diff --git "a/last-checkpoint/trainer_state.json" "b/last-checkpoint/trainer_state.json"
--- "a/last-checkpoint/trainer_state.json"
+++ "b/last-checkpoint/trainer_state.json"
@@ -1,9 +1,9 @@
 {
   "best_metric": null,
   "best_model_checkpoint": null,
-  "epoch": 0.9015680844898091,
+  "epoch": 0.9999678011398396,
   "eval_steps": 5,
-  "global_step": 3500,
+  "global_step": 3882,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
@@ -36424,6 +36424,3972 @@
       "eval_samples_per_second": 2.587,
       "eval_steps_per_second": 0.176,
       "step": 3500
+    },
+    {
+      "epoch": 0.9018256753710918,
+      "grad_norm": 0.14090469611175638,
+      "learning_rate": 2.90695757911017e-06,
+      "loss": 0.3593,
+      "step": 3501
+    },
+    {
+      "epoch": 0.9020832662523747,
+      "grad_norm": 0.14977731555923768,
+      "learning_rate": 2.8918666204450106e-06,
+      "loss": 0.3779,
+      "step": 3502
+    },
+    {
+      "epoch": 0.9023408571336574,
+      "grad_norm": 0.14409493620439406,
+      "learning_rate": 2.8768137682253837e-06,
+      "loss": 0.3847,
+      "step": 3503
+    },
+    {
+      "epoch": 0.9025984480149403,
+      "grad_norm": 0.17852245830856514,
+      "learning_rate": 2.8617990346277657e-06,
+      "loss": 0.3825,
+      "step": 3504
+    },
+    {
+      "epoch": 0.9028560388962231,
+      "grad_norm": 0.2081650506236442,
+      "learning_rate": 2.8468224317977743e-06,
+      "loss": 0.4107,
+      "step": 3505
+    },
+    {
+      "epoch": 0.9028560388962231,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8862275449101796,
+      "eval_PRM F1 AUC": 0.7283918281822945,
+      "eval_PRM F1 AUC (fixed)": 0.858040859088528,
+      "eval_PRM F1 Neg": 0.5777777777777777,
+      "eval_PRM NPV": 0.5909090909090909,
+      "eval_PRM Precision": 0.8809523809523809,
+      "eval_PRM Recall": 0.891566265060241,
+      "eval_PRM Specificty": 0.5652173913043478,
+      "eval_loss": 0.3613725006580353,
+      "eval_runtime": 17.0225,
+      "eval_samples_per_second": 2.585,
+      "eval_steps_per_second": 0.176,
+      "step": 3505
+    },
+    {
+      "epoch": 0.9031136297775059,
+      "grad_norm": 0.1304410803037061,
+      "learning_rate": 2.831883971850213e-06,
+      "loss": 0.3199,
+      "step": 3506
+    },
+    {
+      "epoch": 0.9033712206587887,
+      "grad_norm": 0.3183315368975275,
+      "learning_rate": 2.8169836668690165e-06,
+      "loss": 0.517,
+      "step": 3507
+    },
+    {
+      "epoch": 0.9036288115400715,
+      "grad_norm": 0.14265599880347749,
+      "learning_rate": 2.802121528907242e-06,
+      "loss": 0.3872,
+      "step": 3508
+    },
+    {
+      "epoch": 0.9038864024213543,
+      "grad_norm": 0.15673814385728924,
+      "learning_rate": 2.787297569987107e-06,
+      "loss": 0.3693,
+      "step": 3509
+    },
+    {
+      "epoch": 0.9041439933026371,
+      "grad_norm": 0.17147029433004354,
+      "learning_rate": 2.7725118020998973e-06,
+      "loss": 0.4265,
+      "step": 3510
+    },
+    {
+      "epoch": 0.9041439933026371,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.891566265060241,
+      "eval_PRM F1 AUC": 0.7501309586170771,
+      "eval_PRM F1 AUC (fixed)": 0.8580408590885279,
+      "eval_PRM F1 Neg": 0.6086956521739131,
+      "eval_PRM NPV": 0.6086956521739131,
+      "eval_PRM Precision": 0.891566265060241,
+      "eval_PRM Recall": 0.891566265060241,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3582652807235718,
+      "eval_runtime": 17.1274,
+      "eval_samples_per_second": 2.569,
+      "eval_steps_per_second": 0.175,
+      "step": 3510
+    },
+    {
+      "epoch": 0.9044015841839199,
+      "grad_norm": 0.1294599974767688,
+      "learning_rate": 2.7577642372060673e-06,
+      "loss": 0.3366,
+      "step": 3511
+    },
+    {
+      "epoch": 0.9046591750652027,
+      "grad_norm": 0.10739872504317904,
+      "learning_rate": 2.7430548872351135e-06,
+      "loss": 0.2895,
+      "step": 3512
+    },
+    {
+      "epoch": 0.9049167659464855,
+      "grad_norm": 0.1337098599209687,
+      "learning_rate": 2.7283837640856346e-06,
+      "loss": 0.3331,
+      "step": 3513
+    },
+    {
+      "epoch": 0.9051743568277683,
+      "grad_norm": 0.15326670322972885,
+      "learning_rate": 2.7137508796253208e-06,
+      "loss": 0.3547,
+      "step": 3514
+    },
+    {
+      "epoch": 0.9054319477090511,
+      "grad_norm": 0.17439375071989902,
+      "learning_rate": 2.6991562456909205e-06,
+      "loss": 0.4062,
+      "step": 3515
+    },
+    {
+      "epoch": 0.9054319477090511,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8780487804878049,
+      "eval_PRM F1 AUC": 0.7380827658459927,
+      "eval_PRM F1 AUC (fixed)": 0.8543740178103719,
+      "eval_PRM F1 Neg": 0.5833333333333334,
+      "eval_PRM NPV": 0.56,
+      "eval_PRM Precision": 0.8888888888888888,
+      "eval_PRM Recall": 0.8674698795180723,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3610174059867859,
+      "eval_runtime": 17.031,
+      "eval_samples_per_second": 2.584,
+      "eval_steps_per_second": 0.176,
+      "step": 3515
+    },
+    {
+      "epoch": 0.9056895385903339,
+      "grad_norm": 0.14757466196871588,
+      "learning_rate": 2.684599874088256e-06,
+      "loss": 0.4083,
+      "step": 3516
+    },
+    {
+      "epoch": 0.9059471294716167,
+      "grad_norm": 0.13038628864544433,
+      "learning_rate": 2.670081776592165e-06,
+      "loss": 0.3017,
+      "step": 3517
+    },
+    {
+      "epoch": 0.9062047203528996,
+      "grad_norm": 0.1362571289704275,
+      "learning_rate": 2.6556019649465525e-06,
+      "loss": 0.2883,
+      "step": 3518
+    },
+    {
+      "epoch": 0.9064623112341823,
+      "grad_norm": 0.19400127869035802,
+      "learning_rate": 2.641160450864355e-06,
+      "loss": 0.455,
+      "step": 3519
+    },
+    {
+      "epoch": 0.9067199021154652,
+      "grad_norm": 0.1487023689144612,
+      "learning_rate": 2.626757246027506e-06,
+      "loss": 0.3572,
+      "step": 3520
+    },
+    {
+      "epoch": 0.9067199021154652,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8780487804878049,
+      "eval_PRM F1 AUC": 0.7380827658459927,
+      "eval_PRM F1 AUC (fixed)": 0.8559455212152959,
+      "eval_PRM F1 Neg": 0.5833333333333334,
+      "eval_PRM NPV": 0.56,
+      "eval_PRM Precision": 0.8888888888888888,
+      "eval_PRM Recall": 0.8674698795180723,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.36106178164482117,
+      "eval_runtime": 17.6175,
+      "eval_samples_per_second": 2.498,
+      "eval_steps_per_second": 0.17,
+      "step": 3520
+    },
+    {
+      "epoch": 0.9069774929967479,
+      "grad_norm": 0.1758953345140761,
+      "learning_rate": 2.6123923620869795e-06,
+      "loss": 0.4051,
+      "step": 3521
+    },
+    {
+      "epoch": 0.9072350838780308,
+      "grad_norm": 0.1603008294335897,
+      "learning_rate": 2.59806581066272e-06,
+      "loss": 0.3914,
+      "step": 3522
+    },
+    {
+      "epoch": 0.9074926747593135,
+      "grad_norm": 0.14475537508184771,
+      "learning_rate": 2.5837776033436954e-06,
+      "loss": 0.3153,
+      "step": 3523
+    },
+    {
+      "epoch": 0.9077502656405964,
+      "grad_norm": 0.1312022526870313,
+      "learning_rate": 2.5695277516878336e-06,
+      "loss": 0.2982,
+      "step": 3524
+    },
+    {
+      "epoch": 0.9080078565218791,
+      "grad_norm": 0.21749204403301972,
+      "learning_rate": 2.5553162672220465e-06,
+      "loss": 0.3939,
+      "step": 3525
+    },
+    {
+      "epoch": 0.9080078565218791,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8533263488737559,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3616832494735718,
+      "eval_runtime": 16.4985,
+      "eval_samples_per_second": 2.667,
+      "eval_steps_per_second": 0.182,
+      "step": 3525
+    },
+    {
+      "epoch": 0.908265447403162,
+      "grad_norm": 0.15509138928369606,
+      "learning_rate": 2.5411431614422053e-06,
+      "loss": 0.3359,
+      "step": 3526
+    },
+    {
+      "epoch": 0.9085230382844447,
+      "grad_norm": 0.12699302222708414,
+      "learning_rate": 2.5270084458131395e-06,
+      "loss": 0.2976,
+      "step": 3527
+    },
+    {
+      "epoch": 0.9087806291657275,
+      "grad_norm": 0.1781909885669849,
+      "learning_rate": 2.5129121317686356e-06,
+      "loss": 0.3845,
+      "step": 3528
+    },
+    {
+      "epoch": 0.9090382200470103,
+      "grad_norm": 0.15216059727780382,
+      "learning_rate": 2.4988542307113837e-06,
+      "loss": 0.3444,
+      "step": 3529
+    },
+    {
+      "epoch": 0.9092958109282931,
+      "grad_norm": 0.16200325190532375,
+      "learning_rate": 2.4848347540130434e-06,
+      "loss": 0.2597,
+      "step": 3530
+    },
+    {
+      "epoch": 0.9092958109282931,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8780487804878049,
+      "eval_PRM F1 AUC": 0.7380827658459927,
+      "eval_PRM F1 AUC (fixed)": 0.8530644316396019,
+      "eval_PRM F1 Neg": 0.5833333333333334,
+      "eval_PRM NPV": 0.56,
+      "eval_PRM Precision": 0.8888888888888888,
+      "eval_PRM Recall": 0.8674698795180723,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.36110618710517883,
+      "eval_runtime": 16.989,
+      "eval_samples_per_second": 2.59,
+      "eval_steps_per_second": 0.177,
+      "step": 3530
+    },
+    {
+      "epoch": 0.9095534018095759,
+      "grad_norm": 0.15168951575416167,
+      "learning_rate": 2.4708537130141485e-06,
+      "loss": 0.3409,
+      "step": 3531
+    },
+    {
+      "epoch": 0.9098109926908587,
+      "grad_norm": 0.15496414073672882,
+      "learning_rate": 2.45691111902418e-06,
+      "loss": 0.3449,
+      "step": 3532
+    },
+    {
+      "epoch": 0.9100685835721416,
+      "grad_norm": 0.3104512486482539,
+      "learning_rate": 2.4430069833214996e-06,
+      "loss": 0.4948,
+      "step": 3533
+    },
+    {
+      "epoch": 0.9103261744534243,
+      "grad_norm": 0.20727427059037964,
+      "learning_rate": 2.429141317153355e-06,
+      "loss": 0.455,
+      "step": 3534
+    },
+    {
+      "epoch": 0.9105837653347072,
+      "grad_norm": 0.1696170445047076,
+      "learning_rate": 2.415314131735885e-06,
+      "loss": 0.368,
+      "step": 3535
+    },
+    {
+      "epoch": 0.9105837653347072,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8567312729177579,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3587535619735718,
+      "eval_runtime": 16.9861,
+      "eval_samples_per_second": 2.59,
+      "eval_steps_per_second": 0.177,
+      "step": 3535
+    },
+    {
+      "epoch": 0.9108413562159899,
+      "grad_norm": 0.16578645048029902,
+      "learning_rate": 2.4015254382541043e-06,
+      "loss": 0.3967,
+      "step": 3536
+    },
+    {
+      "epoch": 0.9110989470972728,
+      "grad_norm": 0.14286368348379852,
+      "learning_rate": 2.38777524786189e-06,
+      "loss": 0.2688,
+      "step": 3537
+    },
+    {
+      "epoch": 0.9113565379785555,
+      "grad_norm": 0.12512391214687887,
+      "learning_rate": 2.374063571681956e-06,
+      "loss": 0.3187,
+      "step": 3538
+    },
+    {
+      "epoch": 0.9116141288598384,
+      "grad_norm": 0.17653831694965627,
+      "learning_rate": 2.360390420805869e-06,
+      "loss": 0.391,
+      "step": 3539
+    },
+    {
+      "epoch": 0.9118717197411211,
+      "grad_norm": 0.14848179498309094,
+      "learning_rate": 2.34675580629406e-06,
+      "loss": 0.4207,
+      "step": 3540
+    },
+    {
+      "epoch": 0.9118717197411211,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8780487804878049,
+      "eval_PRM F1 AUC": 0.7380827658459927,
+      "eval_PRM F1 AUC (fixed)": 0.8517548454688318,
+      "eval_PRM F1 Neg": 0.5833333333333334,
+      "eval_PRM NPV": 0.56,
+      "eval_PRM Precision": 0.8888888888888888,
+      "eval_PRM Recall": 0.8674698795180723,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35910865664482117,
+      "eval_runtime": 17.5619,
+      "eval_samples_per_second": 2.505,
+      "eval_steps_per_second": 0.171,
+      "step": 3540
+    },
+    {
+      "epoch": 0.912129310622404,
+      "grad_norm": 0.17358589157100177,
+      "learning_rate": 2.3331597391757442e-06,
+      "loss": 0.2874,
+      "step": 3541
+    },
+    {
+      "epoch": 0.9123869015036867,
+      "grad_norm": 0.17014683888786977,
+      "learning_rate": 2.3196022304489864e-06,
+      "loss": 0.3243,
+      "step": 3542
+    },
+    {
+      "epoch": 0.9126444923849696,
+      "grad_norm": 0.219578585265958,
+      "learning_rate": 2.306083291080641e-06,
+      "loss": 0.4471,
+      "step": 3543
+    },
+    {
+      "epoch": 0.9129020832662523,
+      "grad_norm": 0.13052088921236898,
+      "learning_rate": 2.292602932006377e-06,
+      "loss": 0.3245,
+      "step": 3544
+    },
+    {
+      "epoch": 0.9131596741475352,
+      "grad_norm": 0.1485674479766076,
+      "learning_rate": 2.2791611641306555e-06,
+      "loss": 0.305,
+      "step": 3545
+    },
+    {
+      "epoch": 0.9131596741475352,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8780487804878049,
+      "eval_PRM F1 AUC": 0.7380827658459927,
+      "eval_PRM F1 AUC (fixed)": 0.8533263488737558,
+      "eval_PRM F1 Neg": 0.5833333333333334,
+      "eval_PRM NPV": 0.56,
+      "eval_PRM Precision": 0.8888888888888888,
+      "eval_PRM Recall": 0.8674698795180723,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3580433130264282,
+      "eval_runtime": 17.4401,
+      "eval_samples_per_second": 2.523,
+      "eval_steps_per_second": 0.172,
+      "step": 3545
+    },
+    {
+      "epoch": 0.9134172650288179,
+      "grad_norm": 0.28276776229349954,
+      "learning_rate": 2.2657579983267064e-06,
+      "loss": 0.4108,
+      "step": 3546
+    },
+    {
+      "epoch": 0.9136748559101008,
+      "grad_norm": 0.15989949777721657,
+      "learning_rate": 2.252393445436546e-06,
+      "loss": 0.429,
+      "step": 3547
+    },
+    {
+      "epoch": 0.9139324467913836,
+      "grad_norm": 0.16830071915691178,
+      "learning_rate": 2.239067516270954e-06,
+      "loss": 0.3874,
+      "step": 3548
+    },
+    {
+      "epoch": 0.9141900376726664,
+      "grad_norm": 0.14365545128953014,
+      "learning_rate": 2.2257802216094746e-06,
+      "loss": 0.3974,
+      "step": 3549
+    },
+    {
+      "epoch": 0.9144476285539492,
+      "grad_norm": 0.10287303681402393,
+      "learning_rate": 2.212531572200377e-06,
+      "loss": 0.298,
+      "step": 3550
+    },
+    {
+      "epoch": 0.9144476285539492,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8780487804878049,
+      "eval_PRM F1 AUC": 0.7380827658459927,
+      "eval_PRM F1 AUC (fixed)": 0.8509690937663698,
+      "eval_PRM F1 Neg": 0.5833333333333334,
+      "eval_PRM NPV": 0.56,
+      "eval_PRM Precision": 0.8888888888888888,
+      "eval_PRM Recall": 0.8674698795180723,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3582652807235718,
+      "eval_runtime": 16.8653,
+      "eval_samples_per_second": 2.609,
+      "eval_steps_per_second": 0.178,
+      "step": 3550
+    },
+    {
+      "epoch": 0.914705219435232,
+      "grad_norm": 0.1414417999527504,
+      "learning_rate": 2.1993215787606936e-06,
+      "loss": 0.3516,
+      "step": 3551
+    },
+    {
+      "epoch": 0.9149628103165148,
+      "grad_norm": 0.2086615373938124,
+      "learning_rate": 2.186150251976171e-06,
+      "loss": 0.4494,
+      "step": 3552
+    },
+    {
+      "epoch": 0.9152204011977976,
+      "grad_norm": 0.15183181236880697,
+      "learning_rate": 2.1730176025012816e-06,
+      "loss": 0.3249,
+      "step": 3553
+    },
+    {
+      "epoch": 0.9154779920790804,
+      "grad_norm": 0.17879774948444485,
+      "learning_rate": 2.1599236409592327e-06,
+      "loss": 0.3687,
+      "step": 3554
+    },
+    {
+      "epoch": 0.9157355829603632,
+      "grad_norm": 0.22149387381725968,
+      "learning_rate": 2.1468683779418965e-06,
+      "loss": 0.4342,
+      "step": 3555
+    },
+    {
+      "epoch": 0.9157355829603632,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8780487804878049,
+      "eval_PRM F1 AUC": 0.7380827658459927,
+      "eval_PRM F1 AUC (fixed)": 0.8499214248297537,
+      "eval_PRM F1 Neg": 0.5833333333333334,
+      "eval_PRM NPV": 0.56,
+      "eval_PRM Precision": 0.8888888888888888,
+      "eval_PRM Recall": 0.8674698795180723,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3606622815132141,
+      "eval_runtime": 17.747,
+      "eval_samples_per_second": 2.479,
+      "eval_steps_per_second": 0.169,
+      "step": 3555
+    },
+    {
+      "epoch": 0.915993173841646,
+      "grad_norm": 0.15702597463200188,
+      "learning_rate": 2.1338518240098694e-06,
+      "loss": 0.4099,
+      "step": 3556
+    },
+    {
+      "epoch": 0.9162507647229288,
+      "grad_norm": 0.15451074192970757,
+      "learning_rate": 2.1208739896924345e-06,
+      "loss": 0.3728,
+      "step": 3557
+    },
+    {
+      "epoch": 0.9165083556042116,
+      "grad_norm": 0.15794402817072664,
+      "learning_rate": 2.1079348854875505e-06,
+      "loss": 0.3392,
+      "step": 3558
+    },
+    {
+      "epoch": 0.9167659464854944,
+      "grad_norm": 0.19678681465612674,
+      "learning_rate": 2.0950345218618328e-06,
+      "loss": 0.3692,
+      "step": 3559
+    },
+    {
+      "epoch": 0.9170235373667772,
+      "grad_norm": 0.1361028667540031,
+      "learning_rate": 2.082172909250568e-06,
+      "loss": 0.3296,
+      "step": 3560
+    },
+    {
+      "epoch": 0.9170235373667772,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8780487804878049,
+      "eval_PRM F1 AUC": 0.7380827658459927,
+      "eval_PRM F1 AUC (fixed)": 0.858040859088528,
+      "eval_PRM F1 Neg": 0.5833333333333334,
+      "eval_PRM NPV": 0.56,
+      "eval_PRM Precision": 0.8888888888888888,
+      "eval_PRM Recall": 0.8674698795180723,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35537996888160706,
+      "eval_runtime": 16.9677,
+      "eval_samples_per_second": 2.593,
+      "eval_steps_per_second": 0.177,
+      "step": 3560
+    },
+    {
+      "epoch": 0.9172811282480601,
+      "grad_norm": 0.12828516715375252,
+      "learning_rate": 2.069350058057723e-06,
+      "loss": 0.284,
+      "step": 3561
+    },
+    {
+      "epoch": 0.9175387191293428,
+      "grad_norm": 0.15589654522386157,
+      "learning_rate": 2.056565978655861e-06,
+      "loss": 0.2953,
+      "step": 3562
+    },
+    {
+      "epoch": 0.9177963100106257,
+      "grad_norm": 0.1922738703624335,
+      "learning_rate": 2.043820681386227e-06,
+      "loss": 0.3538,
+      "step": 3563
+    },
+    {
+      "epoch": 0.9180539008919084,
+      "grad_norm": 0.20272910116447893,
+      "learning_rate": 2.0311141765586573e-06,
+      "loss": 0.4282,
+      "step": 3564
+    },
+    {
+      "epoch": 0.9183114917731913,
+      "grad_norm": 0.16346346255570265,
+      "learning_rate": 2.0184464744516353e-06,
+      "loss": 0.3704,
+      "step": 3565
+    },
+    {
+      "epoch": 0.9183114917731913,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8780487804878049,
+      "eval_PRM F1 AUC": 0.7380827658459927,
+      "eval_PRM F1 AUC (fixed)": 0.8528025144054479,
+      "eval_PRM F1 Neg": 0.5833333333333334,
+      "eval_PRM NPV": 0.56,
+      "eval_PRM Precision": 0.8888888888888888,
+      "eval_PRM Recall": 0.8674698795180723,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35910865664482117,
+      "eval_runtime": 17.618,
+      "eval_samples_per_second": 2.497,
+      "eval_steps_per_second": 0.17,
+      "step": 3565
+    },
+    {
+      "epoch": 0.918569082654474,
+      "grad_norm": 0.1457485536576306,
+      "learning_rate": 2.005817585312253e-06,
+      "loss": 0.334,
+      "step": 3566
+    },
+    {
+      "epoch": 0.9188266735357569,
+      "grad_norm": 0.15443730771182385,
+      "learning_rate": 1.993227519356189e-06,
+      "loss": 0.3371,
+      "step": 3567
+    },
+    {
+      "epoch": 0.9190842644170396,
+      "grad_norm": 0.1680000043483095,
+      "learning_rate": 1.9806762867677296e-06,
+      "loss": 0.4377,
+      "step": 3568
+    },
+    {
+      "epoch": 0.9193418552983225,
+      "grad_norm": 0.14717374054767854,
+      "learning_rate": 1.9681638976997486e-06,
+      "loss": 0.3451,
+      "step": 3569
+    },
+    {
+      "epoch": 0.9195994461796052,
+      "grad_norm": 0.1242823379036716,
+      "learning_rate": 1.95569036227371e-06,
+      "loss": 0.3423,
+      "step": 3570
+    },
+    {
+      "epoch": 0.9195994461796052,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.85620743844945,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3561345934867859,
+      "eval_runtime": 17.0785,
+      "eval_samples_per_second": 2.576,
+      "eval_steps_per_second": 0.176,
+      "step": 3570
+    },
+    {
+      "epoch": 0.9198570370608881,
+      "grad_norm": 0.1169699351085616,
+      "learning_rate": 1.943255690579615e-06,
+      "loss": 0.2595,
+      "step": 3571
+    },
+    {
+      "epoch": 0.9201146279421708,
+      "grad_norm": 0.11198757782737513,
+      "learning_rate": 1.9308598926760625e-06,
+      "loss": 0.3334,
+      "step": 3572
+    },
+    {
+      "epoch": 0.9203722188234537,
+      "grad_norm": 0.14045385474378944,
+      "learning_rate": 1.9185029785901867e-06,
+      "loss": 0.3475,
+      "step": 3573
+    },
+    {
+      "epoch": 0.9206298097047364,
+      "grad_norm": 0.1654475938787749,
+      "learning_rate": 1.906184958317664e-06,
+      "loss": 0.3333,
+      "step": 3574
+    },
+    {
+      "epoch": 0.9208874005860193,
+      "grad_norm": 0.16327096309656713,
+      "learning_rate": 1.8939058418227406e-06,
+      "loss": 0.4313,
+      "step": 3575
+    },
+    {
+      "epoch": 0.9208874005860193,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.891566265060241,
+      "eval_PRM F1 AUC": 0.7501309586170771,
+      "eval_PRM F1 AUC (fixed)": 0.8548978522786799,
+      "eval_PRM F1 Neg": 0.6086956521739131,
+      "eval_PRM NPV": 0.6086956521739131,
+      "eval_PRM Precision": 0.891566265060241,
+      "eval_PRM Recall": 0.891566265060241,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35595703125,
+      "eval_runtime": 17.2808,
+      "eval_samples_per_second": 2.546,
+      "eval_steps_per_second": 0.174,
+      "step": 3575
+    },
+    {
+      "epoch": 0.9211449914673021,
+      "grad_norm": 0.17906720956662106,
+      "learning_rate": 1.8816656390381437e-06,
+      "loss": 0.3375,
+      "step": 3576
+    },
+    {
+      "epoch": 0.9214025823485849,
+      "grad_norm": 0.17656544233826593,
+      "learning_rate": 1.8694643598651695e-06,
+      "loss": 0.3924,
+      "step": 3577
+    },
+    {
+      "epoch": 0.9216601732298677,
+      "grad_norm": 0.13320038375003498,
+      "learning_rate": 1.8573020141736008e-06,
+      "loss": 0.3321,
+      "step": 3578
+    },
+    {
+      "epoch": 0.9219177641111505,
+      "grad_norm": 0.15680397389075254,
+      "learning_rate": 1.8451786118017234e-06,
+      "loss": 0.3439,
+      "step": 3579
+    },
+    {
+      "epoch": 0.9221753549924333,
+      "grad_norm": 0.1473731597244704,
+      "learning_rate": 1.8330941625563535e-06,
+      "loss": 0.4521,
+      "step": 3580
+    },
+    {
+      "epoch": 0.9221753549924333,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8575170246202201,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35458096861839294,
+      "eval_runtime": 16.9061,
+      "eval_samples_per_second": 2.603,
+      "eval_steps_per_second": 0.177,
+      "step": 3580
+    },
+    {
+      "epoch": 0.922432945873716,
+      "grad_norm": 0.13627264372745382,
+      "learning_rate": 1.8210486762127499e-06,
+      "loss": 0.3084,
+      "step": 3581
+    },
+    {
+      "epoch": 0.9226905367549989,
+      "grad_norm": 0.16310370490610962,
+      "learning_rate": 1.8090421625147013e-06,
+      "loss": 0.3037,
+      "step": 3582
+    },
+    {
+      "epoch": 0.9229481276362816,
+      "grad_norm": 0.14387730487937045,
+      "learning_rate": 1.7970746311744336e-06,
+      "loss": 0.4041,
+      "step": 3583
+    },
+    {
+      "epoch": 0.9232057185175645,
+      "grad_norm": 0.14068108820423889,
+      "learning_rate": 1.7851460918726748e-06,
+      "loss": 0.3223,
+      "step": 3584
+    },
+    {
+      "epoch": 0.9234633093988472,
+      "grad_norm": 0.1302142737116632,
+      "learning_rate": 1.7732565542585678e-06,
+      "loss": 0.315,
+      "step": 3585
+    },
+    {
+      "epoch": 0.9234633093988472,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.891566265060241,
+      "eval_PRM F1 AUC": 0.7501309586170771,
+      "eval_PRM F1 AUC (fixed)": 0.856993190151912,
+      "eval_PRM F1 Neg": 0.6086956521739131,
+      "eval_PRM NPV": 0.6086956521739131,
+      "eval_PRM Precision": 0.891566265060241,
+      "eval_PRM Recall": 0.891566265060241,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35946378111839294,
+      "eval_runtime": 17.6897,
+      "eval_samples_per_second": 2.487,
+      "eval_steps_per_second": 0.17,
+      "step": 3585
+    },
+    {
+      "epoch": 0.9237209002801301,
+      "grad_norm": 0.1446479054321807,
+      "learning_rate": 1.7614060279497413e-06,
+      "loss": 0.3067,
+      "step": 3586
+    },
+    {
+      "epoch": 0.9239784911614128,
+      "grad_norm": 0.13510928230906516,
+      "learning_rate": 1.7495945225322607e-06,
+      "loss": 0.2805,
+      "step": 3587
+    },
+    {
+      "epoch": 0.9242360820426957,
+      "grad_norm": 0.15398682236678907,
+      "learning_rate": 1.737822047560611e-06,
+      "loss": 0.3182,
+      "step": 3588
+    },
+    {
+      "epoch": 0.9244936729239784,
+      "grad_norm": 0.1598075897613883,
+      "learning_rate": 1.7260886125577247e-06,
+      "loss": 0.4169,
+      "step": 3589
+    },
+    {
+      "epoch": 0.9247512638052613,
+      "grad_norm": 0.15869750034866761,
+      "learning_rate": 1.7143942270149427e-06,
+      "loss": 0.3449,
+      "step": 3590
+    },
+    {
+      "epoch": 0.9247512638052613,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.891566265060241,
+      "eval_PRM F1 AUC": 0.7501309586170771,
+      "eval_PRM F1 AUC (fixed)": 0.8585646935568361,
+      "eval_PRM F1 Neg": 0.6086956521739131,
+      "eval_PRM NPV": 0.6086956521739131,
+      "eval_PRM Precision": 0.891566265060241,
+      "eval_PRM Recall": 0.891566265060241,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35830965638160706,
+      "eval_runtime": 16.3997,
+      "eval_samples_per_second": 2.683,
+      "eval_steps_per_second": 0.183,
+      "step": 3590
+    },
+    {
+      "epoch": 0.9250088546865441,
+      "grad_norm": 0.15858603689819276,
+      "learning_rate": 1.7027389003920313e-06,
+      "loss": 0.3533,
+      "step": 3591
+    },
+    {
+      "epoch": 0.9252664455678269,
+      "grad_norm": 0.15268138246759289,
+      "learning_rate": 1.6911226421171378e-06,
+      "loss": 0.3639,
+      "step": 3592
+    },
+    {
+      "epoch": 0.9255240364491097,
+      "grad_norm": 0.1692437619504843,
+      "learning_rate": 1.6795454615868234e-06,
+      "loss": 0.3816,
+      "step": 3593
+    },
+    {
+      "epoch": 0.9257816273303925,
+      "grad_norm": 0.16953837399894867,
+      "learning_rate": 1.6680073681660358e-06,
+      "loss": 0.401,
+      "step": 3594
+    },
+    {
+      "epoch": 0.9260392182116753,
+      "grad_norm": 0.19482660347644706,
+      "learning_rate": 1.656508371188109e-06,
+      "loss": 0.2827,
+      "step": 3595
+    },
+    {
+      "epoch": 0.9260392182116753,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.859874279727606,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35520240664482117,
+      "eval_runtime": 17.2187,
+      "eval_samples_per_second": 2.555,
+      "eval_steps_per_second": 0.174,
+      "step": 3595
+    },
+    {
+      "epoch": 0.9262968090929581,
+      "grad_norm": 0.13849159515883916,
+      "learning_rate": 1.6450484799547473e-06,
+      "loss": 0.4034,
+      "step": 3596
+    },
+    {
+      "epoch": 0.9265543999742409,
+      "grad_norm": 0.17046521701031161,
+      "learning_rate": 1.6336277037360127e-06,
+      "loss": 0.3568,
+      "step": 3597
+    },
+    {
+      "epoch": 0.9268119908555237,
+      "grad_norm": 0.17147074585372724,
+      "learning_rate": 1.622246051770343e-06,
+      "loss": 0.4252,
+      "step": 3598
+    },
+    {
+      "epoch": 0.9270695817368065,
+      "grad_norm": 0.17687896763490749,
+      "learning_rate": 1.6109035332645183e-06,
+      "loss": 0.3575,
+      "step": 3599
+    },
+    {
+      "epoch": 0.9273271726180893,
+      "grad_norm": 0.1830855633473594,
+      "learning_rate": 1.5996001573936604e-06,
+      "loss": 0.3733,
+      "step": 3600
+    },
+    {
+      "epoch": 0.9273271726180893,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8593504452592982,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3565784692764282,
+      "eval_runtime": 16.8833,
+      "eval_samples_per_second": 2.606,
+      "eval_steps_per_second": 0.178,
+      "step": 3600
+    },
+    {
+      "epoch": 0.9275847634993721,
+      "grad_norm": 0.17321490179978577,
+      "learning_rate": 1.5883359333012438e-06,
+      "loss": 0.3897,
+      "step": 3601
+    },
+    {
+      "epoch": 0.9278423543806549,
+      "grad_norm": 0.15898382741093173,
+      "learning_rate": 1.5771108700990412e-06,
+      "loss": 0.3738,
+      "step": 3602
+    },
+    {
+      "epoch": 0.9280999452619377,
+      "grad_norm": 0.1391967307095472,
+      "learning_rate": 1.5659249768671835e-06,
+      "loss": 0.3337,
+      "step": 3603
+    },
+    {
+      "epoch": 0.9283575361432206,
+      "grad_norm": 0.12481251556632249,
+      "learning_rate": 1.5547782626540941e-06,
+      "loss": 0.3264,
+      "step": 3604
+    },
+    {
+      "epoch": 0.9286151270245033,
+      "grad_norm": 0.1582670706657496,
+      "learning_rate": 1.5436707364765213e-06,
+      "loss": 0.3687,
+      "step": 3605
+    },
+    {
+      "epoch": 0.9286151270245033,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8606600314300681,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3534712493419647,
+      "eval_runtime": 17.7365,
+      "eval_samples_per_second": 2.481,
+      "eval_steps_per_second": 0.169,
+      "step": 3605
+    },
+    {
+      "epoch": 0.9288727179057862,
+      "grad_norm": 0.1962104659388427,
+      "learning_rate": 1.5326024073194834e-06,
+      "loss": 0.4026,
+      "step": 3606
+    },
+    {
+      "epoch": 0.9291303087870689,
+      "grad_norm": 0.13502688190956327,
+      "learning_rate": 1.5215732841363239e-06,
+      "loss": 0.2946,
+      "step": 3607
+    },
+    {
+      "epoch": 0.9293878996683518,
+      "grad_norm": 0.11718572355068893,
+      "learning_rate": 1.5105833758486621e-06,
+      "loss": 0.298,
+      "step": 3608
+    },
+    {
+      "epoch": 0.9296454905496345,
+      "grad_norm": 0.16677425259863735,
+      "learning_rate": 1.4996326913463754e-06,
+      "loss": 0.2994,
+      "step": 3609
+    },
+    {
+      "epoch": 0.9299030814309174,
+      "grad_norm": 0.15552722462552468,
+      "learning_rate": 1.4887212394876503e-06,
+      "loss": 0.3641,
+      "step": 3610
+    },
+    {
+      "epoch": 0.9299030814309174,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8593504452592982,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3546253442764282,
+      "eval_runtime": 17.1911,
+      "eval_samples_per_second": 2.559,
+      "eval_steps_per_second": 0.175,
+      "step": 3610
+    },
+    {
+      "epoch": 0.9301606723122001,
+      "grad_norm": 0.1307387602339093,
+      "learning_rate": 1.4778490290988934e-06,
+      "loss": 0.3407,
+      "step": 3611
+    },
+    {
+      "epoch": 0.930418263193483,
+      "grad_norm": 0.2251240226399944,
+      "learning_rate": 1.467016068974819e-06,
+      "loss": 0.4252,
+      "step": 3612
+    },
+    {
+      "epoch": 0.9306758540747657,
+      "grad_norm": 0.1225706200048114,
+      "learning_rate": 1.4562223678783516e-06,
+      "loss": 0.2744,
+      "step": 3613
+    },
+    {
+      "epoch": 0.9309334449560486,
+      "grad_norm": 0.14564700385711848,
+      "learning_rate": 1.4454679345406619e-06,
+      "loss": 0.3934,
+      "step": 3614
+    },
+    {
+      "epoch": 0.9311910358373313,
+      "grad_norm": 0.16876178194128658,
+      "learning_rate": 1.4347527776611746e-06,
+      "loss": 0.3533,
+      "step": 3615
+    },
+    {
+      "epoch": 0.9311910358373313,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.859350445259298,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3512073755264282,
+      "eval_runtime": 17.5668,
+      "eval_samples_per_second": 2.505,
+      "eval_steps_per_second": 0.171,
+      "step": 3615
+    },
+    {
+      "epoch": 0.9314486267186142,
+      "grad_norm": 0.1525862990991277,
+      "learning_rate": 1.4240769059075342e-06,
+      "loss": 0.2566,
+      "step": 3616
+    },
+    {
+      "epoch": 0.9317062175998969,
+      "grad_norm": 0.14940505072760127,
+      "learning_rate": 1.4134403279156106e-06,
+      "loss": 0.2961,
+      "step": 3617
+    },
+    {
+      "epoch": 0.9319638084811798,
+      "grad_norm": 0.15374299685837137,
+      "learning_rate": 1.4028430522894765e-06,
+      "loss": 0.3751,
+      "step": 3618
+    },
+    {
+      "epoch": 0.9322213993624626,
+      "grad_norm": 0.1341987628806026,
+      "learning_rate": 1.3922850876014192e-06,
+      "loss": 0.343,
+      "step": 3619
+    },
+    {
+      "epoch": 0.9324789902437454,
+      "grad_norm": 0.13654672843056637,
+      "learning_rate": 1.3817664423919351e-06,
+      "loss": 0.3156,
+      "step": 3620
+    },
+    {
+      "epoch": 0.9324789902437454,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.891566265060241,
+      "eval_PRM F1 AUC": 0.7501309586170771,
+      "eval_PRM F1 AUC (fixed)": 0.8588266107909901,
+      "eval_PRM F1 Neg": 0.6086956521739131,
+      "eval_PRM NPV": 0.6086956521739131,
+      "eval_PRM Precision": 0.891566265060241,
+      "eval_PRM Recall": 0.891566265060241,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.353515625,
+      "eval_runtime": 17.5501,
+      "eval_samples_per_second": 2.507,
+      "eval_steps_per_second": 0.171,
+      "step": 3620
+    },
+    {
+      "epoch": 0.9327365811250282,
+      "grad_norm": 0.16672504814855243,
+      "learning_rate": 1.3712871251697012e-06,
+      "loss": 0.3681,
+      "step": 3621
+    },
+    {
+      "epoch": 0.932994172006311,
+      "grad_norm": 0.19594028408110487,
+      "learning_rate": 1.3608471444115977e-06,
+      "loss": 0.3699,
+      "step": 3622
+    },
+    {
+      "epoch": 0.9332517628875938,
+      "grad_norm": 0.17086512939183604,
+      "learning_rate": 1.3504465085626638e-06,
+      "loss": 0.4609,
+      "step": 3623
+    },
+    {
+      "epoch": 0.9335093537688766,
+      "grad_norm": 0.17268770687487686,
+      "learning_rate": 1.3400852260361252e-06,
+      "loss": 0.3939,
+      "step": 3624
+    },
+    {
+      "epoch": 0.9337669446501594,
+      "grad_norm": 0.14195089254333754,
+      "learning_rate": 1.3297633052133718e-06,
+      "loss": 0.3224,
+      "step": 3625
+    },
+    {
+      "epoch": 0.9337669446501594,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8643268727082242,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3525390625,
+      "eval_runtime": 17.1593,
+      "eval_samples_per_second": 2.564,
+      "eval_steps_per_second": 0.175,
+      "step": 3625
+    },
+    {
+      "epoch": 0.9340245355314422,
+      "grad_norm": 0.15892233470683617,
+      "learning_rate": 1.3194807544439635e-06,
+      "loss": 0.373,
+      "step": 3626
+    },
+    {
+      "epoch": 0.934282126412725,
+      "grad_norm": 0.13537545563985226,
+      "learning_rate": 1.3092375820455916e-06,
+      "loss": 0.3124,
+      "step": 3627
+    },
+    {
+      "epoch": 0.9345397172940078,
+      "grad_norm": 0.15225779088304373,
+      "learning_rate": 1.299033796304111e-06,
+      "loss": 0.382,
+      "step": 3628
+    },
+    {
+      "epoch": 0.9347973081752906,
+      "grad_norm": 0.16354081536121157,
+      "learning_rate": 1.2888694054735085e-06,
+      "loss": 0.3384,
+      "step": 3629
+    },
+    {
+      "epoch": 0.9350548990565734,
+      "grad_norm": 0.15028260915943015,
+      "learning_rate": 1.2787444177759068e-06,
+      "loss": 0.3891,
+      "step": 3630
+    },
+    {
+      "epoch": 0.9350548990565734,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8795180722891566,
+      "eval_PRM F1 AUC": 0.7223677317967523,
+      "eval_PRM F1 AUC (fixed)": 0.8596123624934522,
+      "eval_PRM F1 Neg": 0.5652173913043478,
+      "eval_PRM NPV": 0.5652173913043478,
+      "eval_PRM Precision": 0.8795180722891566,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.5652173913043478,
+      "eval_loss": 0.3546253442764282,
+      "eval_runtime": 17.0396,
+      "eval_samples_per_second": 2.582,
+      "eval_steps_per_second": 0.176,
+      "step": 3630
+    },
+    {
+      "epoch": 0.9353124899378562,
+      "grad_norm": 0.15976214590838345,
+      "learning_rate": 1.2686588414015543e-06,
+      "loss": 0.3122,
+      "step": 3631
+    },
+    {
+      "epoch": 0.935570080819139,
+      "grad_norm": 0.13884197323278308,
+      "learning_rate": 1.2586126845088086e-06,
+      "loss": 0.3325,
+      "step": 3632
+    },
+    {
+      "epoch": 0.9358276717004218,
+      "grad_norm": 0.15873924916917725,
+      "learning_rate": 1.248605955224169e-06,
+      "loss": 0.3719,
+      "step": 3633
+    },
+    {
+      "epoch": 0.9360852625817047,
+      "grad_norm": 0.17844417291816023,
+      "learning_rate": 1.2386386616422164e-06,
+      "loss": 0.3619,
+      "step": 3634
+    },
+    {
+      "epoch": 0.9363428534629874,
+      "grad_norm": 0.1429127723093715,
+      "learning_rate": 1.228710811825623e-06,
+      "loss": 0.3363,
+      "step": 3635
+    },
+    {
+      "epoch": 0.9363428534629874,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.891566265060241,
+      "eval_PRM F1 AUC": 0.7501309586170771,
+      "eval_PRM F1 AUC (fixed)": 0.8577789418543739,
+      "eval_PRM F1 Neg": 0.6086956521739131,
+      "eval_PRM NPV": 0.6086956521739131,
+      "eval_PRM Precision": 0.891566265060241,
+      "eval_PRM Recall": 0.891566265060241,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3564453125,
+      "eval_runtime": 17.1988,
+      "eval_samples_per_second": 2.558,
+      "eval_steps_per_second": 0.174,
+      "step": 3635
+    },
+    {
+      "epoch": 0.9366004443442703,
+      "grad_norm": 0.143553936055149,
+      "learning_rate": 1.2188224138051874e-06,
+      "loss": 0.3573,
+      "step": 3636
+    },
+    {
+      "epoch": 0.936858035225553,
+      "grad_norm": 0.1570089595566499,
+      "learning_rate": 1.208973475579761e-06,
+      "loss": 0.3635,
+      "step": 3637
+    },
+    {
+      "epoch": 0.9371156261068359,
+      "grad_norm": 0.1636449864279838,
+      "learning_rate": 1.19916400511631e-06,
+      "loss": 0.4452,
+      "step": 3638
+    },
+    {
+      "epoch": 0.9373732169881186,
+      "grad_norm": 0.1434591093458333,
+      "learning_rate": 1.189394010349848e-06,
+      "loss": 0.3512,
+      "step": 3639
+    },
+    {
+      "epoch": 0.9376308078694015,
+      "grad_norm": 0.1845146332799933,
+      "learning_rate": 1.1796634991834476e-06,
+      "loss": 0.3987,
+      "step": 3640
+    },
+    {
+      "epoch": 0.9376308078694015,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.891566265060241,
+      "eval_PRM F1 AUC": 0.7501309586170771,
+      "eval_PRM F1 AUC (fixed)": 0.8609219486642221,
+      "eval_PRM F1 Neg": 0.6086956521739131,
+      "eval_PRM NPV": 0.6086956521739131,
+      "eval_PRM Precision": 0.891566265060241,
+      "eval_PRM Recall": 0.891566265060241,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3558238744735718,
+      "eval_runtime": 17.0517,
+      "eval_samples_per_second": 2.58,
+      "eval_steps_per_second": 0.176,
+      "step": 3640
+    },
+    {
+      "epoch": 0.9378883987506842,
+      "grad_norm": 0.1850052147374829,
+      "learning_rate": 1.1699724794882849e-06,
+      "loss": 0.4093,
+      "step": 3641
+    },
+    {
+      "epoch": 0.938145989631967,
+      "grad_norm": 0.140018455666938,
+      "learning_rate": 1.1603209591035447e-06,
+      "loss": 0.3434,
+      "step": 3642
+    },
+    {
+      "epoch": 0.9384035805132498,
+      "grad_norm": 0.19723646319820293,
+      "learning_rate": 1.1507089458364984e-06,
+      "loss": 0.3912,
+      "step": 3643
+    },
+    {
+      "epoch": 0.9386611713945326,
+      "grad_norm": 0.13679557778401036,
+      "learning_rate": 1.1411364474624264e-06,
+      "loss": 0.2239,
+      "step": 3644
+    },
+    {
+      "epoch": 0.9389187622758154,
+      "grad_norm": 0.11434361612107999,
+      "learning_rate": 1.1316034717246626e-06,
+      "loss": 0.3308,
+      "step": 3645
+    },
+    {
+      "epoch": 0.9389187622758154,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.891566265060241,
+      "eval_PRM F1 AUC": 0.7501309586170771,
+      "eval_PRM F1 AUC (fixed)": 0.8624934520691461,
+      "eval_PRM F1 Neg": 0.6086956521739131,
+      "eval_PRM NPV": 0.6086956521739131,
+      "eval_PRM Precision": 0.891566265060241,
+      "eval_PRM Recall": 0.891566265060241,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35205078125,
+      "eval_runtime": 17.0238,
+      "eval_samples_per_second": 2.585,
+      "eval_steps_per_second": 0.176,
+      "step": 3645
+    },
+    {
+      "epoch": 0.9391763531570982,
+      "grad_norm": 0.13629942244093662,
+      "learning_rate": 1.1221100263345773e-06,
+      "loss": 0.3062,
+      "step": 3646
+    },
+    {
+      "epoch": 0.939433944038381,
+      "grad_norm": 0.19287017509156754,
+      "learning_rate": 1.1126561189715502e-06,
+      "loss": 0.3842,
+      "step": 3647
+    },
+    {
+      "epoch": 0.9396915349196638,
+      "grad_norm": 0.13998401829250906,
+      "learning_rate": 1.1032417572829745e-06,
+      "loss": 0.3509,
+      "step": 3648
+    },
+    {
+      "epoch": 0.9399491258009467,
+      "grad_norm": 0.17528116018871204,
+      "learning_rate": 1.0938669488842702e-06,
+      "loss": 0.4325,
+      "step": 3649
+    },
+    {
+      "epoch": 0.9402067166822294,
+      "grad_norm": 0.15921787864544712,
+      "learning_rate": 1.0845317013588596e-06,
+      "loss": 0.3764,
+      "step": 3650
+    },
+    {
+      "epoch": 0.9402067166822294,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.891566265060241,
+      "eval_PRM F1 AUC": 0.7501309586170771,
+      "eval_PRM F1 AUC (fixed)": 0.859874279727606,
+      "eval_PRM F1 Neg": 0.6086956521739131,
+      "eval_PRM NPV": 0.6086956521739131,
+      "eval_PRM Precision": 0.891566265060241,
+      "eval_PRM Recall": 0.891566265060241,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35324928164482117,
+      "eval_runtime": 17.04,
+      "eval_samples_per_second": 2.582,
+      "eval_steps_per_second": 0.176,
+      "step": 3650
+    },
+    {
+      "epoch": 0.9404643075635123,
+      "grad_norm": 0.193952286619382,
+      "learning_rate": 1.075236022258147e-06,
+      "loss": 0.3405,
+      "step": 3651
+    },
+    {
+      "epoch": 0.940721898444795,
+      "grad_norm": 0.26518998616866324,
+      "learning_rate": 1.0659799191015507e-06,
+      "loss": 0.4643,
+      "step": 3652
+    },
+    {
+      "epoch": 0.9409794893260779,
+      "grad_norm": 0.13543714158708117,
+      "learning_rate": 1.0567633993764537e-06,
+      "loss": 0.273,
+      "step": 3653
+    },
+    {
+      "epoch": 0.9412370802073606,
+      "grad_norm": 0.13373048613174426,
+      "learning_rate": 1.0475864705382422e-06,
+      "loss": 0.3098,
+      "step": 3654
+    },
+    {
+      "epoch": 0.9414946710886435,
+      "grad_norm": 0.15415902218239402,
+      "learning_rate": 1.0384491400102614e-06,
+      "loss": 0.3667,
+      "step": 3655
+    },
+    {
+      "epoch": 0.9414946710886435,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8609219486642221,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3531605005264282,
+      "eval_runtime": 16.5343,
+      "eval_samples_per_second": 2.661,
+      "eval_steps_per_second": 0.181,
+      "step": 3655
+    },
+    {
+      "epoch": 0.9417522619699262,
+      "grad_norm": 0.16902528190442034,
+      "learning_rate": 1.0293514151838268e-06,
+      "loss": 0.4194,
+      "step": 3656
+    },
+    {
+      "epoch": 0.9420098528512091,
+      "grad_norm": 0.1837710083377154,
+      "learning_rate": 1.0202933034182237e-06,
+      "loss": 0.3673,
+      "step": 3657
+    },
+    {
+      "epoch": 0.9422674437324918,
+      "grad_norm": 0.155257115669239,
+      "learning_rate": 1.0112748120406856e-06,
+      "loss": 0.3239,
+      "step": 3658
+    },
+    {
+      "epoch": 0.9425250346137747,
+      "grad_norm": 0.1603155333470742,
+      "learning_rate": 1.0022959483464045e-06,
+      "loss": 0.3068,
+      "step": 3659
+    },
+    {
+      "epoch": 0.9427826254950574,
+      "grad_norm": 0.19221021847619366,
+      "learning_rate": 9.933567195985094e-07,
+      "loss": 0.4736,
+      "step": 3660
+    },
+    {
+      "epoch": 0.9427826254950574,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8619696176008381,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3556463122367859,
+      "eval_runtime": 17.1711,
+      "eval_samples_per_second": 2.562,
+      "eval_steps_per_second": 0.175,
+      "step": 3660
+    },
+    {
+      "epoch": 0.9430402163763403,
+      "grad_norm": 0.16379670288583872,
+      "learning_rate": 9.844571330280549e-07,
+      "loss": 0.4013,
+      "step": 3661
+    },
+    {
+      "epoch": 0.9432978072576231,
+      "grad_norm": 0.17665040568388132,
+      "learning_rate": 9.75597195834077e-07,
+      "loss": 0.3604,
+      "step": 3662
+    },
+    {
+      "epoch": 0.9435553981389059,
+      "grad_norm": 0.1694285296370909,
+      "learning_rate": 9.66776915183476e-07,
+      "loss": 0.3601,
+      "step": 3663
+    },
+    {
+      "epoch": 0.9438129890201887,
+      "grad_norm": 0.12650901480029578,
+      "learning_rate": 9.579962982111224e-07,
+      "loss": 0.3239,
+      "step": 3664
+    },
+    {
+      "epoch": 0.9440705799014715,
+      "grad_norm": 0.14756402561710022,
+      "learning_rate": 9.492553520197733e-07,
+      "loss": 0.4006,
+      "step": 3665
+    },
+    {
+      "epoch": 0.9440705799014715,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8609219486642221,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3529829680919647,
+      "eval_runtime": 17.0353,
+      "eval_samples_per_second": 2.583,
+      "eval_steps_per_second": 0.176,
+      "step": 3665
+    },
+    {
+      "epoch": 0.9443281707827543,
+      "grad_norm": 0.17854362312944283,
+      "learning_rate": 9.405540836801119e-07,
+      "loss": 0.3465,
+      "step": 3666
+    },
+    {
+      "epoch": 0.9445857616640371,
+      "grad_norm": 0.17709937002340748,
+      "learning_rate": 9.318925002307188e-07,
+      "loss": 0.3149,
+      "step": 3667
+    },
+    {
+      "epoch": 0.9448433525453199,
+      "grad_norm": 0.172973032323487,
+      "learning_rate": 9.232706086780619e-07,
+      "loss": 0.3452,
+      "step": 3668
+    },
+    {
+      "epoch": 0.9451009434266027,
+      "grad_norm": 0.1345121785044483,
+      "learning_rate": 9.146884159965286e-07,
+      "loss": 0.2891,
+      "step": 3669
+    },
+    {
+      "epoch": 0.9453585343078855,
+      "grad_norm": 0.13728174842331684,
+      "learning_rate": 9.061459291283658e-07,
+      "loss": 0.3148,
+      "step": 3670
+    },
+    {
+      "epoch": 0.9453585343078855,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.858040859088528,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3525390625,
+      "eval_runtime": 16.9543,
+      "eval_samples_per_second": 2.595,
+      "eval_steps_per_second": 0.177,
+      "step": 3670
+    },
+    {
+      "epoch": 0.9456161251891683,
+      "grad_norm": 0.12179028171270745,
+      "learning_rate": 8.976431549837239e-07,
+      "loss": 0.2727,
+      "step": 3671
+    },
+    {
+      "epoch": 0.9458737160704511,
+      "grad_norm": 0.20292642131673685,
+      "learning_rate": 8.891801004406119e-07,
+      "loss": 0.4476,
+      "step": 3672
+    },
+    {
+      "epoch": 0.9461313069517339,
+      "grad_norm": 0.2117910852732907,
+      "learning_rate": 8.807567723449206e-07,
+      "loss": 0.4602,
+      "step": 3673
+    },
+    {
+      "epoch": 0.9463888978330167,
+      "grad_norm": 0.17428962139928333,
+      "learning_rate": 8.723731775103938e-07,
+      "loss": 0.4595,
+      "step": 3674
+    },
+    {
+      "epoch": 0.9466464887142995,
+      "grad_norm": 0.2090642216310755,
+      "learning_rate": 8.640293227186513e-07,
+      "loss": 0.4209,
+      "step": 3675
+    },
+    {
+      "epoch": 0.9466464887142995,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.859874279727606,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35422584414482117,
+      "eval_runtime": 16.9944,
+      "eval_samples_per_second": 2.589,
+      "eval_steps_per_second": 0.177,
+      "step": 3675
+    },
+    {
+      "epoch": 0.9469040795955823,
+      "grad_norm": 0.1414186990108163,
+      "learning_rate": 8.557252147191497e-07,
+      "loss": 0.3868,
+      "step": 3676
+    },
+    {
+      "epoch": 0.9471616704768652,
+      "grad_norm": 0.14781947639985626,
+      "learning_rate": 8.474608602292044e-07,
+      "loss": 0.2981,
+      "step": 3677
+    },
+    {
+      "epoch": 0.9474192613581479,
+      "grad_norm": 0.14013584779347527,
+      "learning_rate": 8.392362659339681e-07,
+      "loss": 0.3697,
+      "step": 3678
+    },
+    {
+      "epoch": 0.9476768522394308,
+      "grad_norm": 0.15392211696114294,
+      "learning_rate": 8.31051438486441e-07,
+      "loss": 0.3278,
+      "step": 3679
+    },
+    {
+      "epoch": 0.9479344431207135,
+      "grad_norm": 0.2299225601108455,
+      "learning_rate": 8.22906384507438e-07,
+      "loss": 0.422,
+      "step": 3680
+    },
+    {
+      "epoch": 0.9479344431207135,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8548978522786799,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3544478118419647,
+      "eval_runtime": 17.2464,
+      "eval_samples_per_second": 2.551,
+      "eval_steps_per_second": 0.174,
+      "step": 3680
+    },
+    {
+      "epoch": 0.9481920340019964,
+      "grad_norm": 0.16274285836944163,
+      "learning_rate": 8.148011105856168e-07,
+      "loss": 0.4064,
+      "step": 3681
+    },
+    {
+      "epoch": 0.9484496248832791,
+      "grad_norm": 0.20910361953817333,
+      "learning_rate": 8.067356232774437e-07,
+      "loss": 0.473,
+      "step": 3682
+    },
+    {
+      "epoch": 0.948707215764562,
+      "grad_norm": 0.17748660504950942,
+      "learning_rate": 7.987099291072109e-07,
+      "loss": 0.3576,
+      "step": 3683
+    },
+    {
+      "epoch": 0.9489648066458447,
+      "grad_norm": 0.1578814384448594,
+      "learning_rate": 7.907240345670142e-07,
+      "loss": 0.3753,
+      "step": 3684
+    },
+    {
+      "epoch": 0.9492223975271276,
+      "grad_norm": 0.15212521856275274,
+      "learning_rate": 7.827779461167695e-07,
+      "loss": 0.3193,
+      "step": 3685
+    },
+    {
+      "epoch": 0.9492223975271276,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.891566265060241,
+      "eval_PRM F1 AUC": 0.7501309586170771,
+      "eval_PRM F1 AUC (fixed)": 0.861707700366684,
+      "eval_PRM F1 Neg": 0.6086956521739131,
+      "eval_PRM NPV": 0.6086956521739131,
+      "eval_PRM Precision": 0.891566265060241,
+      "eval_PRM Recall": 0.891566265060241,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3528497815132141,
+      "eval_runtime": 17.6253,
+      "eval_samples_per_second": 2.496,
+      "eval_steps_per_second": 0.17,
+      "step": 3685
+    },
+    {
+      "epoch": 0.9494799884084103,
+      "grad_norm": 0.18749221830312565,
+      "learning_rate": 7.748716701841685e-07,
+      "loss": 0.3724,
+      "step": 3686
+    },
+    {
+      "epoch": 0.9497375792896932,
+      "grad_norm": 0.1891080708391055,
+      "learning_rate": 7.670052131647121e-07,
+      "loss": 0.389,
+      "step": 3687
+    },
+    {
+      "epoch": 0.9499951701709759,
+      "grad_norm": 0.21227202340748516,
+      "learning_rate": 7.591785814217046e-07,
+      "loss": 0.3401,
+      "step": 3688
+    },
+    {
+      "epoch": 0.9502527610522588,
+      "grad_norm": 0.1648754080971513,
+      "learning_rate": 7.513917812862037e-07,
+      "loss": 0.3453,
+      "step": 3689
+    },
+    {
+      "epoch": 0.9505103519335415,
+      "grad_norm": 0.19280860950145984,
+      "learning_rate": 7.436448190570766e-07,
+      "loss": 0.3472,
+      "step": 3690
+    },
+    {
+      "epoch": 0.9505103519335415,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.859874279727606,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3533824682235718,
+      "eval_runtime": 17.2241,
+      "eval_samples_per_second": 2.555,
+      "eval_steps_per_second": 0.174,
+      "step": 3690
+    },
+    {
+      "epoch": 0.9507679428148244,
+      "grad_norm": 0.14333195087175796,
+      "learning_rate": 7.359377010009383e-07,
+      "loss": 0.2865,
+      "step": 3691
+    },
+    {
+      "epoch": 0.9510255336961072,
+      "grad_norm": 0.18370111816292833,
+      "learning_rate": 7.282704333522017e-07,
+      "loss": 0.3927,
+      "step": 3692
+    },
+    {
+      "epoch": 0.95128312457739,
+      "grad_norm": 0.1728863568975308,
+      "learning_rate": 7.206430223130278e-07,
+      "loss": 0.3012,
+      "step": 3693
+    },
+    {
+      "epoch": 0.9515407154586728,
+      "grad_norm": 0.15169059013242978,
+      "learning_rate": 7.130554740533313e-07,
+      "loss": 0.3035,
+      "step": 3694
+    },
+    {
+      "epoch": 0.9517983063399555,
+      "grad_norm": 0.13965884246701635,
+      "learning_rate": 7.055077947107858e-07,
+      "loss": 0.3257,
+      "step": 3695
+    },
+    {
+      "epoch": 0.9517983063399555,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.859874279727606,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35205078125,
+      "eval_runtime": 16.9261,
+      "eval_samples_per_second": 2.6,
+      "eval_steps_per_second": 0.177,
+      "step": 3695
+    },
+    {
+      "epoch": 0.9520558972212384,
+      "grad_norm": 0.15234151942918903,
+      "learning_rate": 6.979999903908297e-07,
+      "loss": 0.2756,
+      "step": 3696
+    },
+    {
+      "epoch": 0.9523134881025211,
+      "grad_norm": 0.17011987219811123,
+      "learning_rate": 6.905320671666326e-07,
+      "loss": 0.3531,
+      "step": 3697
+    },
+    {
+      "epoch": 0.952571078983804,
+      "grad_norm": 0.20705912072551072,
+      "learning_rate": 6.831040310790959e-07,
+      "loss": 0.4757,
+      "step": 3698
+    },
+    {
+      "epoch": 0.9528286698650867,
+      "grad_norm": 0.12767561963277557,
+      "learning_rate": 6.757158881368686e-07,
+      "loss": 0.3342,
+      "step": 3699
+    },
+    {
+      "epoch": 0.9530862607463696,
+      "grad_norm": 0.17796177316584283,
+      "learning_rate": 6.683676443163311e-07,
+      "loss": 0.3029,
+      "step": 3700
+    },
+    {
+      "epoch": 0.9530862607463696,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8780487804878049,
+      "eval_PRM F1 AUC": 0.7380827658459927,
+      "eval_PRM F1 AUC (fixed)": 0.8622315348349922,
+      "eval_PRM F1 Neg": 0.5833333333333334,
+      "eval_PRM NPV": 0.56,
+      "eval_PRM Precision": 0.8888888888888888,
+      "eval_PRM Recall": 0.8674698795180723,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3525390625,
+      "eval_runtime": 17.2548,
+      "eval_samples_per_second": 2.55,
+      "eval_steps_per_second": 0.174,
+      "step": 3700
+    },
+    {
+      "epoch": 0.9533438516276523,
+      "grad_norm": 0.1310624988332114,
+      "learning_rate": 6.610593055615733e-07,
+      "loss": 0.3261,
+      "step": 3701
+    },
+    {
+      "epoch": 0.9536014425089352,
+      "grad_norm": 0.13505877689353882,
+      "learning_rate": 6.537908777844215e-07,
+      "loss": 0.3194,
+      "step": 3702
+    },
+    {
+      "epoch": 0.9538590333902179,
+      "grad_norm": 0.14249633091361077,
+      "learning_rate": 6.465623668644117e-07,
+      "loss": 0.3034,
+      "step": 3703
+    },
+    {
+      "epoch": 0.9541166242715008,
+      "grad_norm": 0.14268727301342632,
+      "learning_rate": 6.393737786487885e-07,
+      "loss": 0.3654,
+      "step": 3704
+    },
+    {
+      "epoch": 0.9543742151527836,
+      "grad_norm": 0.1564181616551385,
+      "learning_rate": 6.322251189525008e-07,
+      "loss": 0.3605,
+      "step": 3705
+    },
+    {
+      "epoch": 0.9543742151527836,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.891566265060241,
+      "eval_PRM F1 AUC": 0.7501309586170771,
+      "eval_PRM F1 AUC (fixed)": 0.8590885280251441,
+      "eval_PRM F1 Neg": 0.6086956521739131,
+      "eval_PRM NPV": 0.6086956521739131,
+      "eval_PRM Precision": 0.891566265060241,
+      "eval_PRM Recall": 0.891566265060241,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3514293432235718,
+      "eval_runtime": 17.4001,
+      "eval_samples_per_second": 2.529,
+      "eval_steps_per_second": 0.172,
+      "step": 3705
+    },
+    {
+      "epoch": 0.9546318060340664,
+      "grad_norm": 0.16752144840241906,
+      "learning_rate": 6.251163935582116e-07,
+      "loss": 0.3424,
+      "step": 3706
+    },
+    {
+      "epoch": 0.9548893969153492,
+      "grad_norm": 0.1731135089318462,
+      "learning_rate": 6.180476082162656e-07,
+      "loss": 0.3848,
+      "step": 3707
+    },
+    {
+      "epoch": 0.955146987796632,
+      "grad_norm": 0.15070275416486573,
+      "learning_rate": 6.110187686447e-07,
+      "loss": 0.3657,
+      "step": 3708
+    },
+    {
+      "epoch": 0.9554045786779148,
+      "grad_norm": 0.18480108877326346,
+      "learning_rate": 6.040298805292499e-07,
+      "loss": 0.3594,
+      "step": 3709
+    },
+    {
+      "epoch": 0.9556621695591976,
+      "grad_norm": 0.1451269599547431,
+      "learning_rate": 5.970809495233265e-07,
+      "loss": 0.3382,
+      "step": 3710
+    },
+    {
+      "epoch": 0.9556621695591976,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8588266107909901,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3533380627632141,
+      "eval_runtime": 17.1366,
+      "eval_samples_per_second": 2.568,
+      "eval_steps_per_second": 0.175,
+      "step": 3710
+    },
+    {
+      "epoch": 0.9559197604404804,
+      "grad_norm": 0.204747373910136,
+      "learning_rate": 5.901719812480166e-07,
+      "loss": 0.4478,
+      "step": 3711
+    },
+    {
+      "epoch": 0.9561773513217632,
+      "grad_norm": 0.12067191724353114,
+      "learning_rate": 5.833029812920831e-07,
+      "loss": 0.3148,
+      "step": 3712
+    },
+    {
+      "epoch": 0.956434942203046,
+      "grad_norm": 0.15870357656342368,
+      "learning_rate": 5.764739552119702e-07,
+      "loss": 0.4068,
+      "step": 3713
+    },
+    {
+      "epoch": 0.9566925330843288,
+      "grad_norm": 0.1973206592364747,
+      "learning_rate": 5.696849085317646e-07,
+      "loss": 0.3478,
+      "step": 3714
+    },
+    {
+      "epoch": 0.9569501239656116,
+      "grad_norm": 0.2018788475412071,
+      "learning_rate": 5.629358467432289e-07,
+      "loss": 0.4264,
+      "step": 3715
+    },
+    {
+      "epoch": 0.9569501239656116,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8780487804878049,
+      "eval_PRM F1 AUC": 0.7380827658459927,
+      "eval_PRM F1 AUC (fixed)": 0.86013619696176,
+      "eval_PRM F1 Neg": 0.5833333333333334,
+      "eval_PRM NPV": 0.56,
+      "eval_PRM Precision": 0.8888888888888888,
+      "eval_PRM Recall": 0.8674698795180723,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3541370630264282,
+      "eval_runtime": 17.3697,
+      "eval_samples_per_second": 2.533,
+      "eval_steps_per_second": 0.173,
+      "step": 3715
+    },
+    {
+      "epoch": 0.9572077148468944,
+      "grad_norm": 0.15152106035513985,
+      "learning_rate": 5.562267753057626e-07,
+      "loss": 0.3867,
+      "step": 3716
+    },
+    {
+      "epoch": 0.9574653057281772,
+      "grad_norm": 0.13845243247216737,
+      "learning_rate": 5.495576996464468e-07,
+      "loss": 0.3161,
+      "step": 3717
+    },
+    {
+      "epoch": 0.95772289660946,
+      "grad_norm": 0.13210334465496898,
+      "learning_rate": 5.429286251599885e-07,
+      "loss": 0.3148,
+      "step": 3718
+    },
+    {
+      "epoch": 0.9579804874907428,
+      "grad_norm": 0.1571066170842066,
+      "learning_rate": 5.363395572087371e-07,
+      "loss": 0.3541,
+      "step": 3719
+    },
+    {
+      "epoch": 0.9582380783720257,
+      "grad_norm": 0.15295813994550886,
+      "learning_rate": 5.297905011226845e-07,
+      "loss": 0.4044,
+      "step": 3720
+    },
+    {
+      "epoch": 0.9582380783720257,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8780487804878049,
+      "eval_PRM F1 AUC": 0.7380827658459927,
+      "eval_PRM F1 AUC (fixed)": 0.86013619696176,
+      "eval_PRM F1 Neg": 0.5833333333333334,
+      "eval_PRM NPV": 0.56,
+      "eval_PRM Precision": 0.8888888888888888,
+      "eval_PRM Recall": 0.8674698795180723,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3545365631580353,
+      "eval_runtime": 17.1569,
+      "eval_samples_per_second": 2.565,
+      "eval_steps_per_second": 0.175,
+      "step": 3720
+    },
+    {
+      "epoch": 0.9584956692533084,
+      "grad_norm": 0.19608383974821886,
+      "learning_rate": 5.232814621994598e-07,
+      "loss": 0.4614,
+      "step": 3721
+    },
+    {
+      "epoch": 0.9587532601345913,
+      "grad_norm": 0.16729845368017604,
+      "learning_rate": 5.168124457043178e-07,
+      "loss": 0.3416,
+      "step": 3722
+    },
+    {
+      "epoch": 0.959010851015874,
+      "grad_norm": 0.1385253217089452,
+      "learning_rate": 5.103834568701338e-07,
+      "loss": 0.3307,
+      "step": 3723
+    },
+    {
+      "epoch": 0.9592684418971569,
+      "grad_norm": 0.13269673976828575,
+      "learning_rate": 5.039945008974145e-07,
+      "loss": 0.3276,
+      "step": 3724
+    },
+    {
+      "epoch": 0.9595260327784396,
+      "grad_norm": 0.1505006388364205,
+      "learning_rate": 4.976455829542704e-07,
+      "loss": 0.3554,
+      "step": 3725
+    },
+    {
+      "epoch": 0.9595260327784396,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8606600314300681,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3522283434867859,
+      "eval_runtime": 17.3805,
+      "eval_samples_per_second": 2.532,
+      "eval_steps_per_second": 0.173,
+      "step": 3725
+    },
+    {
+      "epoch": 0.9597836236597225,
+      "grad_norm": 0.15422220137904377,
+      "learning_rate": 4.91336708176443e-07,
+      "loss": 0.307,
+      "step": 3726
+    },
+    {
+      "epoch": 0.9600412145410052,
+      "grad_norm": 0.1948051781206229,
+      "learning_rate": 4.850678816672671e-07,
+      "loss": 0.3772,
+      "step": 3727
+    },
+    {
+      "epoch": 0.9602988054222881,
+      "grad_norm": 0.11942577613308271,
+      "learning_rate": 4.788391084976862e-07,
+      "loss": 0.3148,
+      "step": 3728
+    },
+    {
+      "epoch": 0.9605563963035708,
+      "grad_norm": 0.16581696493096246,
+      "learning_rate": 4.726503937062421e-07,
+      "loss": 0.3584,
+      "step": 3729
+    },
+    {
+      "epoch": 0.9608139871848537,
+      "grad_norm": 0.18270653336601703,
+      "learning_rate": 4.6650174229908026e-07,
+      "loss": 0.3931,
+      "step": 3730
+    },
+    {
+      "epoch": 0.9608139871848537,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.859350445259298,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35400390625,
+      "eval_runtime": 17.6109,
+      "eval_samples_per_second": 2.498,
+      "eval_steps_per_second": 0.17,
+      "step": 3730
+    },
+    {
+      "epoch": 0.9610715780661364,
+      "grad_norm": 0.1656800706537147,
+      "learning_rate": 4.6039315924992774e-07,
+      "loss": 0.337,
+      "step": 3731
+    },
+    {
+      "epoch": 0.9613291689474193,
+      "grad_norm": 0.15687605108049824,
+      "learning_rate": 4.543246495001097e-07,
+      "loss": 0.3087,
+      "step": 3732
+    },
+    {
+      "epoch": 0.961586759828702,
+      "grad_norm": 0.41979890458747227,
+      "learning_rate": 4.4829621795852705e-07,
+      "loss": 0.4024,
+      "step": 3733
+    },
+    {
+      "epoch": 0.9618443507099849,
+      "grad_norm": 0.18933485246564027,
+      "learning_rate": 4.42307869501668e-07,
+      "loss": 0.4332,
+      "step": 3734
+    },
+    {
+      "epoch": 0.9621019415912677,
+      "grad_norm": 0.1369765834241001,
+      "learning_rate": 4.363596089735911e-07,
+      "loss": 0.3336,
+      "step": 3735
+    },
+    {
+      "epoch": 0.9621019415912677,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.858040859088528,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35520240664482117,
+      "eval_runtime": 16.9809,
+      "eval_samples_per_second": 2.591,
+      "eval_steps_per_second": 0.177,
+      "step": 3735
+    },
+    {
+      "epoch": 0.9623595324725505,
+      "grad_norm": 0.1743975417300076,
+      "learning_rate": 4.304514411859306e-07,
+      "loss": 0.4265,
+      "step": 3736
+    },
+    {
+      "epoch": 0.9626171233538333,
+      "grad_norm": 0.18163224348449938,
+      "learning_rate": 4.2458337091788593e-07,
+      "loss": 0.4083,
+      "step": 3737
+    },
+    {
+      "epoch": 0.9628747142351161,
+      "grad_norm": 0.15285926097764635,
+      "learning_rate": 4.1875540291622106e-07,
+      "loss": 0.3955,
+      "step": 3738
+    },
+    {
+      "epoch": 0.9631323051163989,
+      "grad_norm": 0.16429583098588607,
+      "learning_rate": 4.129675418952761e-07,
+      "loss": 0.3556,
+      "step": 3739
+    },
+    {
+      "epoch": 0.9633898959976817,
+      "grad_norm": 0.22238495157892732,
+      "learning_rate": 4.0721979253692254e-07,
+      "loss": 0.4598,
+      "step": 3740
+    },
+    {
+      "epoch": 0.9633898959976817,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8780487804878049,
+      "eval_PRM F1 AUC": 0.7380827658459927,
+      "eval_PRM F1 AUC (fixed)": 0.8588266107909901,
+      "eval_PRM F1 Neg": 0.5833333333333334,
+      "eval_PRM NPV": 0.56,
+      "eval_PRM Precision": 0.8888888888888888,
+      "eval_PRM Recall": 0.8674698795180723,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35422584414482117,
+      "eval_runtime": 16.9862,
+      "eval_samples_per_second": 2.59,
+      "eval_steps_per_second": 0.177,
+      "step": 3740
+    },
+    {
+      "epoch": 0.9636474868789645,
+      "grad_norm": 0.17674656977462916,
+      "learning_rate": 4.015121594906024e-07,
+      "loss": 0.3965,
+      "step": 3741
+    },
+    {
+      "epoch": 0.9639050777602473,
+      "grad_norm": 0.16840117530400994,
+      "learning_rate": 3.958446473733002e-07,
+      "loss": 0.4132,
+      "step": 3742
+    },
+    {
+      "epoch": 0.9641626686415301,
+      "grad_norm": 0.13867074302418367,
+      "learning_rate": 3.9021726076954867e-07,
+      "loss": 0.3525,
+      "step": 3743
+    },
+    {
+      "epoch": 0.9644202595228129,
+      "grad_norm": 0.1459626178307732,
+      "learning_rate": 3.8463000423142326e-07,
+      "loss": 0.3487,
+      "step": 3744
+    },
+    {
+      "epoch": 0.9646778504040957,
+      "grad_norm": 0.15807260960562433,
+      "learning_rate": 3.7908288227853085e-07,
+      "loss": 0.3266,
+      "step": 3745
+    },
+    {
+      "epoch": 0.9646778504040957,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.891566265060241,
+      "eval_PRM F1 AUC": 0.7501309586170771,
+      "eval_PRM F1 AUC (fixed)": 0.859088528025144,
+      "eval_PRM F1 Neg": 0.6086956521739131,
+      "eval_PRM NPV": 0.6086956521739131,
+      "eval_PRM Precision": 0.891566265060241,
+      "eval_PRM Recall": 0.891566265060241,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35506924986839294,
+      "eval_runtime": 17.1534,
+      "eval_samples_per_second": 2.565,
+      "eval_steps_per_second": 0.175,
+      "step": 3745
+    },
+    {
+      "epoch": 0.9649354412853784,
+      "grad_norm": 0.1452604989808499,
+      "learning_rate": 3.73575899398021e-07,
+      "loss": 0.3457,
+      "step": 3746
+    },
+    {
+      "epoch": 0.9651930321666613,
+      "grad_norm": 0.19165309239817846,
+      "learning_rate": 3.681090600445747e-07,
+      "loss": 0.4019,
+      "step": 3747
+    },
+    {
+      "epoch": 0.9654506230479442,
+      "grad_norm": 0.1471576365965977,
+      "learning_rate": 3.626823686403935e-07,
+      "loss": 0.3234,
+      "step": 3748
+    },
+    {
+      "epoch": 0.9657082139292269,
+      "grad_norm": 0.18331570304808137,
+      "learning_rate": 3.572958295752049e-07,
+      "loss": 0.4197,
+      "step": 3749
+    },
+    {
+      "epoch": 0.9659658048105098,
+      "grad_norm": 0.1497592311083479,
+      "learning_rate": 3.519494472062568e-07,
+      "loss": 0.2976,
+      "step": 3750
+    },
+    {
+      "epoch": 0.9659658048105098,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8624934520691461,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3533824682235718,
+      "eval_runtime": 17.4882,
+      "eval_samples_per_second": 2.516,
+      "eval_steps_per_second": 0.172,
+      "step": 3750
+    },
+    {
+      "epoch": 0.9662233956917925,
+      "grad_norm": 0.18542590579616175,
+      "learning_rate": 3.466432258583174e-07,
+      "loss": 0.2955,
+      "step": 3751
+    },
+    {
+      "epoch": 0.9664809865730754,
+      "grad_norm": 0.14196607964512953,
+      "learning_rate": 3.4137716982366453e-07,
+      "loss": 0.2848,
+      "step": 3752
+    },
+    {
+      "epoch": 0.9667385774543581,
+      "grad_norm": 0.16210220602760117,
+      "learning_rate": 3.3615128336209054e-07,
+      "loss": 0.4347,
+      "step": 3753
+    },
+    {
+      "epoch": 0.966996168335641,
+      "grad_norm": 0.12357408535083989,
+      "learning_rate": 3.309655707008863e-07,
+      "loss": 0.3585,
+      "step": 3754
+    },
+    {
+      "epoch": 0.9672537592169237,
+      "grad_norm": 0.13618250690592645,
+      "learning_rate": 3.258200360348462e-07,
+      "loss": 0.3281,
+      "step": 3755
+    },
+    {
+      "epoch": 0.9672537592169237,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8588266107909899,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35360440611839294,
+      "eval_runtime": 17.0303,
+      "eval_samples_per_second": 2.584,
+      "eval_steps_per_second": 0.176,
+      "step": 3755
+    },
+    {
+      "epoch": 0.9675113500982065,
+      "grad_norm": 0.13654383235663448,
+      "learning_rate": 3.207146835262742e-07,
+      "loss": 0.3063,
+      "step": 3756
+    },
+    {
+      "epoch": 0.9677689409794893,
+      "grad_norm": 0.15999728729569007,
+      "learning_rate": 3.1564951730495006e-07,
+      "loss": 0.3554,
+      "step": 3757
+    },
+    {
+      "epoch": 0.9680265318607721,
+      "grad_norm": 0.16211117805780945,
+      "learning_rate": 3.106245414681741e-07,
+      "loss": 0.3693,
+      "step": 3758
+    },
+    {
+      "epoch": 0.9682841227420549,
+      "grad_norm": 0.21034146330880885,
+      "learning_rate": 3.0563976008071706e-07,
+      "loss": 0.4134,
+      "step": 3759
+    },
+    {
+      "epoch": 0.9685417136233377,
+      "grad_norm": 0.14652568706774757,
+      "learning_rate": 3.006951771748423e-07,
+      "loss": 0.3347,
+      "step": 3760
+    },
+    {
+      "epoch": 0.9685417136233377,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.857778941854374,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3546697497367859,
+      "eval_runtime": 16.9428,
+      "eval_samples_per_second": 2.597,
+      "eval_steps_per_second": 0.177,
+      "step": 3760
+    },
+    {
+      "epoch": 0.9687993045046205,
+      "grad_norm": 0.14316930791436056,
+      "learning_rate": 2.9579079675028374e-07,
+      "loss": 0.3233,
+      "step": 3761
+    },
+    {
+      "epoch": 0.9690568953859033,
+      "grad_norm": 0.2000551747105647,
+      "learning_rate": 2.9092662277427906e-07,
+      "loss": 0.4459,
+      "step": 3762
+    },
+    {
+      "epoch": 0.9693144862671862,
+      "grad_norm": 0.165468342566276,
+      "learning_rate": 2.8610265918151414e-07,
+      "loss": 0.3376,
+      "step": 3763
+    },
+    {
+      "epoch": 0.9695720771484689,
+      "grad_norm": 0.11885486319727619,
+      "learning_rate": 2.813189098741731e-07,
+      "loss": 0.2936,
+      "step": 3764
+    },
+    {
+      "epoch": 0.9698296680297518,
+      "grad_norm": 0.18333947791187652,
+      "learning_rate": 2.7657537872189386e-07,
+      "loss": 0.4077,
+      "step": 3765
+    },
+    {
+      "epoch": 0.9698296680297518,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8601361969617601,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35245028138160706,
+      "eval_runtime": 17.0879,
+      "eval_samples_per_second": 2.575,
+      "eval_steps_per_second": 0.176,
+      "step": 3765
+    },
+    {
+      "epoch": 0.9700872589110345,
+      "grad_norm": 0.18535954788255296,
+      "learning_rate": 2.7187206956179047e-07,
+      "loss": 0.404,
+      "step": 3766
+    },
+    {
+      "epoch": 0.9703448497923174,
+      "grad_norm": 0.17033575330351897,
+      "learning_rate": 2.6720898619843615e-07,
+      "loss": 0.3333,
+      "step": 3767
+    },
+    {
+      "epoch": 0.9706024406736001,
+      "grad_norm": 0.1340335465058161,
+      "learning_rate": 2.625861324038692e-07,
+      "loss": 0.3344,
+      "step": 3768
+    },
+    {
+      "epoch": 0.970860031554883,
+      "grad_norm": 0.1539518992267126,
+      "learning_rate": 2.580035119175761e-07,
+      "loss": 0.3689,
+      "step": 3769
+    },
+    {
+      "epoch": 0.9711176224361657,
+      "grad_norm": 0.22640026655678336,
+      "learning_rate": 2.534611284465083e-07,
+      "loss": 0.404,
+      "step": 3770
+    },
+    {
+      "epoch": 0.9711176224361657,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8601361969617599,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35391512513160706,
+      "eval_runtime": 17.1355,
+      "eval_samples_per_second": 2.568,
+      "eval_steps_per_second": 0.175,
+      "step": 3770
+    },
+    {
+      "epoch": 0.9713752133174486,
+      "grad_norm": 0.1467583746360964,
+      "learning_rate": 2.4895898566505983e-07,
+      "loss": 0.3658,
+      "step": 3771
+    },
+    {
+      "epoch": 0.9716328041987313,
+      "grad_norm": 0.15289702694249147,
+      "learning_rate": 2.4449708721508424e-07,
+      "loss": 0.3454,
+      "step": 3772
+    },
+    {
+      "epoch": 0.9718903950800142,
+      "grad_norm": 0.16047779634622722,
+      "learning_rate": 2.4007543670587774e-07,
+      "loss": 0.4495,
+      "step": 3773
+    },
+    {
+      "epoch": 0.9721479859612969,
+      "grad_norm": 0.1665765188938278,
+      "learning_rate": 2.356940377141681e-07,
+      "loss": 0.4329,
+      "step": 3774
+    },
+    {
+      "epoch": 0.9724055768425798,
+      "grad_norm": 0.1179113612466018,
+      "learning_rate": 2.3135289378413694e-07,
+      "loss": 0.3069,
+      "step": 3775
+    },
+    {
+      "epoch": 0.9724055768425798,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8569931901519119,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3548029065132141,
+      "eval_runtime": 17.5981,
+      "eval_samples_per_second": 2.5,
+      "eval_steps_per_second": 0.17,
+      "step": 3775
+    },
+    {
+      "epoch": 0.9726631677238625,
+      "grad_norm": 0.16314819661869495,
+      "learning_rate": 2.2705200842740305e-07,
+      "loss": 0.2938,
+      "step": 3776
+    },
+    {
+      "epoch": 0.9729207586051454,
+      "grad_norm": 0.20572881699364856,
+      "learning_rate": 2.2279138512300567e-07,
+      "loss": 0.3926,
+      "step": 3777
+    },
+    {
+      "epoch": 0.9731783494864282,
+      "grad_norm": 0.16351923007215768,
+      "learning_rate": 2.1857102731742684e-07,
+      "loss": 0.309,
+      "step": 3778
+    },
+    {
+      "epoch": 0.973435940367711,
+      "grad_norm": 0.15430164950532926,
+      "learning_rate": 2.143909384245746e-07,
+      "loss": 0.34,
+      "step": 3779
+    },
+    {
+      "epoch": 0.9736935312489938,
+      "grad_norm": 0.17881733482234016,
+      "learning_rate": 2.10251121825783e-07,
+      "loss": 0.3813,
+      "step": 3780
+    },
+    {
+      "epoch": 0.9736935312489938,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.857778941854374,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3528497815132141,
+      "eval_runtime": 16.9415,
+      "eval_samples_per_second": 2.597,
+      "eval_steps_per_second": 0.177,
+      "step": 3780
+    },
+    {
+      "epoch": 0.9739511221302766,
+      "grad_norm": 0.1759713240969221,
+      "learning_rate": 2.0615158086981222e-07,
+      "loss": 0.4305,
+      "step": 3781
+    },
+    {
+      "epoch": 0.9742087130115594,
+      "grad_norm": 0.15528693139533326,
+      "learning_rate": 2.020923188728374e-07,
+      "loss": 0.3215,
+      "step": 3782
+    },
+    {
+      "epoch": 0.9744663038928422,
+      "grad_norm": 0.1394952976316927,
+      "learning_rate": 1.980733391184486e-07,
+      "loss": 0.335,
+      "step": 3783
+    },
+    {
+      "epoch": 0.974723894774125,
+      "grad_norm": 0.16330749731661878,
+      "learning_rate": 1.940946448576675e-07,
+      "loss": 0.3388,
+      "step": 3784
+    },
+    {
+      "epoch": 0.9749814856554078,
+      "grad_norm": 0.13626155155802247,
+      "learning_rate": 1.9015623930890846e-07,
+      "loss": 0.3177,
+      "step": 3785
+    },
+    {
+      "epoch": 0.9749814856554078,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8617077003666841,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35196200013160706,
+      "eval_runtime": 16.6163,
+      "eval_samples_per_second": 2.648,
+      "eval_steps_per_second": 0.181,
+      "step": 3785
+    },
+    {
+      "epoch": 0.9752390765366906,
+      "grad_norm": 0.16720574020751214,
+      "learning_rate": 1.8625812565800648e-07,
+      "loss": 0.4074,
+      "step": 3786
+    },
+    {
+      "epoch": 0.9754966674179734,
+      "grad_norm": 0.16792236698504823,
+      "learning_rate": 1.8240030705820032e-07,
+      "loss": 0.3667,
+      "step": 3787
+    },
+    {
+      "epoch": 0.9757542582992562,
+      "grad_norm": 0.1932284534720809,
+      "learning_rate": 1.7858278663013817e-07,
+      "loss": 0.3507,
+      "step": 3788
+    },
+    {
+      "epoch": 0.976011849180539,
+      "grad_norm": 0.15949009758829952,
+      "learning_rate": 1.748055674618665e-07,
+      "loss": 0.2286,
+      "step": 3789
+    },
+    {
+      "epoch": 0.9762694400618218,
+      "grad_norm": 0.17407811866438883,
+      "learning_rate": 1.710686526088301e-07,
+      "loss": 0.3552,
+      "step": 3790
+    },
+    {
+      "epoch": 0.9762694400618218,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8780487804878049,
+      "eval_PRM F1 AUC": 0.7380827658459927,
+      "eval_PRM F1 AUC (fixed)": 0.8609219486642221,
+      "eval_PRM F1 Neg": 0.5833333333333334,
+      "eval_PRM NPV": 0.56,
+      "eval_PRM Precision": 0.8888888888888888,
+      "eval_PRM Recall": 0.8674698795180723,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35324928164482117,
+      "eval_runtime": 17.4588,
+      "eval_samples_per_second": 2.52,
+      "eval_steps_per_second": 0.172,
+      "step": 3790
+    },
+    {
+      "epoch": 0.9765270309431047,
+      "grad_norm": 0.18141142441348804,
+      "learning_rate": 1.6737204509387206e-07,
+      "loss": 0.4585,
+      "step": 3791
+    },
+    {
+      "epoch": 0.9767846218243874,
+      "grad_norm": 0.17238267023227896,
+      "learning_rate": 1.6371574790723377e-07,
+      "loss": 0.3291,
+      "step": 3792
+    },
+    {
+      "epoch": 0.9770422127056703,
+      "grad_norm": 0.18083413404662643,
+      "learning_rate": 1.6009976400654937e-07,
+      "loss": 0.3972,
+      "step": 3793
+    },
+    {
+      "epoch": 0.977299803586953,
+      "grad_norm": 0.1659465500339923,
+      "learning_rate": 1.5652409631682908e-07,
+      "loss": 0.3672,
+      "step": 3794
+    },
+    {
+      "epoch": 0.9775573944682359,
+      "grad_norm": 0.13994735128114089,
+      "learning_rate": 1.529887477304981e-07,
+      "loss": 0.4038,
+      "step": 3795
+    },
+    {
+      "epoch": 0.9775573944682359,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8795180722891566,
+      "eval_PRM F1 AUC": 0.7223677317967523,
+      "eval_PRM F1 AUC (fixed)": 0.8572551073860659,
+      "eval_PRM F1 Neg": 0.5652173913043478,
+      "eval_PRM NPV": 0.5652173913043478,
+      "eval_PRM Precision": 0.8795180722891566,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.5652173913043478,
+      "eval_loss": 0.3546253442764282,
+      "eval_runtime": 16.6268,
+      "eval_samples_per_second": 2.646,
+      "eval_steps_per_second": 0.18,
+      "step": 3795
+    },
+    {
+      "epoch": 0.9778149853495186,
+      "grad_norm": 0.14847747360892083,
+      "learning_rate": 1.494937211073355e-07,
+      "loss": 0.3311,
+      "step": 3796
+    },
+    {
+      "epoch": 0.9780725762308015,
+      "grad_norm": 0.14264566608027543,
+      "learning_rate": 1.4603901927452978e-07,
+      "loss": 0.3381,
+      "step": 3797
+    },
+    {
+      "epoch": 0.9783301671120842,
+      "grad_norm": 0.13899999288446177,
+      "learning_rate": 1.4262464502663443e-07,
+      "loss": 0.3626,
+      "step": 3798
+    },
+    {
+      "epoch": 0.9785877579933671,
+      "grad_norm": 0.13291754858024754,
+      "learning_rate": 1.3925060112557898e-07,
+      "loss": 0.3594,
+      "step": 3799
+    },
+    {
+      "epoch": 0.9788453488746498,
+      "grad_norm": 0.15571500003310815,
+      "learning_rate": 1.3591689030068577e-07,
+      "loss": 0.3405,
+      "step": 3800
+    },
+    {
+      "epoch": 0.9788453488746498,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8795180722891566,
+      "eval_PRM F1 AUC": 0.7223677317967523,
+      "eval_PRM F1 AUC (fixed)": 0.8590885280251439,
+      "eval_PRM F1 Neg": 0.5652173913043478,
+      "eval_PRM NPV": 0.5652173913043478,
+      "eval_PRM Precision": 0.8795180722891566,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.5652173913043478,
+      "eval_loss": 0.3543146252632141,
+      "eval_runtime": 17.097,
+      "eval_samples_per_second": 2.574,
+      "eval_steps_per_second": 0.175,
+      "step": 3800
+    },
+    {
+      "epoch": 0.9791029397559327,
+      "grad_norm": 0.18620757337239352,
+      "learning_rate": 1.3262351524864213e-07,
+      "loss": 0.4272,
+      "step": 3801
+    },
+    {
+      "epoch": 0.9793605306372154,
+      "grad_norm": 0.1756420655693183,
+      "learning_rate": 1.2937047863350037e-07,
+      "loss": 0.2808,
+      "step": 3802
+    },
+    {
+      "epoch": 0.9796181215184983,
+      "grad_norm": 0.14244491271889062,
+      "learning_rate": 1.2615778308668892e-07,
+      "loss": 0.2852,
+      "step": 3803
+    },
+    {
+      "epoch": 0.979875712399781,
+      "grad_norm": 0.1516494617807825,
+      "learning_rate": 1.2298543120700113e-07,
+      "loss": 0.3433,
+      "step": 3804
+    },
+    {
+      "epoch": 0.9801333032810639,
+      "grad_norm": 0.14925646654192018,
+      "learning_rate": 1.1985342556060652e-07,
+      "loss": 0.2899,
+      "step": 3805
+    },
+    {
+      "epoch": 0.9801333032810639,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8609219486642221,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3538263440132141,
+      "eval_runtime": 17.8024,
+      "eval_samples_per_second": 2.472,
+      "eval_steps_per_second": 0.169,
+      "step": 3805
+    },
+    {
+      "epoch": 0.9803908941623467,
+      "grad_norm": 0.14432809565504431,
+      "learning_rate": 1.1676176868102295e-07,
+      "loss": 0.3247,
+      "step": 3806
+    },
+    {
+      "epoch": 0.9806484850436294,
+      "grad_norm": 0.17548929482352033,
+      "learning_rate": 1.1371046306914435e-07,
+      "loss": 0.3986,
+      "step": 3807
+    },
+    {
+      "epoch": 0.9809060759249123,
+      "grad_norm": 0.13571487128816537,
+      "learning_rate": 1.1069951119320188e-07,
+      "loss": 0.3344,
+      "step": 3808
+    },
+    {
+      "epoch": 0.981163666806195,
+      "grad_norm": 0.14559430886412886,
+      "learning_rate": 1.0772891548880282e-07,
+      "loss": 0.3374,
+      "step": 3809
+    },
+    {
+      "epoch": 0.9814212576874779,
+      "grad_norm": 0.15790207391145736,
+      "learning_rate": 1.0479867835891389e-07,
+      "loss": 0.3819,
+      "step": 3810
+    },
+    {
+      "epoch": 0.9814212576874779,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.85620743844945,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3541814684867859,
+      "eval_runtime": 17.5142,
+      "eval_samples_per_second": 2.512,
+      "eval_steps_per_second": 0.171,
+      "step": 3810
+    },
+    {
+      "epoch": 0.9816788485687606,
+      "grad_norm": 0.14105410419460254,
+      "learning_rate": 1.0190880217383347e-07,
+      "loss": 0.3418,
+      "step": 3811
+    },
+    {
+      "epoch": 0.9819364394500435,
+      "grad_norm": 0.16032004791537616,
+      "learning_rate": 9.905928927123609e-08,
+      "loss": 0.2992,
+      "step": 3812
+    },
+    {
+      "epoch": 0.9821940303313262,
+      "grad_norm": 0.17342208377038257,
+      "learning_rate": 9.625014195612236e-08,
+      "loss": 0.3601,
+      "step": 3813
+    },
+    {
+      "epoch": 0.9824516212126091,
+      "grad_norm": 0.13949111908780418,
+      "learning_rate": 9.348136250085793e-08,
+      "loss": 0.3036,
+      "step": 3814
+    },
+    {
+      "epoch": 0.9827092120938918,
+      "grad_norm": 0.1604200708844017,
+      "learning_rate": 9.075295314514565e-08,
+      "loss": 0.3832,
+      "step": 3815
+    },
+    {
+      "epoch": 0.9827092120938918,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.859350445259298,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3556019067764282,
+      "eval_runtime": 17.1716,
+      "eval_samples_per_second": 2.562,
+      "eval_steps_per_second": 0.175,
+      "step": 3815
+    },
+    {
+      "epoch": 0.9829668029751747,
+      "grad_norm": 0.15596615102074293,
+      "learning_rate": 8.806491609603673e-08,
+      "loss": 0.3538,
+      "step": 3816
+    },
+    {
+      "epoch": 0.9832243938564574,
+      "grad_norm": 0.1798691547838207,
+      "learning_rate": 8.541725352791963e-08,
+      "loss": 0.3654,
+      "step": 3817
+    },
+    {
+      "epoch": 0.9834819847377403,
+      "grad_norm": 0.16031802102742165,
+      "learning_rate": 8.280996758253112e-08,
+      "loss": 0.363,
+      "step": 3818
+    },
+    {
+      "epoch": 0.983739575619023,
+      "grad_norm": 0.13947517080899272,
+      "learning_rate": 8.02430603689397e-08,
+      "loss": 0.315,
+      "step": 3819
+    },
+    {
+      "epoch": 0.9839971665003059,
+      "grad_norm": 0.16446868867314984,
+      "learning_rate": 7.77165339635566e-08,
+      "loss": 0.394,
+      "step": 3820
+    },
+    {
+      "epoch": 0.9839971665003059,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8619696176008381,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35427024960517883,
+      "eval_runtime": 17.0867,
+      "eval_samples_per_second": 2.575,
+      "eval_steps_per_second": 0.176,
+      "step": 3820
+    },
+    {
+      "epoch": 0.9842547573815887,
+      "grad_norm": 0.1385861107711507,
+      "learning_rate": 7.523039041012481e-08,
+      "loss": 0.3465,
+      "step": 3821
+    },
+    {
+      "epoch": 0.9845123482628715,
+      "grad_norm": 0.12561988471237037,
+      "learning_rate": 7.27846317197134e-08,
+      "loss": 0.3569,
+      "step": 3822
+    },
+    {
+      "epoch": 0.9847699391441543,
+      "grad_norm": 0.19075340599019874,
+      "learning_rate": 7.037925987074534e-08,
+      "loss": 0.324,
+      "step": 3823
+    },
+    {
+      "epoch": 0.9850275300254371,
+      "grad_norm": 0.1658793046587931,
+      "learning_rate": 6.801427680894756e-08,
+      "loss": 0.3466,
+      "step": 3824
+    },
+    {
+      "epoch": 0.9852851209067199,
+      "grad_norm": 0.1456440297940814,
+      "learning_rate": 6.568968444739532e-08,
+      "loss": 0.4245,
+      "step": 3825
+    },
+    {
+      "epoch": 0.9852851209067199,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8862275449101796,
+      "eval_PRM F1 AUC": 0.7283918281822945,
+      "eval_PRM F1 AUC (fixed)": 0.8588266107909899,
+      "eval_PRM F1 Neg": 0.5777777777777777,
+      "eval_PRM NPV": 0.5909090909090909,
+      "eval_PRM Precision": 0.8809523809523809,
+      "eval_PRM Recall": 0.891566265060241,
+      "eval_PRM Specificty": 0.5652173913043478,
+      "eval_loss": 0.35573509335517883,
+      "eval_runtime": 17.1337,
+      "eval_samples_per_second": 2.568,
+      "eval_steps_per_second": 0.175,
+      "step": 3825
+    },
+    {
+      "epoch": 0.9855427117880027,
+      "grad_norm": 0.13283828143284623,
+      "learning_rate": 6.340548466648443e-08,
+      "loss": 0.2988,
+      "step": 3826
+    },
+    {
+      "epoch": 0.9858003026692855,
+      "grad_norm": 0.1839927084008868,
+      "learning_rate": 6.116167931393691e-08,
+      "loss": 0.3724,
+      "step": 3827
+    },
+    {
+      "epoch": 0.9860578935505683,
+      "grad_norm": 0.14082341113721372,
+      "learning_rate": 5.895827020479527e-08,
+      "loss": 0.276,
+      "step": 3828
+    },
+    {
+      "epoch": 0.9863154844318511,
+      "grad_norm": 0.14357908740989736,
+      "learning_rate": 5.6795259121439345e-08,
+      "loss": 0.2945,
+      "step": 3829
+    },
+    {
+      "epoch": 0.9865730753131339,
+      "grad_norm": 0.15451102193521904,
+      "learning_rate": 5.4672647813547305e-08,
+      "loss": 0.3223,
+      "step": 3830
+    },
+    {
+      "epoch": 0.9865730753131339,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8551597695128339,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3560901880264282,
+      "eval_runtime": 16.8872,
+      "eval_samples_per_second": 2.606,
+      "eval_steps_per_second": 0.178,
+      "step": 3830
+    },
+    {
+      "epoch": 0.9868306661944167,
+      "grad_norm": 0.17445890619553567,
+      "learning_rate": 5.2590437998134566e-08,
+      "loss": 0.4237,
+      "step": 3831
+    },
+    {
+      "epoch": 0.9870882570756995,
+      "grad_norm": 0.1255277707156797,
+      "learning_rate": 5.054863135953158e-08,
+      "loss": 0.3201,
+      "step": 3832
+    },
+    {
+      "epoch": 0.9873458479569823,
+      "grad_norm": 0.1420320075937339,
+      "learning_rate": 4.8547229549383844e-08,
+      "loss": 0.4258,
+      "step": 3833
+    },
+    {
+      "epoch": 0.9876034388382652,
+      "grad_norm": 0.16849844384390825,
+      "learning_rate": 4.658623418665742e-08,
+      "loss": 0.3963,
+      "step": 3834
+    },
+    {
+      "epoch": 0.9878610297195479,
+      "grad_norm": 0.13801399523631844,
+      "learning_rate": 4.4665646857627864e-08,
+      "loss": 0.3911,
+      "step": 3835
+    },
+    {
+      "epoch": 0.9878610297195479,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.891566265060241,
+      "eval_PRM F1 AUC": 0.7501309586170771,
+      "eval_PRM F1 AUC (fixed)": 0.8619696176008381,
+      "eval_PRM F1 Neg": 0.6086956521739131,
+      "eval_PRM NPV": 0.6086956521739131,
+      "eval_PRM Precision": 0.891566265060241,
+      "eval_PRM Recall": 0.891566265060241,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35400390625,
+      "eval_runtime": 17.5443,
+      "eval_samples_per_second": 2.508,
+      "eval_steps_per_second": 0.171,
+      "step": 3835
+    },
+    {
+      "epoch": 0.9881186206008308,
+      "grad_norm": 0.14497941822115065,
+      "learning_rate": 4.278546911588022e-08,
+      "loss": 0.337,
+      "step": 3836
+    },
+    {
+      "epoch": 0.9883762114821135,
+      "grad_norm": 0.17940620321670397,
+      "learning_rate": 4.094570248232565e-08,
+      "loss": 0.3965,
+      "step": 3837
+    },
+    {
+      "epoch": 0.9886338023633964,
+      "grad_norm": 0.13987601268318273,
+      "learning_rate": 3.9146348445173733e-08,
+      "loss": 0.3451,
+      "step": 3838
+    },
+    {
+      "epoch": 0.9888913932446791,
+      "grad_norm": 0.16282875349357964,
+      "learning_rate": 3.7387408459949035e-08,
+      "loss": 0.3761,
+      "step": 3839
+    },
+    {
+      "epoch": 0.989148984125962,
+      "grad_norm": 0.15513189177962927,
+      "learning_rate": 3.566888394948009e-08,
+      "loss": 0.3252,
+      "step": 3840
+    },
+    {
+      "epoch": 0.989148984125962,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8780487804878049,
+      "eval_PRM F1 AUC": 0.7380827658459927,
+      "eval_PRM F1 AUC (fixed)": 0.8624934520691462,
+      "eval_PRM F1 Neg": 0.5833333333333334,
+      "eval_PRM NPV": 0.56,
+      "eval_PRM Precision": 0.8888888888888888,
+      "eval_PRM Recall": 0.8674698795180723,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3528497815132141,
+      "eval_runtime": 16.8796,
+      "eval_samples_per_second": 2.607,
+      "eval_steps_per_second": 0.178,
+      "step": 3840
+    },
+    {
+      "epoch": 0.9894065750072447,
+      "grad_norm": 0.1551213359199816,
+      "learning_rate": 3.3990776303910456e-08,
+      "loss": 0.3912,
+      "step": 3841
+    },
+    {
+      "epoch": 0.9896641658885276,
+      "grad_norm": 0.15981647033788135,
+      "learning_rate": 3.235308688068206e-08,
+      "loss": 0.3977,
+      "step": 3842
+    },
+    {
+      "epoch": 0.9899217567698103,
+      "grad_norm": 0.16522435580180106,
+      "learning_rate": 3.075581700454633e-08,
+      "loss": 0.2958,
+      "step": 3843
+    },
+    {
+      "epoch": 0.9901793476510932,
+      "grad_norm": 0.23521488165578452,
+      "learning_rate": 2.919896796755861e-08,
+      "loss": 0.4202,
+      "step": 3844
+    },
+    {
+      "epoch": 0.9904369385323759,
+      "grad_norm": 0.1904301992468953,
+      "learning_rate": 2.768254102907819e-08,
+      "loss": 0.4,
+      "step": 3845
+    },
+    {
+      "epoch": 0.9904369385323759,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.858302776322682,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35178443789482117,
+      "eval_runtime": 17.4583,
+      "eval_samples_per_second": 2.52,
+      "eval_steps_per_second": 0.172,
+      "step": 3845
+    },
+    {
+      "epoch": 0.9906945294136588,
+      "grad_norm": 0.1464762142939478,
+      "learning_rate": 2.6206537415762733e-08,
+      "loss": 0.3524,
+      "step": 3846
+    },
+    {
+      "epoch": 0.9909521202949415,
+      "grad_norm": 0.16020884252536802,
+      "learning_rate": 2.4770958321568283e-08,
+      "loss": 0.4248,
+      "step": 3847
+    },
+    {
+      "epoch": 0.9912097111762244,
+      "grad_norm": 0.11633653027179804,
+      "learning_rate": 2.337580490776592e-08,
+      "loss": 0.246,
+      "step": 3848
+    },
+    {
+      "epoch": 0.9914673020575072,
+      "grad_norm": 0.14268793767622032,
+      "learning_rate": 2.2021078302902897e-08,
+      "loss": 0.339,
+      "step": 3849
+    },
+    {
+      "epoch": 0.99172489293879,
+      "grad_norm": 0.1373486811631178,
+      "learning_rate": 2.070677960284706e-08,
+      "loss": 0.3918,
+      "step": 3850
+    },
+    {
+      "epoch": 0.99172489293879,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.859874279727606,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35293856263160706,
+      "eval_runtime": 17.4584,
+      "eval_samples_per_second": 2.52,
+      "eval_steps_per_second": 0.172,
+      "step": 3850
+    },
+    {
+      "epoch": 0.9919824838200728,
+      "grad_norm": 0.12419707771509048,
+      "learning_rate": 1.9432909870747974e-08,
+      "loss": 0.3071,
+      "step": 3851
+    },
+    {
+      "epoch": 0.9922400747013556,
+      "grad_norm": 0.14396651708514002,
+      "learning_rate": 1.819947013705914e-08,
+      "loss": 0.3771,
+      "step": 3852
+    },
+    {
+      "epoch": 0.9924976655826384,
+      "grad_norm": 0.16279186864950249,
+      "learning_rate": 1.7006461399532436e-08,
+      "loss": 0.3324,
+      "step": 3853
+    },
+    {
+      "epoch": 0.9927552564639212,
+      "grad_norm": 0.14552583560564572,
+      "learning_rate": 1.5853884623195925e-08,
+      "loss": 0.306,
+      "step": 3854
+    },
+    {
+      "epoch": 0.993012847345204,
+      "grad_norm": 0.14857715616726758,
+      "learning_rate": 1.4741740740398246e-08,
+      "loss": 0.3462,
+      "step": 3855
+    },
+    {
+      "epoch": 0.993012847345204,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8780487804878049,
+      "eval_PRM F1 AUC": 0.7380827658459927,
+      "eval_PRM F1 AUC (fixed)": 0.8611838658983761,
+      "eval_PRM F1 Neg": 0.5833333333333334,
+      "eval_PRM NPV": 0.56,
+      "eval_PRM Precision": 0.8888888888888888,
+      "eval_PRM Recall": 0.8674698795180723,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35262784361839294,
+      "eval_runtime": 16.7981,
+      "eval_samples_per_second": 2.619,
+      "eval_steps_per_second": 0.179,
+      "step": 3855
+    },
+    {
+      "epoch": 0.9932704382264868,
+      "grad_norm": 0.11861395561824115,
+      "learning_rate": 1.367003065076422e-08,
+      "loss": 0.2956,
+      "step": 3856
+    },
+    {
+      "epoch": 0.9935280291077696,
+      "grad_norm": 0.18274494107078726,
+      "learning_rate": 1.2638755221217047e-08,
+      "loss": 0.3989,
+      "step": 3857
+    },
+    {
+      "epoch": 0.9937856199890523,
+      "grad_norm": 0.15911968143770888,
+      "learning_rate": 1.1647915285967204e-08,
+      "loss": 0.4393,
+      "step": 3858
+    },
+    {
+      "epoch": 0.9940432108703352,
+      "grad_norm": 0.19947833891807695,
+      "learning_rate": 1.069751164652355e-08,
+      "loss": 0.4774,
+      "step": 3859
+    },
+    {
+      "epoch": 0.994300801751618,
+      "grad_norm": 0.1792602435414376,
+      "learning_rate": 9.787545071676674e-09,
+      "loss": 0.4409,
+      "step": 3860
+    },
+    {
+      "epoch": 0.994300801751618,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8596123624934521,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3546697497367859,
+      "eval_runtime": 16.8421,
+      "eval_samples_per_second": 2.612,
+      "eval_steps_per_second": 0.178,
+      "step": 3860
+    },
+    {
+      "epoch": 0.9945583926329008,
+      "grad_norm": 0.16676315030772298,
+      "learning_rate": 8.918016297515541e-09,
+      "loss": 0.3873,
+      "step": 3861
+    },
+    {
+      "epoch": 0.9948159835141835,
+      "grad_norm": 0.1537904133890653,
+      "learning_rate": 8.088926027410848e-09,
+      "loss": 0.3587,
+      "step": 3862
+    },
+    {
+      "epoch": 0.9950735743954664,
+      "grad_norm": 0.14157889090639966,
+      "learning_rate": 7.30027493203167e-09,
+      "loss": 0.3607,
+      "step": 3863
+    },
+    {
+      "epoch": 0.9953311652767493,
+      "grad_norm": 0.19332578555591137,
+      "learning_rate": 6.552063649323259e-09,
+      "loss": 0.4088,
+      "step": 3864
+    },
+    {
+      "epoch": 0.995588756158032,
+      "grad_norm": 0.13129590984808737,
+      "learning_rate": 5.844292784523697e-09,
+      "loss": 0.3371,
+      "step": 3865
+    },
+    {
+      "epoch": 0.995588756158032,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.85751702462022,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35520240664482117,
+      "eval_runtime": 17.1042,
+      "eval_samples_per_second": 2.572,
+      "eval_steps_per_second": 0.175,
+      "step": 3865
+    },
+    {
+      "epoch": 0.9958463470393148,
+      "grad_norm": 0.1591974055243455,
+      "learning_rate": 5.176962910163896e-09,
+      "loss": 0.302,
+      "step": 3866
+    },
+    {
+      "epoch": 0.9961039379205976,
+      "grad_norm": 0.13232354053669065,
+      "learning_rate": 4.5500745660509435e-09,
+      "loss": 0.2751,
+      "step": 3867
+    },
+    {
+      "epoch": 0.9963615288018804,
+      "grad_norm": 0.19267965536010473,
+      "learning_rate": 3.963628259290308e-09,
+      "loss": 0.4389,
+      "step": 3868
+    },
+    {
+      "epoch": 0.9966191196831632,
+      "grad_norm": 0.22194226750555665,
+      "learning_rate": 3.4176244642636355e-09,
+      "loss": 0.4047,
+      "step": 3869
+    },
+    {
+      "epoch": 0.996876710564446,
+      "grad_norm": 0.1667614023688519,
+      "learning_rate": 2.912063622639849e-09,
+      "loss": 0.3986,
+      "step": 3870
+    },
+    {
+      "epoch": 0.996876710564446,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.891566265060241,
+      "eval_PRM F1 AUC": 0.7501309586170771,
+      "eval_PRM F1 AUC (fixed)": 0.8643268727082242,
+      "eval_PRM F1 Neg": 0.6086956521739131,
+      "eval_PRM NPV": 0.6086956521739131,
+      "eval_PRM Precision": 0.891566265060241,
+      "eval_PRM Recall": 0.891566265060241,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35129615664482117,
+      "eval_runtime": 16.8712,
+      "eval_samples_per_second": 2.608,
+      "eval_steps_per_second": 0.178,
+      "step": 3870
+    },
+    {
+      "epoch": 0.9971343014457288,
+      "grad_norm": 0.1728700765894953,
+      "learning_rate": 2.4469461433751507e-09,
+      "loss": 0.3819,
+      "step": 3871
+    },
+    {
+      "epoch": 0.9973918923270116,
+      "grad_norm": 0.21291593566328246,
+      "learning_rate": 2.0222724027074703e-09,
+      "loss": 0.3915,
+      "step": 3872
+    },
+    {
+      "epoch": 0.9976494832082944,
+      "grad_norm": 0.1550645695952983,
+      "learning_rate": 1.6380427441731184e-09,
+      "loss": 0.3743,
+      "step": 3873
+    },
+    {
+      "epoch": 0.9979070740895772,
+      "grad_norm": 0.166816058084006,
+      "learning_rate": 1.2942574785623772e-09,
+      "loss": 0.3438,
+      "step": 3874
+    },
+    {
+      "epoch": 0.99816466497086,
+      "grad_norm": 0.18251919599287586,
+      "learning_rate": 9.90916883986115e-10,
+      "loss": 0.2944,
+      "step": 3875
+    },
+    {
+      "epoch": 0.99816466497086,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.863017286537454,
+      "eval_PRM F1 Neg": 0.5957446808510638,
+      "eval_PRM NPV": 0.5833333333333334,
+      "eval_PRM Precision": 0.8902439024390244,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35129615664482117,
+      "eval_runtime": 16.9038,
+      "eval_samples_per_second": 2.603,
+      "eval_steps_per_second": 0.177,
+      "step": 3875
+    },
+    {
+      "epoch": 0.9984222558521428,
+      "grad_norm": 0.15290702113417942,
+      "learning_rate": 7.280212058091706e-10,
+      "loss": 0.2959,
+      "step": 3876
+    },
+    {
+      "epoch": 0.9986798467334257,
+      "grad_norm": 0.14057128974127117,
+      "learning_rate": 5.055706566947649e-10,
+      "loss": 0.4001,
+      "step": 3877
+    },
+    {
+      "epoch": 0.9989374376147084,
+      "grad_norm": 0.14824828438814447,
+      "learning_rate": 3.2356541658784546e-10,
+      "loss": 0.3468,
+      "step": 3878
+    },
+    {
+      "epoch": 0.9991950284959913,
+      "grad_norm": 0.15676207191010466,
+      "learning_rate": 1.8200563271508765e-10,
+      "loss": 0.4174,
+      "step": 3879
+    },
+    {
+      "epoch": 0.999452619377274,
+      "grad_norm": 0.13897661504636863,
+      "learning_rate": 8.08914195793431e-11,
+      "loss": 0.342,
+      "step": 3880
+    },
+    {
+      "epoch": 0.999452619377274,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.891566265060241,
+      "eval_PRM F1 AUC": 0.7501309586170771,
+      "eval_PRM F1 AUC (fixed)": 0.859088528025144,
+      "eval_PRM F1 Neg": 0.6086956521739131,
+      "eval_PRM NPV": 0.6086956521739131,
+      "eval_PRM Precision": 0.891566265060241,
+      "eval_PRM Recall": 0.891566265060241,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35400390625,
+      "eval_runtime": 17.0944,
+      "eval_samples_per_second": 2.574,
+      "eval_steps_per_second": 0.175,
+      "step": 3880
+    },
+    {
+      "epoch": 0.9997102102585569,
+      "grad_norm": 0.18222748366726835,
+      "learning_rate": 2.0222858987395398e-11,
+      "loss": 0.2709,
+      "step": 3881
+    },
+    {
+      "epoch": 0.9999678011398396,
+      "grad_norm": 0.175363916804034,
+      "learning_rate": 0.0,
+      "loss": 0.4232,
+      "step": 3882
     }
   ],
   "logging_steps": 1,
@@ -36438,12 +40404,12 @@
         "should_evaluate": false,
         "should_log": false,
         "should_save": true,
-        "should_training_stop": false
+        "should_training_stop": true
       },
       "attributes": {}
     }
   },
-  "total_flos": 7639333596430336.0,
+  "total_flos": 8470762821648384.0,
   "train_batch_size": 4,
   "trial_name": null,
   "trial_params": null