{
  "best_global_step": 24,
  "best_metric": 0.04339282959699631,
  "best_model_checkpoint": "/mnt/nvme5n1p1/trained_grpo_distill_14b_rl_70_s3/v3-20250330-200345/checkpoint-24",
  "epoch": 6.842105263157895,
  "eval_steps": 6,
  "global_step": 28,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": false,
  "log_history": [
    {
      "clip_ratio": 0.0,
      "completion_length": 10352.974609375,
      "epoch": 0.21052631578947367,
      "grad_norm": 0.13259537518024445,
      "kl": 0.0,
      "learning_rate": 1.6666666666666667e-05,
      "loss": -0.11016345024108887,
      "memory(GiB)": 186.69,
      "response_clip_ratio": 0.11328125,
      "reward": -0.002658387296833098,
      "reward_std": 0.06134121119976044,
      "rewards/CosineReward": -0.0026579967816360295,
      "rewards/RepetitionPenalty": -3.8975886695880035e-07,
      "step": 1,
      "train_speed(iter/s)": 0.000241
    },
    {
      "clip_ratio": 0.0,
      "epoch": 0.42105263157894735,
      "grad_norm": 0.1320001482963562,
      "kl": 0.0,
      "learning_rate": 3.3333333333333335e-05,
      "loss": -0.11016345024108887,
      "memory(GiB)": 186.69,
      "step": 2,
      "train_speed(iter/s)": 0.000466
    },
    {
      "clip_ratio": 1.3441811461234465e-05,
      "completion_length": 10439.369140625,
      "epoch": 0.631578947368421,
      "grad_norm": 0.08990391343832016,
      "kl": 9.50181856751442e-07,
      "learning_rate": 5e-05,
      "loss": -0.06604708731174469,
      "memory(GiB)": 186.69,
      "response_clip_ratio": 0.13671875,
      "reward": 0.0006296975770965219,
      "reward_std": 0.07172460854053497,
      "rewards/CosineReward": 0.0006298604130279273,
      "rewards/RepetitionPenalty": -1.6200439745261974e-07,
      "step": 3,
      "train_speed(iter/s)": 0.00035
    },
    {
      "clip_ratio": 1.70210253145342e-05,
      "epoch": 0.8421052631578947,
      "grad_norm": 0.0967094898223877,
      "kl": 1.1101365089416504e-05,
      "learning_rate": 6.666666666666667e-05,
      "loss": -0.06727766245603561,
      "memory(GiB)": 186.69,
      "step": 4,
      "train_speed(iter/s)": 0.000458
    },
    {
      "clip_ratio": 1.675608473306056e-05,
      "completion_length": 10092.408203125,
      "epoch": 1.2105263157894737,
      "grad_norm": 0.142837256193161,
      "kl": 0.00017762184143066406,
      "learning_rate": 8.333333333333334e-05,
      "loss": -0.09315311908721924,
      "memory(GiB)": 186.69,
      "response_clip_ratio": 0.119140625,
      "reward": -0.005135859013535082,
      "reward_std": 0.07994875870645046,
      "rewards/CosineReward": -0.005134060338605195,
      "rewards/RepetitionPenalty": -1.7973881654143042e-06,
      "step": 5,
      "train_speed(iter/s)": 0.000387
    },
    {
      "epoch": 1.4210526315789473,
      "grad_norm": 0.18263348937034607,
      "learning_rate": 0.0001,
      "loss": -0.1041698157787323,
      "memory(GiB)": 186.69,
      "step": 6,
      "train_speed(iter/s)": 0.000459
    },
    {
      "epoch": 1.4210526315789473,
      "eval_clip_ratio": 4.069424539920874e-05,
      "eval_completion_length": 12289.0,
      "eval_kl": 0.04833984375,
      "eval_loss": -0.5377416610717773,
      "eval_response_clip_ratio": 1.0,
      "eval_reward": 0.012996690347790718,
      "eval_reward_std": 0.08769983053207397,
      "eval_rewards/CosineReward": 0.012996694073081017,
      "eval_rewards/RepetitionPenalty": 0.0,
      "eval_runtime": 1030.1122,
      "eval_samples_per_second": 0.001,
      "eval_steps_per_second": 0.001,
      "step": 6
    },
    {
      "clip_ratio": 0.0005237623976199757,
      "completion_length": 10448.94921875,
      "epoch": 1.631578947368421,
      "grad_norm": 0.1291271299123764,
      "kl": 0.017406463623046875,
      "learning_rate": 9.991540791356342e-05,
      "loss": -0.051375165581703186,
      "memory(GiB)": 186.69,
      "response_clip_ratio": 0.1484375,
      "reward": 0.004909618757665157,
      "reward_std": 0.08167182095348835,
      "rewards/CosineReward": 0.004909833543933928,
      "rewards/RepetitionPenalty": -2.1478646772266075e-07,
      "step": 7,
      "train_speed(iter/s)": 0.000382
    },
    {
      "clip_ratio": 0.1706484742462635,
      "epoch": 1.8421052631578947,
      "grad_norm": 0.26641014218330383,
      "kl": 0.089599609375,
      "learning_rate": 9.966191788709716e-05,
      "loss": -0.05105742812156677,
      "memory(GiB)": 186.69,
      "step": 8,
      "train_speed(iter/s)": 0.000433
    },
    {
      "clip_ratio": 9.482144946559856e-06,
      "completion_length": 10432.384765625,
      "epoch": 2.2105263157894735,
      "grad_norm": 0.10375155508518219,
      "kl": 0.0963134765625,
      "learning_rate": 9.924038765061042e-05,
      "loss": -0.05842069163918495,
      "memory(GiB)": 186.69,
      "response_clip_ratio": 0.255859375,
      "reward": 0.03643610421568155,
      "reward_std": 0.11898956261575222,
      "rewards/CosineReward": 0.03643618477508426,
      "rewards/RepetitionPenalty": -7.898860587829404e-08,
      "step": 9,
      "train_speed(iter/s)": 0.000396
    },
    {
      "clip_ratio": 0.0036088433116674423,
      "epoch": 2.4210526315789473,
      "grad_norm": 0.09477333724498749,
      "kl": 0.1185302734375,
      "learning_rate": 9.865224352899119e-05,
      "loss": -0.06491819024085999,
      "memory(GiB)": 186.69,
      "step": 10,
      "train_speed(iter/s)": 0.000436
    },
    {
      "clip_ratio": 1.2955343891007942e-05,
      "completion_length": 10559.296875,
      "epoch": 2.6315789473684212,
      "grad_norm": 0.06739140301942825,
      "kl": 0.1275634765625,
      "learning_rate": 9.789947561577445e-05,
      "loss": -0.04600231721997261,
      "memory(GiB)": 186.69,
      "response_clip_ratio": 0.361328125,
      "reward": 0.023204635945148766,
      "reward_std": 0.10593634657561779,
      "rewards/CosineReward": 0.02320496749598533,
      "rewards/RepetitionPenalty": -3.3051759373847744e-07,
      "step": 11,
      "train_speed(iter/s)": 0.000405
    },
    {
      "epoch": 2.8421052631578947,
      "grad_norm": 0.05781339108943939,
      "learning_rate": 9.698463103929542e-05,
      "loss": -0.05069056898355484,
      "memory(GiB)": 186.69,
      "step": 12,
      "train_speed(iter/s)": 0.000439
    },
    {
      "epoch": 2.8421052631578947,
      "eval_clip_ratio": 4.392032860778272e-05,
      "eval_completion_length": 12289.0,
      "eval_kl": 0.2275390625,
      "eval_loss": 0.17524278163909912,
      "eval_response_clip_ratio": 1.0,
      "eval_reward": 0.03234308212995529,
      "eval_reward_std": 0.10685288906097412,
      "eval_rewards/CosineReward": 0.03234308212995529,
      "eval_rewards/RepetitionPenalty": 0.0,
      "eval_runtime": 1025.9053,
      "eval_samples_per_second": 0.001,
      "eval_steps_per_second": 0.001,
      "step": 12
    },
    {
      "clip_ratio": 0.0007908324183745208,
      "completion_length": 10652.939453125,
      "epoch": 3.2105263157894735,
      "grad_norm": 0.01199417095631361,
      "kl": 0.151123046875,
      "learning_rate": 9.591080534401371e-05,
      "loss": -0.02191038429737091,
      "memory(GiB)": 186.69,
      "response_clip_ratio": 0.419921875,
      "reward": 0.035983758978545666,
      "reward_std": 0.11553369648754597,
      "rewards/CosineReward": 0.03598417737521231,
      "rewards/RepetitionPenalty": -4.176556771540163e-07,
      "step": 13,
      "train_speed(iter/s)": 0.000399
    },
    {
      "clip_ratio": 0.0004821276670554653,
      "epoch": 3.4210526315789473,
      "grad_norm": 0.01075426209717989,
      "kl": 0.169189453125,
      "learning_rate": 9.468163201617062e-05,
      "loss": -0.022672578692436218,
      "memory(GiB)": 186.69,
      "step": 14,
      "train_speed(iter/s)": 0.000427
    },
    {
      "clip_ratio": 1.9617403040683712e-05,
      "completion_length": 10482.146484375,
      "epoch": 3.6315789473684212,
      "grad_norm": 0.01779361069202423,
      "kl": 0.166748046875,
      "learning_rate": 9.330127018922194e-05,
      "loss": -0.059799157083034515,
      "memory(GiB)": 186.69,
      "response_clip_ratio": 0.4765625,
      "reward": 0.03584331553429365,
      "reward_std": 0.11829411797225475,
      "rewards/CosineReward": 0.03584346390562132,
      "rewards/RepetitionPenalty": -1.4977952389472193e-07,
      "step": 15,
      "train_speed(iter/s)": 0.000406
    },
    {
      "clip_ratio": 0.00011349086707923561,
      "epoch": 3.8421052631578947,
      "grad_norm": 0.013216385617852211,
      "kl": 0.16748046875,
      "learning_rate": 9.177439057064683e-05,
      "loss": -0.06071458384394646,
      "memory(GiB)": 186.69,
      "step": 16,
      "train_speed(iter/s)": 0.000431
    },
    {
      "clip_ratio": 2.4864069928298704e-05,
      "completion_length": 10822.3515625,
      "epoch": 4.2105263157894735,
      "grad_norm": 0.008352754637598991,
      "kl": 0.1787109375,
      "learning_rate": 9.01061596377522e-05,
      "loss": -0.04504441097378731,
      "memory(GiB)": 186.69,
      "response_clip_ratio": 0.5625,
      "reward": 0.027318883687257767,
      "reward_std": 0.10441224090754986,
      "rewards/CosineReward": 0.027319116634316742,
      "rewards/RepetitionPenalty": -2.338138500590503e-07,
      "step": 17,
      "train_speed(iter/s)": 0.00041
    },
    {
      "epoch": 4.421052631578947,
      "grad_norm": 0.005998397711664438,
      "learning_rate": 8.83022221559489e-05,
      "loss": -0.045487549155950546,
      "memory(GiB)": 186.69,
      "step": 18,
      "train_speed(iter/s)": 0.000432
    },
    {
      "epoch": 4.421052631578947,
      "eval_clip_ratio": 2.286707422172185e-05,
      "eval_completion_length": 12289.0,
      "eval_kl": 0.18359375,
      "eval_loss": -0.38219889998435974,
      "eval_response_clip_ratio": 1.0,
      "eval_reward": 0.03729328140616417,
      "eval_reward_std": 0.10691346973180771,
      "eval_rewards/CosineReward": 0.03729327768087387,
      "eval_rewards/RepetitionPenalty": 0.0,
      "eval_runtime": 1041.2277,
      "eval_samples_per_second": 0.001,
      "eval_steps_per_second": 0.001,
      "step": 18
    },
    {
      "clip_ratio": 6.176384295031312e-05,
      "completion_length": 10454.50390625,
      "epoch": 4.631578947368421,
      "grad_norm": 0.007075619418174028,
      "kl": 0.1820068359375,
      "learning_rate": 8.636868207865244e-05,
      "loss": -0.03466903418302536,
      "memory(GiB)": 186.69,
      "response_clip_ratio": 0.466796875,
      "reward": 0.04069916973821819,
      "reward_std": 0.11991005763411522,
      "rewards/CosineReward": 0.04070046404376626,
      "rewards/RepetitionPenalty": -1.294118249006715e-06,
      "step": 19,
      "train_speed(iter/s)": 0.000404
    },
    {
      "clip_ratio": 6.06911453360226e-05,
      "epoch": 4.842105263157895,
      "grad_norm": 0.005896567367017269,
      "kl": 0.19287109375,
      "learning_rate": 8.43120818934367e-05,
      "loss": -0.03502114117145538,
      "memory(GiB)": 186.69,
      "step": 20,
      "train_speed(iter/s)": 0.000424
    },
    {
      "clip_ratio": 3.8725801914551994e-05,
      "completion_length": 10645.056640625,
      "epoch": 5.2105263157894735,
      "grad_norm": 0.004154536407440901,
      "kl": 0.17626953125,
      "learning_rate": 8.213938048432697e-05,
      "loss": -0.008662773296236992,
      "memory(GiB)": 186.69,
      "response_clip_ratio": 0.5625,
      "reward": 0.04996980866417289,
      "reward_std": 0.13849420100450516,
      "rewards/CosineReward": 0.049969930201768875,
      "rewards/RepetitionPenalty": -1.1864573679076784e-07,
      "step": 21,
      "train_speed(iter/s)": 0.000407
    },
    {
      "clip_ratio": 5.869188044016482e-05,
      "epoch": 5.421052631578947,
      "grad_norm": 0.004300669766962528,
      "kl": 0.178955078125,
      "learning_rate": 7.985792958513931e-05,
      "loss": -0.008743642829358578,
      "memory(GiB)": 186.69,
      "step": 22,
      "train_speed(iter/s)": 0.000426
    },
    {
      "clip_ratio": 4.6346245653694496e-05,
      "completion_length": 10538.072265625,
      "epoch": 5.631578947368421,
      "grad_norm": 0.01327697653323412,
      "kl": 0.1796875,
      "learning_rate": 7.74754489035403e-05,
      "loss": -0.03423420712351799,
      "memory(GiB)": 186.69,
      "response_clip_ratio": 0.583984375,
      "reward": 0.034468831261619925,
      "reward_std": 0.11841745302081108,
      "rewards/CosineReward": 0.03447544714435935,
      "rewards/RepetitionPenalty": -6.612649428916484e-06,
      "step": 23,
      "train_speed(iter/s)": 0.00041
    },
    {
      "epoch": 5.842105263157895,
      "grad_norm": 0.014131724834442139,
      "learning_rate": 7.500000000000001e-05,
      "loss": -0.03426633030176163,
      "memory(GiB)": 186.69,
      "step": 24,
      "train_speed(iter/s)": 0.000427
    },
    {
      "epoch": 5.842105263157895,
      "eval_clip_ratio": 4.0687620639801025e-05,
      "eval_completion_length": 12289.0,
      "eval_kl": 0.1982421875,
      "eval_loss": 0.3612469434738159,
      "eval_response_clip_ratio": 1.0,
      "eval_reward": 0.04339282959699631,
      "eval_reward_std": 0.10456253588199615,
      "eval_rewards/CosineReward": 0.04339282959699631,
      "eval_rewards/RepetitionPenalty": 0.0,
      "eval_runtime": 1045.0653,
      "eval_samples_per_second": 0.001,
      "eval_steps_per_second": 0.001,
      "step": 24
    },
    {
      "clip_ratio": 5.05705434079573e-05,
      "completion_length": 10789.259765625,
      "epoch": 6.2105263157894735,
      "grad_norm": 0.0099335303530097,
      "kl": 0.1800537109375,
      "learning_rate": 7.243995901002312e-05,
      "loss": -0.02097315341234207,
      "memory(GiB)": 186.69,
      "response_clip_ratio": 0.6171875,
      "reward": 0.03010205877944827,
      "reward_std": 0.10742511600255966,
      "rewards/CosineReward": 0.030102317687124014,
      "rewards/RepetitionPenalty": -2.580197531187878e-07,
      "step": 25,
      "train_speed(iter/s)": 0.000406
    },
    {
      "clip_ratio": 4.821802576771006e-05,
      "epoch": 6.421052631578947,
      "grad_norm": 0.00989576056599617,
      "kl": 0.18408203125,
      "learning_rate": 6.980398830195785e-05,
      "loss": -0.02103913575410843,
      "memory(GiB)": 186.69,
      "step": 26,
      "train_speed(iter/s)": 0.000421
    },
    {
      "clip_ratio": 5.442534347821493e-05,
      "completion_length": 10197.099609375,
      "epoch": 6.631578947368421,
      "grad_norm": 0.00436774967238307,
      "kl": 0.174560546875,
      "learning_rate": 6.710100716628344e-05,
      "loss": -0.03593946248292923,
      "memory(GiB)": 186.69,
      "response_clip_ratio": 0.513671875,
      "reward": 0.04752760287374258,
      "reward_std": 0.14935147762298584,
      "rewards/CosineReward": 0.04752839542925358,
      "rewards/RepetitionPenalty": -7.915698745364352e-07,
      "step": 27,
      "train_speed(iter/s)": 0.000408
    },
    {
      "clip_ratio": 5.9543880524870474e-05,
      "epoch": 6.842105263157895,
      "grad_norm": 0.005277659278362989,
      "kl": 0.182373046875,
      "learning_rate": 6.434016163555452e-05,
      "loss": -0.03595500811934471,
      "memory(GiB)": 186.69,
      "step": 28,
      "train_speed(iter/s)": 0.000422
    }
  ],
  "logging_steps": 1,
  "max_steps": 60,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 15,
  "save_steps": 2,
  "stateful_callbacks": {
    "TrainerControl": {
      "args": {
        "should_epoch_stop": false,
        "should_evaluate": false,
        "should_log": false,
        "should_save": true,
        "should_training_stop": false
      },
      "attributes": {}
    }
  },
  "total_flos": 0.0,
  "train_batch_size": 4,
  "trial_name": null,
  "trial_params": null
}