
+      "clip_ratio/high_mean": 0.0019841270986944437,
+      "clip_ratio/low_mean": 0.01378506887704134,
+      "clip_ratio/low_min": 0.01378506887704134,
+      "clip_ratio/region_mean": 0.01576919574290514,
+      "epoch": 1.5970149253731343,
+      "grad_norm": 1.0694122716373746,
+      "kl": 0.8849439112927939,
+      "learning_rate": 2.226442019016739e-05,
+      "loss": 0.0032178200781345367,
+      "memory(GiB)": 79.48,
+      "step": 132,
+      "train_speed(iter/s)": 0.002646
+    },
+    {
+      "clip_ratio/high_max": 0.0014880952658131719,
+      "clip_ratio/high_mean": 0.0014880952658131719,
+      "clip_ratio/low_mean": 0.004777232185006142,
+      "clip_ratio/low_min": 0.004777232185006142,
+      "clip_ratio/region_mean": 0.0062653274508193135,
+      "completions/clipped_ratio": 0.0703125,
+      "completions/max_length": 51.0,
+      "completions/mean_length": 7.703125,
+      "completions/min_length": 3.0,
+      "epoch": 1.608955223880597,
+      "grad_norm": 8.857193213468985,
+      "kl": 6.557242290626164,
+      "learning_rate": 2.1021306073183167e-05,
+      "loss": 0.09046173840761185,
+      "memory(GiB)": 79.48,
+      "reward": 0.5625,
+      "reward_std": 0.3117782771587372,
+      "rewards/Response_no_think_reward_1/mean": 0.5625,
+      "rewards/Response_no_think_reward_1/std": 0.6728714108467102,
+      "step": 133,
+      "train_speed(iter/s)": 0.002636
+    },
+    {
+      "clip_ratio/high_max": 0.0043966451194137335,
+      "clip_ratio/high_mean": 0.0043966451194137335,
+      "clip_ratio/low_mean": 0.0072958533419296145,
+      "clip_ratio/low_min": 0.0072958533419296145,
+      "clip_ratio/region_mean": 0.01169249857775867,
+      "epoch": 1.6208955223880597,
+      "grad_norm": 5.293380513533871,
+      "kl": 4.1661525671806885,
+      "learning_rate": 1.9809814434036e-05,
+      "loss": 0.05591786280274391,
+      "memory(GiB)": 79.48,
+      "step": 134,
+      "train_speed(iter/s)": 0.002646
+    },
+    {
+      "clip_ratio/high_max": 0.0006793478387407959,
+      "clip_ratio/high_mean": 0.0006793478387407959,
+      "clip_ratio/low_mean": 0.008976120705483481,
+      "clip_ratio/low_min": 0.008976120705483481,
+      "clip_ratio/region_mean": 0.009655468544224277,
+      "completions/clipped_ratio": 0.0390625,
+      "completions/max_length": 29.0,
+      "completions/mean_length": 6.125,
+      "completions/min_length": 3.0,
+      "epoch": 1.6328358208955223,
+      "grad_norm": 2.599750460948482,
+      "kl": 2.9140283690503566,
+      "learning_rate": 1.863043034490938e-05,
+      "loss": 0.026728810742497444,
+      "memory(GiB)": 79.48,
+      "reward": 0.3828125,
+      "reward_std": 0.2688143849372864,
+      "rewards/Response_no_think_reward_1/mean": 0.3828125,
+      "rewards/Response_no_think_reward_1/std": 0.7542122602462769,
+      "step": 135,
+      "train_speed(iter/s)": 0.002641
+    },
+    {
+      "clip_ratio/high_max": 0.0006793478387407959,
+      "clip_ratio/high_mean": 0.0006793478387407959,
+      "clip_ratio/low_mean": 0.01892346324166283,
+      "clip_ratio/low_min": 0.01892346324166283,
+      "clip_ratio/region_mean": 0.019602811022195965,
+      "epoch": 1.644776119402985,
+      "grad_norm": 1.7776881222618184,
+      "kl": 2.936662324536883,
+      "learning_rate": 1.748362602236403e-05,
+      "loss": 0.021627021953463554,
+      "memory(GiB)": 79.48,
+      "step": 136,
+      "train_speed(iter/s)": 0.002651
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0003906250058207661,
+      "clip_ratio/low_min": 0.0003906250058207661,
+      "clip_ratio/region_mean": 0.0003906250058207661,
+      "completions/clipped_ratio": 0.0625,
+      "completions/max_length": 24.0,
+      "completions/mean_length": 6.6953125,
+      "completions/min_length": 3.0,
+      "epoch": 1.6567164179104479,
+      "grad_norm": 2.8008543153134777,
+      "kl": 1.5049916390344151,
+      "learning_rate": 1.63698606382661e-05,
+      "loss": 0.022346211597323418,
+      "memory(GiB)": 79.48,
+      "reward": 0.3828125,
+      "reward_std": 0.15467959642410278,
+      "rewards/Response_no_think_reward_1/mean": 0.3828125,
+      "rewards/Response_no_think_reward_1/std": 0.7436988949775696,
+      "step": 137,
+      "train_speed(iter/s)": 0.002647
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.004379986581625417,
+      "clip_ratio/low_min": 0.004379986581625417,
+      "clip_ratio/region_mean": 0.004379986581625417,
+      "epoch": 1.6686567164179105,
+      "grad_norm": 1.5764765403480743,
+      "kl": 1.4290150789947802,
+      "learning_rate": 1.528958013593801e-05,
+      "loss": 0.0181417278945446,
+      "memory(GiB)": 79.48,
+      "step": 138,
+      "train_speed(iter/s)": 0.002657
+    },
+    {
+      "clip_ratio/high_max": 0.0009469697251915932,
+      "clip_ratio/high_mean": 0.0009469697251915932,
+      "clip_ratio/low_mean": 0.009301686193794012,
+      "clip_ratio/low_min": 0.009301686193794012,
+      "clip_ratio/region_mean": 0.010248655918985605,
+      "completions/clipped_ratio": 0.0625,
+      "completions/max_length": 25.0,
+      "completions/mean_length": 5.15625,
+      "completions/min_length": 3.0,
+      "epoch": 1.680597014925373,
+      "grad_norm": 2.987112951446729,
+      "kl": 2.2646760795032606,
+      "learning_rate": 1.4243217051606283e-05,
+      "loss": 0.022022899240255356,
+      "memory(GiB)": 79.48,
+      "reward": 0.5625,
+      "reward_std": 0.3713865876197815,
+      "rewards/Response_no_think_reward_1/mean": 0.5625,
+      "rewards/Response_no_think_reward_1/std": 0.6958821415901184,
+      "step": 139,
+      "train_speed(iter/s)": 0.002652
+    },
+    {
+      "clip_ratio/high_max": 0.0010245901066809893,
+      "clip_ratio/high_mean": 0.0010245901066809893,
+      "clip_ratio/low_mean": 0.008475731243379414,
+      "clip_ratio/low_min": 0.008475731243379414,
+      "clip_ratio/region_mean": 0.009500321350060403,
+      "epoch": 1.6925373134328359,
+      "grad_norm": 2.122472644745234,
+      "kl": 2.2440224051242694,
+      "learning_rate": 1.3231190341217081e-05,
+      "loss": 0.017425674945116043,
+      "memory(GiB)": 79.48,
+      "step": 140,
+      "train_speed(iter/s)": 0.002662
     }
   ],
   "logging_steps": 1,
-  "max_steps": 180,
+  "max_steps": 166,
   "num_input_tokens_seen": 0,
   "num_train_epochs": 2,
   "save_steps": 5,