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": 0.6439772032070065,
+  "epoch": 0.7727726438484077,
   "eval_steps": 5,
-  "global_step": 2500,
+  "global_step": 3000,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
@@ -26024,6 +26024,5206 @@
       "eval_samples_per_second": 2.658,
       "eval_steps_per_second": 0.181,
       "step": 2500
+    },
+    {
+      "epoch": 0.6442347940882893,
+      "grad_norm": 0.14873657646250013,
+      "learning_rate": 3.3857959368118874e-05,
+      "loss": 0.3583,
+      "step": 2501
+    },
+    {
+      "epoch": 0.6444923849695721,
+      "grad_norm": 0.16040972779397752,
+      "learning_rate": 3.3815404055085045e-05,
+      "loss": 0.3418,
+      "step": 2502
+    },
+    {
+      "epoch": 0.6447499758508549,
+      "grad_norm": 0.19628354525783215,
+      "learning_rate": 3.377286183400328e-05,
+      "loss": 0.3715,
+      "step": 2503
+    },
+    {
+      "epoch": 0.6450075667321377,
+      "grad_norm": 0.23333807996267833,
+      "learning_rate": 3.373033273928659e-05,
+      "loss": 0.4204,
+      "step": 2504
+    },
+    {
+      "epoch": 0.6452651576134205,
+      "grad_norm": 0.15238236387599147,
+      "learning_rate": 3.368781680533737e-05,
+      "loss": 0.323,
+      "step": 2505
+    },
+    {
+      "epoch": 0.6452651576134205,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8462545835515978,
+      "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": 0.3501420319080353,
+      "eval_runtime": 16.7055,
+      "eval_samples_per_second": 2.634,
+      "eval_steps_per_second": 0.18,
+      "step": 2505
+    },
+    {
+      "epoch": 0.6455227484947033,
+      "grad_norm": 0.154791537795474,
+      "learning_rate": 3.364531406654736e-05,
+      "loss": 0.3445,
+      "step": 2506
+    },
+    {
+      "epoch": 0.6457803393759861,
+      "grad_norm": 0.13869050511999345,
+      "learning_rate": 3.3602824557297666e-05,
+      "loss": 0.2986,
+      "step": 2507
+    },
+    {
+      "epoch": 0.6460379302572689,
+      "grad_norm": 0.10900705151009309,
+      "learning_rate": 3.356034831195861e-05,
+      "loss": 0.2838,
+      "step": 2508
+    },
+    {
+      "epoch": 0.6462955211385517,
+      "grad_norm": 0.16012915495858335,
+      "learning_rate": 3.35178853648899e-05,
+      "loss": 0.3429,
+      "step": 2509
+    },
+    {
+      "epoch": 0.6465531120198345,
+      "grad_norm": 0.188822535609594,
+      "learning_rate": 3.3475435750440356e-05,
+      "loss": 0.414,
+      "step": 2510
+    },
+    {
+      "epoch": 0.6465531120198345,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.891566265060241,
+      "eval_PRM F1 AUC": 0.7501309586170771,
+      "eval_PRM F1 AUC (fixed)": 0.8567312729177581,
+      "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": 0.3461470305919647,
+      "eval_runtime": 17.6969,
+      "eval_samples_per_second": 2.486,
+      "eval_steps_per_second": 0.17,
+      "step": 2510
+    },
+    {
+      "epoch": 0.6468107029011173,
+      "grad_norm": 0.1709800133075916,
+      "learning_rate": 3.3432999502948154e-05,
+      "loss": 0.3722,
+      "step": 2511
+    },
+    {
+      "epoch": 0.6470682937824,
+      "grad_norm": 0.14942482644257157,
+      "learning_rate": 3.339057665674054e-05,
+      "loss": 0.381,
+      "step": 2512
+    },
+    {
+      "epoch": 0.6473258846636829,
+      "grad_norm": 0.17726803328848934,
+      "learning_rate": 3.334816724613394e-05,
+      "loss": 0.3147,
+      "step": 2513
+    },
+    {
+      "epoch": 0.6475834755449658,
+      "grad_norm": 0.22306420205756,
+      "learning_rate": 3.330577130543399e-05,
+      "loss": 0.4232,
+      "step": 2514
+    },
+    {
+      "epoch": 0.6478410664262485,
+      "grad_norm": 0.13931705972939903,
+      "learning_rate": 3.326338886893534e-05,
+      "loss": 0.309,
+      "step": 2515
+    },
+    {
+      "epoch": 0.6478410664262485,
+      "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": 0.3401544690132141,
+      "eval_runtime": 16.8532,
+      "eval_samples_per_second": 2.611,
+      "eval_steps_per_second": 0.178,
+      "step": 2515
+    },
+    {
+      "epoch": 0.6480986573075314,
+      "grad_norm": 0.20530377006264572,
+      "learning_rate": 3.3221019970921766e-05,
+      "loss": 0.3751,
+      "step": 2516
+    },
+    {
+      "epoch": 0.6483562481888141,
+      "grad_norm": 0.17330808956810748,
+      "learning_rate": 3.3178664645666066e-05,
+      "loss": 0.4153,
+      "step": 2517
+    },
+    {
+      "epoch": 0.648613839070097,
+      "grad_norm": 0.12881910684658696,
+      "learning_rate": 3.3136322927430086e-05,
+      "loss": 0.3523,
+      "step": 2518
+    },
+    {
+      "epoch": 0.6488714299513797,
+      "grad_norm": 0.1374858295452381,
+      "learning_rate": 3.309399485046466e-05,
+      "loss": 0.3593,
+      "step": 2519
+    },
+    {
+      "epoch": 0.6491290208326626,
+      "grad_norm": 0.20584361316619515,
+      "learning_rate": 3.3051680449009535e-05,
+      "loss": 0.4538,
+      "step": 2520
+    },
+    {
+      "epoch": 0.6491290208326626,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.891566265060241,
+      "eval_PRM F1 AUC": 0.7501309586170771,
+      "eval_PRM F1 AUC (fixed)": 0.8567312729177581,
+      "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": 0.34543678164482117,
+      "eval_runtime": 16.8519,
+      "eval_samples_per_second": 2.611,
+      "eval_steps_per_second": 0.178,
+      "step": 2520
+    },
+    {
+      "epoch": 0.6493866117139453,
+      "grad_norm": 0.14377796745919497,
+      "learning_rate": 3.300937975729347e-05,
+      "loss": 0.3428,
+      "step": 2521
+    },
+    {
+      "epoch": 0.6496442025952281,
+      "grad_norm": 0.1522137015431133,
+      "learning_rate": 3.2967092809534083e-05,
+      "loss": 0.3596,
+      "step": 2522
+    },
+    {
+      "epoch": 0.6499017934765109,
+      "grad_norm": 0.18118934661681027,
+      "learning_rate": 3.2924819639937924e-05,
+      "loss": 0.427,
+      "step": 2523
+    },
+    {
+      "epoch": 0.6501593843577937,
+      "grad_norm": 0.17093884238375687,
+      "learning_rate": 3.2882560282700336e-05,
+      "loss": 0.3756,
+      "step": 2524
+    },
+    {
+      "epoch": 0.6504169752390765,
+      "grad_norm": 0.14721751872523076,
+      "learning_rate": 3.284031477200553e-05,
+      "loss": 0.3446,
+      "step": 2525
+    },
+    {
+      "epoch": 0.6504169752390765,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.891566265060241,
+      "eval_PRM F1 AUC": 0.7501309586170771,
+      "eval_PRM F1 AUC (fixed)": 0.8512310110005239,
+      "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": 0.34872159361839294,
+      "eval_runtime": 16.9321,
+      "eval_samples_per_second": 2.599,
+      "eval_steps_per_second": 0.177,
+      "step": 2525
+    },
+    {
+      "epoch": 0.6506745661203593,
+      "grad_norm": 0.14190700115035357,
+      "learning_rate": 3.2798083142026514e-05,
+      "loss": 0.3009,
+      "step": 2526
+    },
+    {
+      "epoch": 0.6509321570016421,
+      "grad_norm": 0.16798833482760195,
+      "learning_rate": 3.275586542692504e-05,
+      "loss": 0.3929,
+      "step": 2527
+    },
+    {
+      "epoch": 0.6511897478829249,
+      "grad_norm": 0.13038401286059076,
+      "learning_rate": 3.271366166085166e-05,
+      "loss": 0.3491,
+      "step": 2528
+    },
+    {
+      "epoch": 0.6514473387642078,
+      "grad_norm": 0.15305136825663915,
+      "learning_rate": 3.267147187794556e-05,
+      "loss": 0.3568,
+      "step": 2529
+    },
+    {
+      "epoch": 0.6517049296454905,
+      "grad_norm": 0.1311699769335781,
+      "learning_rate": 3.262929611233471e-05,
+      "loss": 0.2919,
+      "step": 2530
+    },
+    {
+      "epoch": 0.6517049296454905,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.891566265060241,
+      "eval_PRM F1 AUC": 0.7501309586170771,
+      "eval_PRM F1 AUC (fixed)": 0.8470403352540599,
+      "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": 17.7648,
+      "eval_samples_per_second": 2.477,
+      "eval_steps_per_second": 0.169,
+      "step": 2530
+    },
+    {
+      "epoch": 0.6519625205267734,
+      "grad_norm": 0.18555112501180562,
+      "learning_rate": 3.258713439813566e-05,
+      "loss": 0.4333,
+      "step": 2531
+    },
+    {
+      "epoch": 0.6522201114080561,
+      "grad_norm": 0.1432515596280174,
+      "learning_rate": 3.2544986769453616e-05,
+      "loss": 0.3027,
+      "step": 2532
+    },
+    {
+      "epoch": 0.652477702289339,
+      "grad_norm": 0.16666607778718331,
+      "learning_rate": 3.2502853260382436e-05,
+      "loss": 0.3986,
+      "step": 2533
+    },
+    {
+      "epoch": 0.6527352931706217,
+      "grad_norm": 0.1805640643644997,
+      "learning_rate": 3.246073390500449e-05,
+      "loss": 0.3662,
+      "step": 2534
+    },
+    {
+      "epoch": 0.6529928840519046,
+      "grad_norm": 0.15817752408897212,
+      "learning_rate": 3.241862873739075e-05,
+      "loss": 0.3749,
+      "step": 2535
+    },
+    {
+      "epoch": 0.6529928840519046,
+      "eval_PRM Accuracy": 0.839622641509434,
+      "eval_PRM F1": 0.896969696969697,
+      "eval_PRM F1 AUC": 0.7718700890518595,
+      "eval_PRM F1 AUC (fixed)": 0.8504452592980618,
+      "eval_PRM F1 Neg": 0.6382978723404256,
+      "eval_PRM NPV": 0.625,
+      "eval_PRM Precision": 0.9024390243902439,
+      "eval_PRM Recall": 0.891566265060241,
+      "eval_PRM Specificty": 0.6521739130434783,
+      "eval_loss": 0.3546253442764282,
+      "eval_runtime": 17.0846,
+      "eval_samples_per_second": 2.575,
+      "eval_steps_per_second": 0.176,
+      "step": 2535
+    },
+    {
+      "epoch": 0.6532504749331873,
+      "grad_norm": 0.2069291485258539,
+      "learning_rate": 3.237653779160067e-05,
+      "loss": 0.4628,
+      "step": 2536
+    },
+    {
+      "epoch": 0.6535080658144702,
+      "grad_norm": 0.17594609060190622,
+      "learning_rate": 3.2334461101682246e-05,
+      "loss": 0.4072,
+      "step": 2537
+    },
+    {
+      "epoch": 0.6537656566957529,
+      "grad_norm": 0.1454335036500637,
+      "learning_rate": 3.229239870167191e-05,
+      "loss": 0.3352,
+      "step": 2538
+    },
+    {
+      "epoch": 0.6540232475770358,
+      "grad_norm": 0.1874278194677057,
+      "learning_rate": 3.225035062559452e-05,
+      "loss": 0.3919,
+      "step": 2539
+    },
+    {
+      "epoch": 0.6542808384583185,
+      "grad_norm": 0.16750406533288847,
+      "learning_rate": 3.220831690746339e-05,
+      "loss": 0.41,
+      "step": 2540
+    },
+    {
+      "epoch": 0.6542808384583185,
+      "eval_PRM Accuracy": 0.839622641509434,
+      "eval_PRM F1": 0.896969696969697,
+      "eval_PRM F1 AUC": 0.7718700890518595,
+      "eval_PRM F1 AUC (fixed)": 0.8454688318491357,
+      "eval_PRM F1 Neg": 0.6382978723404256,
+      "eval_PRM NPV": 0.625,
+      "eval_PRM Precision": 0.9024390243902439,
+      "eval_PRM Recall": 0.891566265060241,
+      "eval_PRM Specificty": 0.6521739130434783,
+      "eval_loss": 0.3556019067764282,
+      "eval_runtime": 17.665,
+      "eval_samples_per_second": 2.491,
+      "eval_steps_per_second": 0.17,
+      "step": 2540
+    },
+    {
+      "epoch": 0.6545384293396014,
+      "grad_norm": 0.16608541292512252,
+      "learning_rate": 3.216629758128018e-05,
+      "loss": 0.3982,
+      "step": 2541
+    },
+    {
+      "epoch": 0.6547960202208842,
+      "grad_norm": 0.1763015395163615,
+      "learning_rate": 3.212429268103495e-05,
+      "loss": 0.4252,
+      "step": 2542
+    },
+    {
+      "epoch": 0.655053611102167,
+      "grad_norm": 0.12489844211720086,
+      "learning_rate": 3.208230224070605e-05,
+      "loss": 0.2988,
+      "step": 2543
+    },
+    {
+      "epoch": 0.6553112019834498,
+      "grad_norm": 0.17351401694461402,
+      "learning_rate": 3.204032629426016e-05,
+      "loss": 0.3988,
+      "step": 2544
+    },
+    {
+      "epoch": 0.6555687928647326,
+      "grad_norm": 0.1337855883003999,
+      "learning_rate": 3.199836487565222e-05,
+      "loss": 0.3521,
+      "step": 2545
+    },
+    {
+      "epoch": 0.6555687928647326,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.8902439024390244,
+      "eval_PRM F1 AUC": 0.7658459926663174,
+      "eval_PRM F1 AUC (fixed)": 0.8412781561026715,
+      "eval_PRM F1 Neg": 0.625,
+      "eval_PRM NPV": 0.6,
+      "eval_PRM Precision": 0.9012345679012346,
+      "eval_PRM Recall": 0.8795180722891566,
+      "eval_PRM Specificty": 0.6521739130434783,
+      "eval_loss": 0.35946378111839294,
+      "eval_runtime": 16.9641,
+      "eval_samples_per_second": 2.594,
+      "eval_steps_per_second": 0.177,
+      "step": 2545
+    },
+    {
+      "epoch": 0.6558263837460154,
+      "grad_norm": 0.18725366477704852,
+      "learning_rate": 3.195641801882541e-05,
+      "loss": 0.3138,
+      "step": 2546
+    },
+    {
+      "epoch": 0.6560839746272982,
+      "grad_norm": 0.24688665103473245,
+      "learning_rate": 3.191448575771119e-05,
+      "loss": 0.4933,
+      "step": 2547
+    },
+    {
+      "epoch": 0.656341565508581,
+      "grad_norm": 0.14498573625614256,
+      "learning_rate": 3.1872568126229095e-05,
+      "loss": 0.3223,
+      "step": 2548
+    },
+    {
+      "epoch": 0.6565991563898638,
+      "grad_norm": 0.13504588754413635,
+      "learning_rate": 3.1830665158286976e-05,
+      "loss": 0.3042,
+      "step": 2549
+    },
+    {
+      "epoch": 0.6568567472711466,
+      "grad_norm": 0.122929927826042,
+      "learning_rate": 3.1788776887780684e-05,
+      "loss": 0.3023,
+      "step": 2550
+    },
+    {
+      "epoch": 0.6568567472711466,
+      "eval_PRM Accuracy": 0.839622641509434,
+      "eval_PRM F1": 0.896969696969697,
+      "eval_PRM F1 AUC": 0.7718700890518595,
+      "eval_PRM F1 AUC (fixed)": 0.8452069146149817,
+      "eval_PRM F1 Neg": 0.6382978723404256,
+      "eval_PRM NPV": 0.625,
+      "eval_PRM Precision": 0.9024390243902439,
+      "eval_PRM Recall": 0.891566265060241,
+      "eval_PRM Specificty": 0.6521739130434783,
+      "eval_loss": 0.3597301244735718,
+      "eval_runtime": 17.1128,
+      "eval_samples_per_second": 2.571,
+      "eval_steps_per_second": 0.175,
+      "step": 2550
+    },
+    {
+      "epoch": 0.6571143381524294,
+      "grad_norm": 0.12923353371719942,
+      "learning_rate": 3.17469033485943e-05,
+      "loss": 0.3137,
+      "step": 2551
+    },
+    {
+      "epoch": 0.6573719290337122,
+      "grad_norm": 0.16529377522823094,
+      "learning_rate": 3.170504457459989e-05,
+      "loss": 0.3606,
+      "step": 2552
+    },
+    {
+      "epoch": 0.657629519914995,
+      "grad_norm": 0.12628829466835872,
+      "learning_rate": 3.166320059965762e-05,
+      "loss": 0.3095,
+      "step": 2553
+    },
+    {
+      "epoch": 0.6578871107962778,
+      "grad_norm": 0.16930794155717083,
+      "learning_rate": 3.16213714576157e-05,
+      "loss": 0.3127,
+      "step": 2554
+    },
+    {
+      "epoch": 0.6581447016775606,
+      "grad_norm": 0.11479473280686057,
+      "learning_rate": 3.15795571823103e-05,
+      "loss": 0.2792,
+      "step": 2555
+    },
+    {
+      "epoch": 0.6581447016775606,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8525405971712938,
+      "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": 0.3607066869735718,
+      "eval_runtime": 17.3594,
+      "eval_samples_per_second": 2.535,
+      "eval_steps_per_second": 0.173,
+      "step": 2555
+    },
+    {
+      "epoch": 0.6584022925588434,
+      "grad_norm": 0.1387933607014311,
+      "learning_rate": 3.1537757807565614e-05,
+      "loss": 0.2783,
+      "step": 2556
+    },
+    {
+      "epoch": 0.6586598834401263,
+      "grad_norm": 0.1477684610839945,
+      "learning_rate": 3.1495973367193734e-05,
+      "loss": 0.3637,
+      "step": 2557
+    },
+    {
+      "epoch": 0.658917474321409,
+      "grad_norm": 0.15011071090661027,
+      "learning_rate": 3.1454203894994713e-05,
+      "loss": 0.2746,
+      "step": 2558
+    },
+    {
+      "epoch": 0.6591750652026919,
+      "grad_norm": 0.14828228800010632,
+      "learning_rate": 3.1412449424756474e-05,
+      "loss": 0.3308,
+      "step": 2559
+    },
+    {
+      "epoch": 0.6594326560839746,
+      "grad_norm": 0.20438333584587137,
+      "learning_rate": 3.1370709990254786e-05,
+      "loss": 0.3942,
+      "step": 2560
+    },
+    {
+      "epoch": 0.6594326560839746,
+      "eval_PRM Accuracy": 0.839622641509434,
+      "eval_PRM F1": 0.8982035928143712,
+      "eval_PRM F1 AUC": 0.7561550550026193,
+      "eval_PRM F1 AUC (fixed)": 0.8598742797276061,
+      "eval_PRM F1 Neg": 0.6222222222222222,
+      "eval_PRM NPV": 0.6363636363636364,
+      "eval_PRM Precision": 0.8928571428571429,
+      "eval_PRM Recall": 0.9036144578313253,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3662553131580353,
+      "eval_runtime": 17.0899,
+      "eval_samples_per_second": 2.575,
+      "eval_steps_per_second": 0.176,
+      "step": 2560
+    },
+    {
+      "epoch": 0.6596902469652575,
+      "grad_norm": 0.2516696310753756,
+      "learning_rate": 3.1328985625253304e-05,
+      "loss": 0.4709,
+      "step": 2561
+    },
+    {
+      "epoch": 0.6599478378465402,
+      "grad_norm": 0.16598887701829515,
+      "learning_rate": 3.1287276363503446e-05,
+      "loss": 0.3714,
+      "step": 2562
+    },
+    {
+      "epoch": 0.6602054287278231,
+      "grad_norm": 0.1591790676833675,
+      "learning_rate": 3.1245582238744444e-05,
+      "loss": 0.3606,
+      "step": 2563
+    },
+    {
+      "epoch": 0.6604630196091058,
+      "grad_norm": 0.14624376218522936,
+      "learning_rate": 3.120390328470326e-05,
+      "loss": 0.3444,
+      "step": 2564
+    },
+    {
+      "epoch": 0.6607206104903887,
+      "grad_norm": 0.17024500235389717,
+      "learning_rate": 3.116223953509463e-05,
+      "loss": 0.3374,
+      "step": 2565
+    },
+    {
+      "epoch": 0.6607206104903887,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8619696176008382,
+      "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": 0.3558238744735718,
+      "eval_runtime": 17.573,
+      "eval_samples_per_second": 2.504,
+      "eval_steps_per_second": 0.171,
+      "step": 2565
+    },
+    {
+      "epoch": 0.6609782013716714,
+      "grad_norm": 0.16071851788388838,
+      "learning_rate": 3.112059102362093e-05,
+      "loss": 0.4286,
+      "step": 2566
+    },
+    {
+      "epoch": 0.6612357922529543,
+      "grad_norm": 0.17033435391975124,
+      "learning_rate": 3.107895778397224e-05,
+      "loss": 0.4324,
+      "step": 2567
+    },
+    {
+      "epoch": 0.661493383134237,
+      "grad_norm": 0.16323632109513192,
+      "learning_rate": 3.1037339849826315e-05,
+      "loss": 0.3228,
+      "step": 2568
+    },
+    {
+      "epoch": 0.6617509740155199,
+      "grad_norm": 0.2536498211462177,
+      "learning_rate": 3.099573725484844e-05,
+      "loss": 0.4613,
+      "step": 2569
+    },
+    {
+      "epoch": 0.6620085648968026,
+      "grad_norm": 0.16031947266740823,
+      "learning_rate": 3.095415003269163e-05,
+      "loss": 0.3737,
+      "step": 2570
+    },
+    {
+      "epoch": 0.6620085648968026,
+      "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": 0.3516956567764282,
+      "eval_runtime": 17.0658,
+      "eval_samples_per_second": 2.578,
+      "eval_steps_per_second": 0.176,
+      "step": 2570
+    },
+    {
+      "epoch": 0.6622661557780855,
+      "grad_norm": 0.12609206435852544,
+      "learning_rate": 3.0912578216996315e-05,
+      "loss": 0.2846,
+      "step": 2571
+    },
+    {
+      "epoch": 0.6625237466593683,
+      "grad_norm": 0.16046671530905887,
+      "learning_rate": 3.087102184139059e-05,
+      "loss": 0.3817,
+      "step": 2572
+    },
+    {
+      "epoch": 0.662781337540651,
+      "grad_norm": 0.17389660168352186,
+      "learning_rate": 3.082948093948997e-05,
+      "loss": 0.4318,
+      "step": 2573
+    },
+    {
+      "epoch": 0.6630389284219339,
+      "grad_norm": 0.14324737030946294,
+      "learning_rate": 3.078795554489748e-05,
+      "loss": 0.3111,
+      "step": 2574
+    },
+    {
+      "epoch": 0.6632965193032166,
+      "grad_norm": 0.14173912099647726,
+      "learning_rate": 3.074644569120364e-05,
+      "loss": 0.3851,
+      "step": 2575
+    },
+    {
+      "epoch": 0.6632965193032166,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.856993190151912,
+      "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": 17.0289,
+      "eval_samples_per_second": 2.584,
+      "eval_steps_per_second": 0.176,
+      "step": 2575
+    },
+    {
+      "epoch": 0.6635541101844995,
+      "grad_norm": 0.11450529030492015,
+      "learning_rate": 3.070495141198634e-05,
+      "loss": 0.3292,
+      "step": 2576
+    },
+    {
+      "epoch": 0.6638117010657822,
+      "grad_norm": 0.16428345459596733,
+      "learning_rate": 3.066347274081091e-05,
+      "loss": 0.403,
+      "step": 2577
+    },
+    {
+      "epoch": 0.6640692919470651,
+      "grad_norm": 0.12045040511375114,
+      "learning_rate": 3.0622009711230036e-05,
+      "loss": 0.3525,
+      "step": 2578
+    },
+    {
+      "epoch": 0.6643268828283478,
+      "grad_norm": 0.12680063256725685,
+      "learning_rate": 3.058056235678378e-05,
+      "loss": 0.3788,
+      "step": 2579
+    },
+    {
+      "epoch": 0.6645844737096307,
+      "grad_norm": 0.16746684595194214,
+      "learning_rate": 3.053913071099947e-05,
+      "loss": 0.3487,
+      "step": 2580
+    },
+    {
+      "epoch": 0.6645844737096307,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8780487804878049,
+      "eval_PRM F1 AUC": 0.7380827658459927,
+      "eval_PRM F1 AUC (fixed)": 0.851754845468832,
+      "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": 0.3593306243419647,
+      "eval_runtime": 17.14,
+      "eval_samples_per_second": 2.567,
+      "eval_steps_per_second": 0.175,
+      "step": 2580
+    },
+    {
+      "epoch": 0.6648420645909134,
+      "grad_norm": 0.14915604575310126,
+      "learning_rate": 3.0497714807391774e-05,
+      "loss": 0.4035,
+      "step": 2581
+    },
+    {
+      "epoch": 0.6650996554721963,
+      "grad_norm": 0.2030432434088284,
+      "learning_rate": 3.0456314679462627e-05,
+      "loss": 0.4354,
+      "step": 2582
+    },
+    {
+      "epoch": 0.665357246353479,
+      "grad_norm": 0.16173799671708103,
+      "learning_rate": 3.041493036070115e-05,
+      "loss": 0.4492,
+      "step": 2583
+    },
+    {
+      "epoch": 0.6656148372347619,
+      "grad_norm": 0.13206398818964307,
+      "learning_rate": 3.0373561884583744e-05,
+      "loss": 0.3664,
+      "step": 2584
+    },
+    {
+      "epoch": 0.6658724281160447,
+      "grad_norm": 0.15272377410733054,
+      "learning_rate": 3.0332209284573958e-05,
+      "loss": 0.3629,
+      "step": 2585
+    },
+    {
+      "epoch": 0.6658724281160447,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.891566265060241,
+      "eval_PRM F1 AUC": 0.7501309586170771,
+      "eval_PRM F1 AUC (fixed)": 0.8473022524882138,
+      "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": 0.36257103085517883,
+      "eval_runtime": 16.9612,
+      "eval_samples_per_second": 2.594,
+      "eval_steps_per_second": 0.177,
+      "step": 2585
+    },
+    {
+      "epoch": 0.6661300189973275,
+      "grad_norm": 0.13067819540064435,
+      "learning_rate": 3.0290872594122487e-05,
+      "loss": 0.3481,
+      "step": 2586
+    },
+    {
+      "epoch": 0.6663876098786103,
+      "grad_norm": 0.1511483043515646,
+      "learning_rate": 3.0249551846667207e-05,
+      "loss": 0.3417,
+      "step": 2587
+    },
+    {
+      "epoch": 0.6666452007598931,
+      "grad_norm": 0.15426699334791533,
+      "learning_rate": 3.020824707563301e-05,
+      "loss": 0.3745,
+      "step": 2588
+    },
+    {
+      "epoch": 0.6669027916411759,
+      "grad_norm": 0.13848777860046718,
+      "learning_rate": 3.0166958314431985e-05,
+      "loss": 0.3631,
+      "step": 2589
+    },
+    {
+      "epoch": 0.6671603825224587,
+      "grad_norm": 0.19336063860804856,
+      "learning_rate": 3.0125685596463137e-05,
+      "loss": 0.3885,
+      "step": 2590
+    },
+    {
+      "epoch": 0.6671603825224587,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.891566265060241,
+      "eval_PRM F1 AUC": 0.7501309586170771,
+      "eval_PRM F1 AUC (fixed)": 0.8488737558931377,
+      "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": 0.3624378442764282,
+      "eval_runtime": 17.2346,
+      "eval_samples_per_second": 2.553,
+      "eval_steps_per_second": 0.174,
+      "step": 2590
+    },
+    {
+      "epoch": 0.6674179734037415,
+      "grad_norm": 0.14882306434604864,
+      "learning_rate": 3.0084428955112626e-05,
+      "loss": 0.4,
+      "step": 2591
+    },
+    {
+      "epoch": 0.6676755642850243,
+      "grad_norm": 0.15076421723970607,
+      "learning_rate": 3.0043188423753498e-05,
+      "loss": 0.3672,
+      "step": 2592
+    },
+    {
+      "epoch": 0.6679331551663071,
+      "grad_norm": 0.15032960970703027,
+      "learning_rate": 3.000196403574581e-05,
+      "loss": 0.4062,
+      "step": 2593
+    },
+    {
+      "epoch": 0.6681907460475899,
+      "grad_norm": 0.13129512333552704,
+      "learning_rate": 2.996075582443658e-05,
+      "loss": 0.345,
+      "step": 2594
+    },
+    {
+      "epoch": 0.6684483369288727,
+      "grad_norm": 0.2087041615741322,
+      "learning_rate": 2.9919563823159713e-05,
+      "loss": 0.3311,
+      "step": 2595
+    },
+    {
+      "epoch": 0.6684483369288727,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.891566265060241,
+      "eval_PRM F1 AUC": 0.7501309586170771,
+      "eval_PRM F1 AUC (fixed)": 0.854374017810372,
+      "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": 0.3663884997367859,
+      "eval_runtime": 16.993,
+      "eval_samples_per_second": 2.589,
+      "eval_steps_per_second": 0.177,
+      "step": 2595
+    },
+    {
+      "epoch": 0.6687059278101555,
+      "grad_norm": 0.1439844716504298,
+      "learning_rate": 2.9878388065236017e-05,
+      "loss": 0.3744,
+      "step": 2596
+    },
+    {
+      "epoch": 0.6689635186914383,
+      "grad_norm": 0.12077440029401869,
+      "learning_rate": 2.9837228583973143e-05,
+      "loss": 0.3422,
+      "step": 2597
+    },
+    {
+      "epoch": 0.6692211095727211,
+      "grad_norm": 0.13144361481789127,
+      "learning_rate": 2.97960854126656e-05,
+      "loss": 0.2947,
+      "step": 2598
+    },
+    {
+      "epoch": 0.6694787004540039,
+      "grad_norm": 0.15236813351012043,
+      "learning_rate": 2.9754958584594694e-05,
+      "loss": 0.3451,
+      "step": 2599
+    },
+    {
+      "epoch": 0.6697362913352868,
+      "grad_norm": 0.13679736408617219,
+      "learning_rate": 2.971384813302847e-05,
+      "loss": 0.2935,
+      "step": 2600
+    },
+    {
+      "epoch": 0.6697362913352868,
+      "eval_PRM Accuracy": 0.839622641509434,
+      "eval_PRM F1": 0.8994082840236687,
+      "eval_PRM F1 AUC": 0.7404400209533788,
+      "eval_PRM F1 AUC (fixed)": 0.8624934520691461,
+      "eval_PRM F1 Neg": 0.6046511627906976,
+      "eval_PRM NPV": 0.65,
+      "eval_PRM Precision": 0.8837209302325582,
+      "eval_PRM Recall": 0.9156626506024096,
+      "eval_PRM Specificty": 0.5652173913043478,
+      "eval_loss": 0.3698064684867859,
+      "eval_runtime": 17.0757,
+      "eval_samples_per_second": 2.577,
+      "eval_steps_per_second": 0.176,
+      "step": 2600
+    },
+    {
+      "epoch": 0.6699938822165695,
+      "grad_norm": 0.14424842156304019,
+      "learning_rate": 2.9672754091221805e-05,
+      "loss": 0.3358,
+      "step": 2601
+    },
+    {
+      "epoch": 0.6702514730978524,
+      "grad_norm": 0.271654280797804,
+      "learning_rate": 2.9631676492416228e-05,
+      "loss": 0.3578,
+      "step": 2602
+    },
+    {
+      "epoch": 0.6705090639791351,
+      "grad_norm": 0.16214664750967217,
+      "learning_rate": 2.9590615369840025e-05,
+      "loss": 0.3556,
+      "step": 2603
+    },
+    {
+      "epoch": 0.670766654860418,
+      "grad_norm": 0.1341168484593591,
+      "learning_rate": 2.9549570756708105e-05,
+      "loss": 0.3674,
+      "step": 2604
+    },
+    {
+      "epoch": 0.6710242457417007,
+      "grad_norm": 0.14645915108003033,
+      "learning_rate": 2.9508542686222067e-05,
+      "loss": 0.3712,
+      "step": 2605
+    },
+    {
+      "epoch": 0.6710242457417007,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.8928571428571429,
+      "eval_PRM F1 AUC": 0.7344159245678366,
+      "eval_PRM F1 AUC (fixed)": 0.8664222105814563,
+      "eval_PRM F1 Neg": 0.5909090909090909,
+      "eval_PRM NPV": 0.6190476190476191,
+      "eval_PRM Precision": 0.8823529411764706,
+      "eval_PRM Recall": 0.9036144578313253,
+      "eval_PRM Specificty": 0.5652173913043478,
+      "eval_loss": 0.3648792505264282,
+      "eval_runtime": 17.011,
+      "eval_samples_per_second": 2.587,
+      "eval_steps_per_second": 0.176,
+      "step": 2605
+    },
+    {
+      "epoch": 0.6712818366229836,
+      "grad_norm": 0.18243098306205288,
+      "learning_rate": 2.9467531191570098e-05,
+      "loss": 0.4229,
+      "step": 2606
+    },
+    {
+      "epoch": 0.6715394275042663,
+      "grad_norm": 0.16135374167770358,
+      "learning_rate": 2.9426536305926968e-05,
+      "loss": 0.4295,
+      "step": 2607
+    },
+    {
+      "epoch": 0.6717970183855492,
+      "grad_norm": 0.15824305201161903,
+      "learning_rate": 2.938555806245406e-05,
+      "loss": 0.3339,
+      "step": 2608
+    },
+    {
+      "epoch": 0.6720546092668319,
+      "grad_norm": 0.1285985863255223,
+      "learning_rate": 2.934459649429922e-05,
+      "loss": 0.3539,
+      "step": 2609
+    },
+    {
+      "epoch": 0.6723122001481148,
+      "grad_norm": 0.14343018905641086,
+      "learning_rate": 2.9303651634596906e-05,
+      "loss": 0.3753,
+      "step": 2610
+    },
+    {
+      "epoch": 0.6723122001481148,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8528025144054479,
+      "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": 0.3631036877632141,
+      "eval_runtime": 17.0075,
+      "eval_samples_per_second": 2.587,
+      "eval_steps_per_second": 0.176,
+      "step": 2610
+    },
+    {
+      "epoch": 0.6725697910293975,
+      "grad_norm": 0.1289836886527462,
+      "learning_rate": 2.9262723516467938e-05,
+      "loss": 0.3051,
+      "step": 2611
+    },
+    {
+      "epoch": 0.6728273819106804,
+      "grad_norm": 0.17466695118132106,
+      "learning_rate": 2.9221812173019725e-05,
+      "loss": 0.435,
+      "step": 2612
+    },
+    {
+      "epoch": 0.6730849727919631,
+      "grad_norm": 0.16372684068951407,
+      "learning_rate": 2.9180917637345994e-05,
+      "loss": 0.3557,
+      "step": 2613
+    },
+    {
+      "epoch": 0.673342563673246,
+      "grad_norm": 0.18436446474435458,
+      "learning_rate": 2.9140039942526932e-05,
+      "loss": 0.4669,
+      "step": 2614
+    },
+    {
+      "epoch": 0.6736001545545288,
+      "grad_norm": 0.18459854773941667,
+      "learning_rate": 2.9099179121629117e-05,
+      "loss": 0.3882,
+      "step": 2615
+    },
+    {
+      "epoch": 0.6736001545545288,
+      "eval_PRM Accuracy": 0.8018867924528302,
+      "eval_PRM F1": 0.8711656441717791,
+      "eval_PRM F1 AUC": 0.7320586694604505,
+      "eval_PRM F1 AUC (fixed)": 0.8436354112100576,
+      "eval_PRM F1 Neg": 0.5714285714285714,
+      "eval_PRM NPV": 0.5384615384615384,
+      "eval_PRM Precision": 0.8875,
+      "eval_PRM Recall": 0.8554216867469879,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3697620630264282,
+      "eval_runtime": 17.1329,
+      "eval_samples_per_second": 2.568,
+      "eval_steps_per_second": 0.175,
+      "step": 2615
+    },
+    {
+      "epoch": 0.6738577454358116,
+      "grad_norm": 0.17466422864277772,
+      "learning_rate": 2.9058335207705406e-05,
+      "loss": 0.3659,
+      "step": 2616
+    },
+    {
+      "epoch": 0.6741153363170944,
+      "grad_norm": 0.235969171921171,
+      "learning_rate": 2.901750823379506e-05,
+      "loss": 0.3938,
+      "step": 2617
+    },
+    {
+      "epoch": 0.6743729271983772,
+      "grad_norm": 0.1705277553156338,
+      "learning_rate": 2.8976698232923592e-05,
+      "loss": 0.3951,
+      "step": 2618
+    },
+    {
+      "epoch": 0.67463051807966,
+      "grad_norm": 0.15408788625297992,
+      "learning_rate": 2.8935905238102834e-05,
+      "loss": 0.3954,
+      "step": 2619
+    },
+    {
+      "epoch": 0.6748881089609428,
+      "grad_norm": 0.13469875580433766,
+      "learning_rate": 2.889512928233078e-05,
+      "loss": 0.3426,
+      "step": 2620
+    },
+    {
+      "epoch": 0.6748881089609428,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8780487804878049,
+      "eval_PRM F1 AUC": 0.7380827658459927,
+      "eval_PRM F1 AUC (fixed)": 0.8336825563122053,
+      "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": 0.3745560944080353,
+      "eval_runtime": 17.4474,
+      "eval_samples_per_second": 2.522,
+      "eval_steps_per_second": 0.172,
+      "step": 2620
+    },
+    {
+      "epoch": 0.6751456998422256,
+      "grad_norm": 0.11833188829156593,
+      "learning_rate": 2.8854370398591696e-05,
+      "loss": 0.3412,
+      "step": 2621
+    },
+    {
+      "epoch": 0.6754032907235084,
+      "grad_norm": 0.139442085857202,
+      "learning_rate": 2.881362861985606e-05,
+      "loss": 0.3481,
+      "step": 2622
+    },
+    {
+      "epoch": 0.6756608816047912,
+      "grad_norm": 0.14973091363678256,
+      "learning_rate": 2.877290397908043e-05,
+      "loss": 0.4451,
+      "step": 2623
+    },
+    {
+      "epoch": 0.675918472486074,
+      "grad_norm": 0.13551257135595993,
+      "learning_rate": 2.87321965092076e-05,
+      "loss": 0.37,
+      "step": 2624
+    },
+    {
+      "epoch": 0.6761760633673568,
+      "grad_norm": 0.1608691455557866,
+      "learning_rate": 2.8691506243166365e-05,
+      "loss": 0.3478,
+      "step": 2625
+    },
+    {
+      "epoch": 0.6761760633673568,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.891566265060241,
+      "eval_PRM F1 AUC": 0.7501309586170771,
+      "eval_PRM F1 AUC (fixed)": 0.8336825563122053,
+      "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": 0.37930575013160706,
+      "eval_runtime": 17.7761,
+      "eval_samples_per_second": 2.475,
+      "eval_steps_per_second": 0.169,
+      "step": 2625
+    },
+    {
+      "epoch": 0.6764336542486395,
+      "grad_norm": 0.25653898986144474,
+      "learning_rate": 2.8650833213871754e-05,
+      "loss": 0.4363,
+      "step": 2626
+    },
+    {
+      "epoch": 0.6766912451299224,
+      "grad_norm": 0.14824584340574268,
+      "learning_rate": 2.8610177454224718e-05,
+      "loss": 0.3604,
+      "step": 2627
+    },
+    {
+      "epoch": 0.6769488360112053,
+      "grad_norm": 0.13973305765078076,
+      "learning_rate": 2.856953899711226e-05,
+      "loss": 0.3751,
+      "step": 2628
+    },
+    {
+      "epoch": 0.677206426892488,
+      "grad_norm": 0.13738295217547944,
+      "learning_rate": 2.8528917875407433e-05,
+      "loss": 0.2925,
+      "step": 2629
+    },
+    {
+      "epoch": 0.6774640177737709,
+      "grad_norm": 0.18460469582575612,
+      "learning_rate": 2.848831412196925e-05,
+      "loss": 0.2771,
+      "step": 2630
+    },
+    {
+      "epoch": 0.6774640177737709,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8315872184389732,
+      "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": 0.3798828125,
+      "eval_runtime": 16.881,
+      "eval_samples_per_second": 2.606,
+      "eval_steps_per_second": 0.178,
+      "step": 2630
+    },
+    {
+      "epoch": 0.6777216086550536,
+      "grad_norm": 0.14662183229158146,
+      "learning_rate": 2.8447727769642697e-05,
+      "loss": 0.2824,
+      "step": 2631
+    },
+    {
+      "epoch": 0.6779791995363365,
+      "grad_norm": 0.1560827145014329,
+      "learning_rate": 2.8407158851258598e-05,
+      "loss": 0.3876,
+      "step": 2632
+    },
+    {
+      "epoch": 0.6782367904176192,
+      "grad_norm": 0.1566681964080817,
+      "learning_rate": 2.8366607399633773e-05,
+      "loss": 0.3411,
+      "step": 2633
+    },
+    {
+      "epoch": 0.678494381298902,
+      "grad_norm": 0.14392669910857436,
+      "learning_rate": 2.8326073447570877e-05,
+      "loss": 0.3229,
+      "step": 2634
+    },
+    {
+      "epoch": 0.6787519721801848,
+      "grad_norm": 0.15826679514916564,
+      "learning_rate": 2.828555702785838e-05,
+      "loss": 0.3544,
+      "step": 2635
+    },
+    {
+      "epoch": 0.6787519721801848,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8368255631220535,
+      "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": 0.37668678164482117,
+      "eval_runtime": 17.1599,
+      "eval_samples_per_second": 2.564,
+      "eval_steps_per_second": 0.175,
+      "step": 2635
+    },
+    {
+      "epoch": 0.6790095630614676,
+      "grad_norm": 0.1599950181363227,
+      "learning_rate": 2.8245058173270622e-05,
+      "loss": 0.3794,
+      "step": 2636
+    },
+    {
+      "epoch": 0.6792671539427504,
+      "grad_norm": 0.12302097922258666,
+      "learning_rate": 2.8204576916567683e-05,
+      "loss": 0.3474,
+      "step": 2637
+    },
+    {
+      "epoch": 0.6795247448240332,
+      "grad_norm": 0.14502002261600558,
+      "learning_rate": 2.816411329049544e-05,
+      "loss": 0.3901,
+      "step": 2638
+    },
+    {
+      "epoch": 0.679782335705316,
+      "grad_norm": 0.17181379234799013,
+      "learning_rate": 2.812366732778553e-05,
+      "loss": 0.446,
+      "step": 2639
+    },
+    {
+      "epoch": 0.6800399265865988,
+      "grad_norm": 0.10429743608354924,
+      "learning_rate": 2.8083239061155232e-05,
+      "loss": 0.2877,
+      "step": 2640
+    },
+    {
+      "epoch": 0.6800399265865988,
+      "eval_PRM Accuracy": 0.8018867924528302,
+      "eval_PRM F1": 0.8711656441717791,
+      "eval_PRM F1 AUC": 0.7320586694604505,
+      "eval_PRM F1 AUC (fixed)": 0.8376113148245155,
+      "eval_PRM F1 Neg": 0.5714285714285714,
+      "eval_PRM NPV": 0.5384615384615384,
+      "eval_PRM Precision": 0.8875,
+      "eval_PRM Recall": 0.8554216867469879,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3638139069080353,
+      "eval_runtime": 16.8599,
+      "eval_samples_per_second": 2.61,
+      "eval_steps_per_second": 0.178,
+      "step": 2640
+    },
+    {
+      "epoch": 0.6802975174678816,
+      "grad_norm": 0.18565690935970933,
+      "learning_rate": 2.804282852330756e-05,
+      "loss": 0.3837,
+      "step": 2641
+    },
+    {
+      "epoch": 0.6805551083491644,
+      "grad_norm": 0.1518437366251305,
+      "learning_rate": 2.8002435746931194e-05,
+      "loss": 0.3514,
+      "step": 2642
+    },
+    {
+      "epoch": 0.6808126992304473,
+      "grad_norm": 0.18587792426939584,
+      "learning_rate": 2.796206076470044e-05,
+      "loss": 0.4258,
+      "step": 2643
+    },
+    {
+      "epoch": 0.68107029011173,
+      "grad_norm": 0.13405425666410486,
+      "learning_rate": 2.792170360927517e-05,
+      "loss": 0.2691,
+      "step": 2644
+    },
+    {
+      "epoch": 0.6813278809930129,
+      "grad_norm": 0.10515530517518289,
+      "learning_rate": 2.7881364313300894e-05,
+      "loss": 0.3167,
+      "step": 2645
+    },
+    {
+      "epoch": 0.6813278809930129,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8780487804878049,
+      "eval_PRM F1 AUC": 0.7380827658459927,
+      "eval_PRM F1 AUC (fixed)": 0.8486118386589837,
+      "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": 0.35830965638160706,
+      "eval_runtime": 17.227,
+      "eval_samples_per_second": 2.554,
+      "eval_steps_per_second": 0.174,
+      "step": 2645
+    },
+    {
+      "epoch": 0.6815854718742956,
+      "grad_norm": 0.11524372373749776,
+      "learning_rate": 2.784104290940862e-05,
+      "loss": 0.3228,
+      "step": 2646
+    },
+    {
+      "epoch": 0.6818430627555785,
+      "grad_norm": 0.15840319147870083,
+      "learning_rate": 2.780073943021493e-05,
+      "loss": 0.3861,
+      "step": 2647
+    },
+    {
+      "epoch": 0.6821006536368612,
+      "grad_norm": 0.14653782775636837,
+      "learning_rate": 2.7760453908321894e-05,
+      "loss": 0.3372,
+      "step": 2648
+    },
+    {
+      "epoch": 0.6823582445181441,
+      "grad_norm": 0.2162995700522589,
+      "learning_rate": 2.7720186376317024e-05,
+      "loss": 0.4424,
+      "step": 2649
+    },
+    {
+      "epoch": 0.6826158353994268,
+      "grad_norm": 0.1404277759035153,
+      "learning_rate": 2.7679936866773315e-05,
+      "loss": 0.3406,
+      "step": 2650
+    },
+    {
+      "epoch": 0.6826158353994268,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8528025144054479,
+      "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": 0.35067471861839294,
+      "eval_runtime": 17.4879,
+      "eval_samples_per_second": 2.516,
+      "eval_steps_per_second": 0.172,
+      "step": 2650
+    },
+    {
+      "epoch": 0.6828734262807097,
+      "grad_norm": 0.14616004490769807,
+      "learning_rate": 2.7639705412249183e-05,
+      "loss": 0.3348,
+      "step": 2651
+    },
+    {
+      "epoch": 0.6831310171619924,
+      "grad_norm": 0.20226492109006747,
+      "learning_rate": 2.759949204528844e-05,
+      "loss": 0.4407,
+      "step": 2652
+    },
+    {
+      "epoch": 0.6833886080432753,
+      "grad_norm": 0.19134254631359318,
+      "learning_rate": 2.7559296798420237e-05,
+      "loss": 0.4001,
+      "step": 2653
+    },
+    {
+      "epoch": 0.683646198924558,
+      "grad_norm": 0.152317104100875,
+      "learning_rate": 2.7519119704159052e-05,
+      "loss": 0.403,
+      "step": 2654
+    },
+    {
+      "epoch": 0.6839037898058409,
+      "grad_norm": 0.15170475473427295,
+      "learning_rate": 2.7478960795004788e-05,
+      "loss": 0.3308,
+      "step": 2655
+    },
+    {
+      "epoch": 0.6839037898058409,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8780487804878049,
+      "eval_PRM F1 AUC": 0.7380827658459927,
+      "eval_PRM F1 AUC (fixed)": 0.855159769512834,
+      "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": 0.34659090638160706,
+      "eval_runtime": 17.5469,
+      "eval_samples_per_second": 2.508,
+      "eval_steps_per_second": 0.171,
+      "step": 2655
+    },
+    {
+      "epoch": 0.6841613806871236,
+      "grad_norm": 0.17995207857884746,
+      "learning_rate": 2.743882010344251e-05,
+      "loss": 0.3558,
+      "step": 2656
+    },
+    {
+      "epoch": 0.6844189715684065,
+      "grad_norm": 0.12577112267556753,
+      "learning_rate": 2.739869766194263e-05,
+      "loss": 0.3235,
+      "step": 2657
+    },
+    {
+      "epoch": 0.6846765624496893,
+      "grad_norm": 0.13602194909859155,
+      "learning_rate": 2.7358593502960727e-05,
+      "loss": 0.3663,
+      "step": 2658
+    },
+    {
+      "epoch": 0.6849341533309721,
+      "grad_norm": 0.14515515627148598,
+      "learning_rate": 2.7318507658937665e-05,
+      "loss": 0.2859,
+      "step": 2659
+    },
+    {
+      "epoch": 0.6851917442122549,
+      "grad_norm": 0.13895407389483255,
+      "learning_rate": 2.7278440162299467e-05,
+      "loss": 0.3585,
+      "step": 2660
+    },
+    {
+      "epoch": 0.6851917442122549,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8780487804878049,
+      "eval_PRM F1 AUC": 0.7380827658459927,
+      "eval_PRM F1 AUC (fixed)": 0.854635935044526,
+      "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": 0.3507190942764282,
+      "eval_runtime": 16.9443,
+      "eval_samples_per_second": 2.597,
+      "eval_steps_per_second": 0.177,
+      "step": 2660
+    },
+    {
+      "epoch": 0.6854493350935377,
+      "grad_norm": 0.17412406091282298,
+      "learning_rate": 2.7238391045457272e-05,
+      "loss": 0.3436,
+      "step": 2661
+    },
+    {
+      "epoch": 0.6857069259748205,
+      "grad_norm": 0.1607797193978022,
+      "learning_rate": 2.7198360340807404e-05,
+      "loss": 0.3526,
+      "step": 2662
+    },
+    {
+      "epoch": 0.6859645168561033,
+      "grad_norm": 0.22489188732297646,
+      "learning_rate": 2.715834808073127e-05,
+      "loss": 0.377,
+      "step": 2663
+    },
+    {
+      "epoch": 0.6862221077373861,
+      "grad_norm": 0.142881788056936,
+      "learning_rate": 2.7118354297595396e-05,
+      "loss": 0.3452,
+      "step": 2664
+    },
+    {
+      "epoch": 0.6864796986186689,
+      "grad_norm": 0.1412582378828825,
+      "learning_rate": 2.7078379023751265e-05,
+      "loss": 0.3499,
+      "step": 2665
+    },
+    {
+      "epoch": 0.6864796986186689,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8514929282346779,
+      "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": 17.1288,
+      "eval_samples_per_second": 2.569,
+      "eval_steps_per_second": 0.175,
+      "step": 2665
+    },
+    {
+      "epoch": 0.6867372894999517,
+      "grad_norm": 0.13549520792037545,
+      "learning_rate": 2.7038422291535505e-05,
+      "loss": 0.3031,
+      "step": 2666
+    },
+    {
+      "epoch": 0.6869948803812345,
+      "grad_norm": 0.12068220882639444,
+      "learning_rate": 2.6998484133269652e-05,
+      "loss": 0.3199,
+      "step": 2667
+    },
+    {
+      "epoch": 0.6872524712625173,
+      "grad_norm": 0.17229188940030932,
+      "learning_rate": 2.6958564581260264e-05,
+      "loss": 0.3949,
+      "step": 2668
+    },
+    {
+      "epoch": 0.6875100621438001,
+      "grad_norm": 0.16154767035065437,
+      "learning_rate": 2.691866366779887e-05,
+      "loss": 0.3745,
+      "step": 2669
+    },
+    {
+      "epoch": 0.6877676530250829,
+      "grad_norm": 0.12551094272328808,
+      "learning_rate": 2.6878781425161858e-05,
+      "loss": 0.29,
+      "step": 2670
+    },
+    {
+      "epoch": 0.6877676530250829,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.853850183342064,
+      "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": 0.35524681210517883,
+      "eval_runtime": 17.4355,
+      "eval_samples_per_second": 2.524,
+      "eval_steps_per_second": 0.172,
+      "step": 2670
+    },
+    {
+      "epoch": 0.6880252439063658,
+      "grad_norm": 0.19720322634325504,
+      "learning_rate": 2.683891788561055e-05,
+      "loss": 0.4223,
+      "step": 2671
+    },
+    {
+      "epoch": 0.6882828347876485,
+      "grad_norm": 0.15366709765352818,
+      "learning_rate": 2.6799073081391147e-05,
+      "loss": 0.3688,
+      "step": 2672
+    },
+    {
+      "epoch": 0.6885404256689314,
+      "grad_norm": 0.1882979421870251,
+      "learning_rate": 2.675924704473469e-05,
+      "loss": 0.3844,
+      "step": 2673
+    },
+    {
+      "epoch": 0.6887980165502141,
+      "grad_norm": 0.16691711318261465,
+      "learning_rate": 2.6719439807857033e-05,
+      "loss": 0.3684,
+      "step": 2674
+    },
+    {
+      "epoch": 0.689055607431497,
+      "grad_norm": 0.15179654874853052,
+      "learning_rate": 2.6679651402958768e-05,
+      "loss": 0.3344,
+      "step": 2675
+    },
+    {
+      "epoch": 0.689055607431497,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8522786799371399,
+      "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": 0.3511185944080353,
+      "eval_runtime": 17.0406,
+      "eval_samples_per_second": 2.582,
+      "eval_steps_per_second": 0.176,
+      "step": 2675
+    },
+    {
+      "epoch": 0.6893131983127797,
+      "grad_norm": 0.1382970160096718,
+      "learning_rate": 2.6639881862225385e-05,
+      "loss": 0.3712,
+      "step": 2676
+    },
+    {
+      "epoch": 0.6895707891940626,
+      "grad_norm": 0.1568546250266237,
+      "learning_rate": 2.6600131217826985e-05,
+      "loss": 0.418,
+      "step": 2677
+    },
+    {
+      "epoch": 0.6898283800753453,
+      "grad_norm": 0.1286400446233427,
+      "learning_rate": 2.6560399501918465e-05,
+      "loss": 0.3097,
+      "step": 2678
+    },
+    {
+      "epoch": 0.6900859709566282,
+      "grad_norm": 0.1658972291537468,
+      "learning_rate": 2.6520686746639357e-05,
+      "loss": 0.337,
+      "step": 2679
+    },
+    {
+      "epoch": 0.6903435618379109,
+      "grad_norm": 0.1245252072082894,
+      "learning_rate": 2.6480992984113877e-05,
+      "loss": 0.3095,
+      "step": 2680
+    },
+    {
+      "epoch": 0.6903435618379109,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8517548454688318,
+      "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": 17.3116,
+      "eval_samples_per_second": 2.542,
+      "eval_steps_per_second": 0.173,
+      "step": 2680
+    },
+    {
+      "epoch": 0.6906011527191938,
+      "grad_norm": 0.1515250169641165,
+      "learning_rate": 2.644131824645092e-05,
+      "loss": 0.3129,
+      "step": 2681
+    },
+    {
+      "epoch": 0.6908587436004765,
+      "grad_norm": 0.16763732918130125,
+      "learning_rate": 2.640166256574389e-05,
+      "loss": 0.3398,
+      "step": 2682
+    },
+    {
+      "epoch": 0.6911163344817594,
+      "grad_norm": 0.1382463052875121,
+      "learning_rate": 2.6362025974070902e-05,
+      "loss": 0.3468,
+      "step": 2683
+    },
+    {
+      "epoch": 0.6913739253630421,
+      "grad_norm": 0.1953974501965342,
+      "learning_rate": 2.6322408503494482e-05,
+      "loss": 0.3896,
+      "step": 2684
+    },
+    {
+      "epoch": 0.691631516244325,
+      "grad_norm": 0.1260736126806096,
+      "learning_rate": 2.6282810186061862e-05,
+      "loss": 0.2979,
+      "step": 2685
+    },
+    {
+      "epoch": 0.691631516244325,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8541121005762178,
+      "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": 0.34854403138160706,
+      "eval_runtime": 17.6506,
+      "eval_samples_per_second": 2.493,
+      "eval_steps_per_second": 0.17,
+      "step": 2685
+    },
+    {
+      "epoch": 0.6918891071256078,
+      "grad_norm": 0.2205195541459199,
+      "learning_rate": 2.624323105380463e-05,
+      "loss": 0.4454,
+      "step": 2686
+    },
+    {
+      "epoch": 0.6921466980068905,
+      "grad_norm": 0.1661657499937164,
+      "learning_rate": 2.6203671138738944e-05,
+      "loss": 0.4094,
+      "step": 2687
+    },
+    {
+      "epoch": 0.6924042888881734,
+      "grad_norm": 0.1811939366544613,
+      "learning_rate": 2.616413047286536e-05,
+      "loss": 0.3772,
+      "step": 2688
+    },
+    {
+      "epoch": 0.6926618797694561,
+      "grad_norm": 0.18520135079202824,
+      "learning_rate": 2.6124609088168905e-05,
+      "loss": 0.3781,
+      "step": 2689
+    },
+    {
+      "epoch": 0.692919470650739,
+      "grad_norm": 0.15492756717661885,
+      "learning_rate": 2.6085107016619015e-05,
+      "loss": 0.3544,
+      "step": 2690
+    },
+    {
+      "epoch": 0.692919470650739,
+      "eval_PRM Accuracy": 0.7924528301886793,
+      "eval_PRM F1": 0.8641975308641975,
+      "eval_PRM F1 AUC": 0.7260345730749084,
+      "eval_PRM F1 AUC (fixed)": 0.8522786799371399,
+      "eval_PRM F1 Neg": 0.56,
+      "eval_PRM NPV": 0.5185185185185185,
+      "eval_PRM Precision": 0.8860759493670886,
+      "eval_PRM Recall": 0.8433734939759037,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.34854403138160706,
+      "eval_runtime": 17.0113,
+      "eval_samples_per_second": 2.587,
+      "eval_steps_per_second": 0.176,
+      "step": 2690
+    },
+    {
+      "epoch": 0.6931770615320217,
+      "grad_norm": 0.22411801493006525,
+      "learning_rate": 2.604562429016944e-05,
+      "loss": 0.4038,
+      "step": 2691
+    },
+    {
+      "epoch": 0.6934346524133046,
+      "grad_norm": 0.1420288648821109,
+      "learning_rate": 2.600616094075835e-05,
+      "loss": 0.3692,
+      "step": 2692
+    },
+    {
+      "epoch": 0.6936922432945873,
+      "grad_norm": 0.1423341400071091,
+      "learning_rate": 2.5966717000308228e-05,
+      "loss": 0.3552,
+      "step": 2693
+    },
+    {
+      "epoch": 0.6939498341758702,
+      "grad_norm": 0.14850724416073355,
+      "learning_rate": 2.5927292500725807e-05,
+      "loss": 0.389,
+      "step": 2694
+    },
+    {
+      "epoch": 0.6942074250571529,
+      "grad_norm": 0.13737435005751336,
+      "learning_rate": 2.5887887473902182e-05,
+      "loss": 0.3552,
+      "step": 2695
+    },
+    {
+      "epoch": 0.6942074250571529,
+      "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": 0.35329368710517883,
+      "eval_runtime": 17.2152,
+      "eval_samples_per_second": 2.556,
+      "eval_steps_per_second": 0.174,
+      "step": 2695
+    },
+    {
+      "epoch": 0.6944650159384358,
+      "grad_norm": 0.15195358397270992,
+      "learning_rate": 2.584850195171259e-05,
+      "loss": 0.4059,
+      "step": 2696
+    },
+    {
+      "epoch": 0.6947226068197185,
+      "grad_norm": 0.13920732343492914,
+      "learning_rate": 2.5809135966016562e-05,
+      "loss": 0.3382,
+      "step": 2697
+    },
+    {
+      "epoch": 0.6949801977010014,
+      "grad_norm": 0.1516401245351728,
+      "learning_rate": 2.576978954865783e-05,
+      "loss": 0.3832,
+      "step": 2698
+    },
+    {
+      "epoch": 0.6952377885822841,
+      "grad_norm": 0.16267808720085594,
+      "learning_rate": 2.5730462731464273e-05,
+      "loss": 0.3678,
+      "step": 2699
+    },
+    {
+      "epoch": 0.695495379463567,
+      "grad_norm": 0.1539062210313814,
+      "learning_rate": 2.569115554624789e-05,
+      "loss": 0.3503,
+      "step": 2700
+    },
+    {
+      "epoch": 0.695495379463567,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8780487804878049,
+      "eval_PRM F1 AUC": 0.7380827658459927,
+      "eval_PRM F1 AUC (fixed)": 0.854112100576218,
+      "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": 0.3552911877632141,
+      "eval_runtime": 17.454,
+      "eval_samples_per_second": 2.521,
+      "eval_steps_per_second": 0.172,
+      "step": 2700
+    },
+    {
+      "epoch": 0.6957529703448498,
+      "grad_norm": 0.13893228245162673,
+      "learning_rate": 2.5651868024804848e-05,
+      "loss": 0.3741,
+      "step": 2701
+    },
+    {
+      "epoch": 0.6960105612261326,
+      "grad_norm": 0.13438617071079145,
+      "learning_rate": 2.5612600198915404e-05,
+      "loss": 0.3658,
+      "step": 2702
+    },
+    {
+      "epoch": 0.6962681521074154,
+      "grad_norm": 0.15143362391177137,
+      "learning_rate": 2.557335210034383e-05,
+      "loss": 0.3986,
+      "step": 2703
+    },
+    {
+      "epoch": 0.6965257429886982,
+      "grad_norm": 0.135959890874589,
+      "learning_rate": 2.553412376083851e-05,
+      "loss": 0.3645,
+      "step": 2704
+    },
+    {
+      "epoch": 0.696783333869981,
+      "grad_norm": 0.2628741371875413,
+      "learning_rate": 2.5494915212131765e-05,
+      "loss": 0.4327,
+      "step": 2705
+    },
+    {
+      "epoch": 0.696783333869981,
+      "eval_PRM Accuracy": 0.7924528301886793,
+      "eval_PRM F1": 0.8641975308641975,
+      "eval_PRM F1 AUC": 0.7260345730749084,
+      "eval_PRM F1 AUC (fixed)": 0.8446830801466737,
+      "eval_PRM F1 Neg": 0.56,
+      "eval_PRM NPV": 0.5185185185185185,
+      "eval_PRM Precision": 0.8860759493670886,
+      "eval_PRM Recall": 0.8433734939759037,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35684481263160706,
+      "eval_runtime": 17.2321,
+      "eval_samples_per_second": 2.553,
+      "eval_steps_per_second": 0.174,
+      "step": 2705
+    },
+    {
+      "epoch": 0.6970409247512638,
+      "grad_norm": 0.17023455064399243,
+      "learning_rate": 2.5455726485940012e-05,
+      "loss": 0.3763,
+      "step": 2706
+    },
+    {
+      "epoch": 0.6972985156325466,
+      "grad_norm": 0.1696502975632842,
+      "learning_rate": 2.541655761396355e-05,
+      "loss": 0.4026,
+      "step": 2707
+    },
+    {
+      "epoch": 0.6975561065138294,
+      "grad_norm": 0.19135637232683286,
+      "learning_rate": 2.537740862788662e-05,
+      "loss": 0.36,
+      "step": 2708
+    },
+    {
+      "epoch": 0.6978136973951122,
+      "grad_norm": 0.1941420190978459,
+      "learning_rate": 2.5338279559377408e-05,
+      "loss": 0.4501,
+      "step": 2709
+    },
+    {
+      "epoch": 0.698071288276395,
+      "grad_norm": 0.18007504455414605,
+      "learning_rate": 2.5299170440087993e-05,
+      "loss": 0.3291,
+      "step": 2710
+    },
+    {
+      "epoch": 0.698071288276395,
+      "eval_PRM Accuracy": 0.7924528301886793,
+      "eval_PRM F1": 0.8641975308641975,
+      "eval_PRM F1 AUC": 0.7260345730749084,
+      "eval_PRM F1 AUC (fixed)": 0.8459926663174437,
+      "eval_PRM F1 Neg": 0.56,
+      "eval_PRM NPV": 0.5185185185185185,
+      "eval_PRM Precision": 0.8860759493670886,
+      "eval_PRM Recall": 0.8433734939759037,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35666725039482117,
+      "eval_runtime": 17.4532,
+      "eval_samples_per_second": 2.521,
+      "eval_steps_per_second": 0.172,
+      "step": 2710
+    },
+    {
+      "epoch": 0.6983288791576778,
+      "grad_norm": 0.14401464521530855,
+      "learning_rate": 2.526008130165432e-05,
+      "loss": 0.4013,
+      "step": 2711
+    },
+    {
+      "epoch": 0.6985864700389606,
+      "grad_norm": 0.13094302556714996,
+      "learning_rate": 2.522101217569611e-05,
+      "loss": 0.3185,
+      "step": 2712
+    },
+    {
+      "epoch": 0.6988440609202434,
+      "grad_norm": 0.17431301504577795,
+      "learning_rate": 2.5181963093816962e-05,
+      "loss": 0.4072,
+      "step": 2713
+    },
+    {
+      "epoch": 0.6991016518015263,
+      "grad_norm": 0.1319443611792694,
+      "learning_rate": 2.514293408760426e-05,
+      "loss": 0.3078,
+      "step": 2714
+    },
+    {
+      "epoch": 0.699359242682809,
+      "grad_norm": 0.16947826226563384,
+      "learning_rate": 2.5103925188629084e-05,
+      "loss": 0.4262,
+      "step": 2715
+    },
+    {
+      "epoch": 0.699359242682809,
+      "eval_PRM Accuracy": 0.8018867924528302,
+      "eval_PRM F1": 0.8711656441717791,
+      "eval_PRM F1 AUC": 0.7320586694604505,
+      "eval_PRM F1 AUC (fixed)": 0.8438973284442116,
+      "eval_PRM F1 Neg": 0.5714285714285714,
+      "eval_PRM NPV": 0.5384615384615384,
+      "eval_PRM Precision": 0.8875,
+      "eval_PRM Recall": 0.8554216867469879,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3564896881580353,
+      "eval_runtime": 17.5267,
+      "eval_samples_per_second": 2.51,
+      "eval_steps_per_second": 0.171,
+      "step": 2715
+    },
+    {
+      "epoch": 0.6996168335640919,
+      "grad_norm": 0.16059942318850942,
+      "learning_rate": 2.506493642844634e-05,
+      "loss": 0.3947,
+      "step": 2716
+    },
+    {
+      "epoch": 0.6998744244453746,
+      "grad_norm": 0.1413370424804344,
+      "learning_rate": 2.5025967838594548e-05,
+      "loss": 0.309,
+      "step": 2717
+    },
+    {
+      "epoch": 0.7001320153266575,
+      "grad_norm": 0.1575776946224474,
+      "learning_rate": 2.498701945059599e-05,
+      "loss": 0.3426,
+      "step": 2718
+    },
+    {
+      "epoch": 0.7003896062079402,
+      "grad_norm": 0.12915501393757087,
+      "learning_rate": 2.4948091295956566e-05,
+      "loss": 0.3574,
+      "step": 2719
+    },
+    {
+      "epoch": 0.7006471970892231,
+      "grad_norm": 0.11879255785567668,
+      "learning_rate": 2.4909183406165836e-05,
+      "loss": 0.3417,
+      "step": 2720
+    },
+    {
+      "epoch": 0.7006471970892231,
+      "eval_PRM Accuracy": 0.8018867924528302,
+      "eval_PRM F1": 0.8711656441717791,
+      "eval_PRM F1 AUC": 0.7320586694604505,
+      "eval_PRM F1 AUC (fixed)": 0.8410162388685175,
+      "eval_PRM F1 Neg": 0.5714285714285714,
+      "eval_PRM NPV": 0.5384615384615384,
+      "eval_PRM Precision": 0.8875,
+      "eval_PRM Recall": 0.8554216867469879,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3572443127632141,
+      "eval_runtime": 17.4714,
+      "eval_samples_per_second": 2.518,
+      "eval_steps_per_second": 0.172,
+      "step": 2720
+    },
+    {
+      "epoch": 0.7009047879705058,
+      "grad_norm": 0.14641264953618238,
+      "learning_rate": 2.4870295812696926e-05,
+      "loss": 0.3337,
+      "step": 2721
+    },
+    {
+      "epoch": 0.7011623788517887,
+      "grad_norm": 0.2179540498024203,
+      "learning_rate": 2.483142854700658e-05,
+      "loss": 0.3628,
+      "step": 2722
+    },
+    {
+      "epoch": 0.7014199697330714,
+      "grad_norm": 0.14540266069521307,
+      "learning_rate": 2.4792581640535105e-05,
+      "loss": 0.3378,
+      "step": 2723
+    },
+    {
+      "epoch": 0.7016775606143543,
+      "grad_norm": 0.16549416288759555,
+      "learning_rate": 2.4753755124706286e-05,
+      "loss": 0.3234,
+      "step": 2724
+    },
+    {
+      "epoch": 0.701935151495637,
+      "grad_norm": 0.16311065413664966,
+      "learning_rate": 2.4714949030927487e-05,
+      "loss": 0.3901,
+      "step": 2725
+    },
+    {
+      "epoch": 0.701935151495637,
+      "eval_PRM Accuracy": 0.8018867924528302,
+      "eval_PRM F1": 0.8727272727272727,
+      "eval_PRM F1 AUC": 0.7163436354112102,
+      "eval_PRM F1 AUC (fixed)": 0.8454688318491357,
+      "eval_PRM F1 Neg": 0.5531914893617021,
+      "eval_PRM NPV": 0.5416666666666666,
+      "eval_PRM Precision": 0.8780487804878049,
+      "eval_PRM Recall": 0.8674698795180723,
+      "eval_PRM Specificty": 0.5652173913043478,
+      "eval_loss": 0.35768821835517883,
+      "eval_runtime": 17.0492,
+      "eval_samples_per_second": 2.581,
+      "eval_steps_per_second": 0.176,
+      "step": 2725
+    },
+    {
+      "epoch": 0.7021927423769199,
+      "grad_norm": 0.14000386797128103,
+      "learning_rate": 2.467616339058945e-05,
+      "loss": 0.329,
+      "step": 2726
+    },
+    {
+      "epoch": 0.7024503332582026,
+      "grad_norm": 0.16026256337928413,
+      "learning_rate": 2.4637398235066527e-05,
+      "loss": 0.3949,
+      "step": 2727
+    },
+    {
+      "epoch": 0.7027079241394855,
+      "grad_norm": 0.15627168101176667,
+      "learning_rate": 2.4598653595716365e-05,
+      "loss": 0.3861,
+      "step": 2728
+    },
+    {
+      "epoch": 0.7029655150207683,
+      "grad_norm": 0.13178721726537074,
+      "learning_rate": 2.455992950388003e-05,
+      "loss": 0.3664,
+      "step": 2729
+    },
+    {
+      "epoch": 0.7032231059020511,
+      "grad_norm": 0.14552517768960338,
+      "learning_rate": 2.452122599088203e-05,
+      "loss": 0.2891,
+      "step": 2730
+    },
+    {
+      "epoch": 0.7032231059020511,
+      "eval_PRM Accuracy": 0.7924528301886793,
+      "eval_PRM F1": 0.8658536585365854,
+      "eval_PRM F1 AUC": 0.710319539025668,
+      "eval_PRM F1 AUC (fixed)": 0.8438973284442117,
+      "eval_PRM F1 Neg": 0.5416666666666666,
+      "eval_PRM NPV": 0.52,
+      "eval_PRM Precision": 0.8765432098765432,
+      "eval_PRM Recall": 0.8554216867469879,
+      "eval_PRM Specificty": 0.5652173913043478,
+      "eval_loss": 0.3614612817764282,
+      "eval_runtime": 16.9344,
+      "eval_samples_per_second": 2.598,
+      "eval_steps_per_second": 0.177,
+      "step": 2730
+    },
+    {
+      "epoch": 0.7034806967833339,
+      "grad_norm": 0.14414489002933717,
+      "learning_rate": 2.4482543088030185e-05,
+      "loss": 0.4093,
+      "step": 2731
+    },
+    {
+      "epoch": 0.7037382876646167,
+      "grad_norm": 0.17593344705568692,
+      "learning_rate": 2.4443880826615665e-05,
+      "loss": 0.4037,
+      "step": 2732
+    },
+    {
+      "epoch": 0.7039958785458995,
+      "grad_norm": 0.1223001057903308,
+      "learning_rate": 2.44052392379129e-05,
+      "loss": 0.3561,
+      "step": 2733
+    },
+    {
+      "epoch": 0.7042534694271823,
+      "grad_norm": 0.17008409291662235,
+      "learning_rate": 2.4366618353179644e-05,
+      "loss": 0.3795,
+      "step": 2734
+    },
+    {
+      "epoch": 0.7045110603084651,
+      "grad_norm": 0.16434909190618796,
+      "learning_rate": 2.43280182036569e-05,
+      "loss": 0.3797,
+      "step": 2735
+    },
+    {
+      "epoch": 0.7045110603084651,
+      "eval_PRM Accuracy": 0.8018867924528302,
+      "eval_PRM F1": 0.8711656441717791,
+      "eval_PRM F1 AUC": 0.7320586694604505,
+      "eval_PRM F1 AUC (fixed)": 0.8415400733368257,
+      "eval_PRM F1 Neg": 0.5714285714285714,
+      "eval_PRM NPV": 0.5384615384615384,
+      "eval_PRM Precision": 0.8875,
+      "eval_PRM Recall": 0.8554216867469879,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3566228747367859,
+      "eval_runtime": 17.0532,
+      "eval_samples_per_second": 2.58,
+      "eval_steps_per_second": 0.176,
+      "step": 2735
+    },
+    {
+      "epoch": 0.7047686511897479,
+      "grad_norm": 0.13000640637875388,
+      "learning_rate": 2.4289438820568845e-05,
+      "loss": 0.3139,
+      "step": 2736
+    },
+    {
+      "epoch": 0.7050262420710307,
+      "grad_norm": 0.15029391695982908,
+      "learning_rate": 2.425088023512293e-05,
+      "loss": 0.2792,
+      "step": 2737
+    },
+    {
+      "epoch": 0.7052838329523134,
+      "grad_norm": 0.22057161618053767,
+      "learning_rate": 2.421234247850972e-05,
+      "loss": 0.3896,
+      "step": 2738
+    },
+    {
+      "epoch": 0.7055414238335963,
+      "grad_norm": 0.18522255265204693,
+      "learning_rate": 2.4173825581902975e-05,
+      "loss": 0.3746,
+      "step": 2739
+    },
+    {
+      "epoch": 0.705799014714879,
+      "grad_norm": 0.1937399712988186,
+      "learning_rate": 2.4135329576459563e-05,
+      "loss": 0.3207,
+      "step": 2740
+    },
+    {
+      "epoch": 0.705799014714879,
+      "eval_PRM Accuracy": 0.8018867924528302,
+      "eval_PRM F1": 0.8711656441717791,
+      "eval_PRM F1 AUC": 0.7320586694604505,
+      "eval_PRM F1 AUC (fixed)": 0.8407543216343635,
+      "eval_PRM F1 Neg": 0.5714285714285714,
+      "eval_PRM NPV": 0.5384615384615384,
+      "eval_PRM Precision": 0.8875,
+      "eval_PRM Recall": 0.8554216867469879,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35768821835517883,
+      "eval_runtime": 17.2614,
+      "eval_samples_per_second": 2.549,
+      "eval_steps_per_second": 0.174,
+      "step": 2740
+    },
+    {
+      "epoch": 0.7060566055961619,
+      "grad_norm": 0.15836049317458623,
+      "learning_rate": 2.4096854493319477e-05,
+      "loss": 0.379,
+      "step": 2741
+    },
+    {
+      "epoch": 0.7063141964774446,
+      "grad_norm": 0.2390188072952708,
+      "learning_rate": 2.4058400363605728e-05,
+      "loss": 0.453,
+      "step": 2742
+    },
+    {
+      "epoch": 0.7065717873587275,
+      "grad_norm": 0.17286721909595393,
+      "learning_rate": 2.4019967218424427e-05,
+      "loss": 0.3891,
+      "step": 2743
+    },
+    {
+      "epoch": 0.7068293782400104,
+      "grad_norm": 0.15192179987331367,
+      "learning_rate": 2.398155508886472e-05,
+      "loss": 0.3095,
+      "step": 2744
+    },
+    {
+      "epoch": 0.7070869691212931,
+      "grad_norm": 0.16477307850149672,
+      "learning_rate": 2.3943164005998697e-05,
+      "loss": 0.384,
+      "step": 2745
+    },
+    {
+      "epoch": 0.7070869691212931,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8780487804878049,
+      "eval_PRM F1 AUC": 0.7380827658459927,
+      "eval_PRM F1 AUC (fixed)": 0.8394447354635935,
+      "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": 0.36208274960517883,
+      "eval_runtime": 17.5058,
+      "eval_samples_per_second": 2.513,
+      "eval_steps_per_second": 0.171,
+      "step": 2745
+    },
+    {
+      "epoch": 0.707344560002576,
+      "grad_norm": 0.16623502164431952,
+      "learning_rate": 2.3904794000881486e-05,
+      "loss": 0.4262,
+      "step": 2746
+    },
+    {
+      "epoch": 0.7076021508838587,
+      "grad_norm": 0.17309520719170254,
+      "learning_rate": 2.3866445104551098e-05,
+      "loss": 0.3891,
+      "step": 2747
+    },
+    {
+      "epoch": 0.7078597417651415,
+      "grad_norm": 0.14559163997947372,
+      "learning_rate": 2.3828117348028528e-05,
+      "loss": 0.3471,
+      "step": 2748
+    },
+    {
+      "epoch": 0.7081173326464243,
+      "grad_norm": 0.11142923579850252,
+      "learning_rate": 2.3789810762317667e-05,
+      "loss": 0.2911,
+      "step": 2749
+    },
+    {
+      "epoch": 0.7083749235277071,
+      "grad_norm": 0.13960134559798204,
+      "learning_rate": 2.375152537840522e-05,
+      "loss": 0.3982,
+      "step": 2750
+    },
+    {
+      "epoch": 0.7083749235277071,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.891566265060241,
+      "eval_PRM F1 AUC": 0.7501309586170771,
+      "eval_PRM F1 AUC (fixed)": 0.8446830801466737,
+      "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": 0.35799893736839294,
+      "eval_runtime": 17.8042,
+      "eval_samples_per_second": 2.471,
+      "eval_steps_per_second": 0.168,
+      "step": 2750
+    },
+    {
+      "epoch": 0.7086325144089899,
+      "grad_norm": 0.1454365178210285,
+      "learning_rate": 2.3713261227260812e-05,
+      "loss": 0.3578,
+      "step": 2751
+    },
+    {
+      "epoch": 0.7088901052902727,
+      "grad_norm": 0.13251829233355425,
+      "learning_rate": 2.3675018339836858e-05,
+      "loss": 0.3375,
+      "step": 2752
+    },
+    {
+      "epoch": 0.7091476961715555,
+      "grad_norm": 0.2148654472903786,
+      "learning_rate": 2.3636796747068595e-05,
+      "loss": 0.4192,
+      "step": 2753
+    },
+    {
+      "epoch": 0.7094052870528383,
+      "grad_norm": 0.14156333761022905,
+      "learning_rate": 2.3598596479874008e-05,
+      "loss": 0.3641,
+      "step": 2754
+    },
+    {
+      "epoch": 0.7096628779341211,
+      "grad_norm": 0.14510469467718043,
+      "learning_rate": 2.3560417569153796e-05,
+      "loss": 0.3197,
+      "step": 2755
+    },
+    {
+      "epoch": 0.7096628779341211,
+      "eval_PRM Accuracy": 0.8018867924528302,
+      "eval_PRM F1": 0.8711656441717791,
+      "eval_PRM F1 AUC": 0.7320586694604505,
+      "eval_PRM F1 AUC (fixed)": 0.8493975903614458,
+      "eval_PRM F1 Neg": 0.5714285714285714,
+      "eval_PRM NPV": 0.5384615384615384,
+      "eval_PRM Precision": 0.8875,
+      "eval_PRM Recall": 0.8554216867469879,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.34792259335517883,
+      "eval_runtime": 17.5069,
+      "eval_samples_per_second": 2.513,
+      "eval_steps_per_second": 0.171,
+      "step": 2755
+    },
+    {
+      "epoch": 0.7099204688154039,
+      "grad_norm": 0.14240734386176662,
+      "learning_rate": 2.352226004579151e-05,
+      "loss": 0.4157,
+      "step": 2756
+    },
+    {
+      "epoch": 0.7101780596966868,
+      "grad_norm": 0.1502415215795503,
+      "learning_rate": 2.3484123940653253e-05,
+      "loss": 0.361,
+      "step": 2757
+    },
+    {
+      "epoch": 0.7104356505779695,
+      "grad_norm": 0.14913094528465853,
+      "learning_rate": 2.3446009284587915e-05,
+      "loss": 0.3628,
+      "step": 2758
+    },
+    {
+      "epoch": 0.7106932414592524,
+      "grad_norm": 0.1393740635701928,
+      "learning_rate": 2.3407916108426946e-05,
+      "loss": 0.3102,
+      "step": 2759
+    },
+    {
+      "epoch": 0.7109508323405351,
+      "grad_norm": 0.15461891812931783,
+      "learning_rate": 2.3369844442984484e-05,
+      "loss": 0.3578,
+      "step": 2760
+    },
+    {
+      "epoch": 0.7109508323405351,
+      "eval_PRM Accuracy": 0.8018867924528302,
+      "eval_PRM F1": 0.8711656441717791,
+      "eval_PRM F1 AUC": 0.7320586694604505,
+      "eval_PRM F1 AUC (fixed)": 0.8473022524882137,
+      "eval_PRM F1 Neg": 0.5714285714285714,
+      "eval_PRM NPV": 0.5384615384615384,
+      "eval_PRM Precision": 0.8875,
+      "eval_PRM Recall": 0.8554216867469879,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.34561434388160706,
+      "eval_runtime": 16.6556,
+      "eval_samples_per_second": 2.642,
+      "eval_steps_per_second": 0.18,
+      "step": 2760
+    },
+    {
+      "epoch": 0.711208423221818,
+      "grad_norm": 0.1620128221439273,
+      "learning_rate": 2.333179431905726e-05,
+      "loss": 0.3858,
+      "step": 2761
+    },
+    {
+      "epoch": 0.7114660141031007,
+      "grad_norm": 0.12484716096081154,
+      "learning_rate": 2.3293765767424537e-05,
+      "loss": 0.3949,
+      "step": 2762
+    },
+    {
+      "epoch": 0.7117236049843836,
+      "grad_norm": 0.12417823636822578,
+      "learning_rate": 2.325575881884819e-05,
+      "loss": 0.3016,
+      "step": 2763
+    },
+    {
+      "epoch": 0.7119811958656663,
+      "grad_norm": 0.1603996671736155,
+      "learning_rate": 2.321777350407252e-05,
+      "loss": 0.339,
+      "step": 2764
+    },
+    {
+      "epoch": 0.7122387867469492,
+      "grad_norm": 0.1635755859000412,
+      "learning_rate": 2.317980985382448e-05,
+      "loss": 0.4179,
+      "step": 2765
+    },
+    {
+      "epoch": 0.7122387867469492,
+      "eval_PRM Accuracy": 0.8018867924528302,
+      "eval_PRM F1": 0.8711656441717791,
+      "eval_PRM F1 AUC": 0.7320586694604505,
+      "eval_PRM F1 AUC (fixed)": 0.85358826610791,
+      "eval_PRM F1 Neg": 0.5714285714285714,
+      "eval_PRM NPV": 0.5384615384615384,
+      "eval_PRM Precision": 0.8875,
+      "eval_PRM Recall": 0.8554216867469879,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3414417505264282,
+      "eval_runtime": 17.0303,
+      "eval_samples_per_second": 2.584,
+      "eval_steps_per_second": 0.176,
+      "step": 2765
+    },
+    {
+      "epoch": 0.7124963776282319,
+      "grad_norm": 0.1578545422391313,
+      "learning_rate": 2.3141867898813358e-05,
+      "loss": 0.34,
+      "step": 2766
+    },
+    {
+      "epoch": 0.7127539685095148,
+      "grad_norm": 0.19707829642740438,
+      "learning_rate": 2.310394766973097e-05,
+      "loss": 0.4205,
+      "step": 2767
+    },
+    {
+      "epoch": 0.7130115593907975,
+      "grad_norm": 0.16862116596731477,
+      "learning_rate": 2.306604919725151e-05,
+      "loss": 0.4188,
+      "step": 2768
+    },
+    {
+      "epoch": 0.7132691502720804,
+      "grad_norm": 0.1173747837754076,
+      "learning_rate": 2.3028172512031604e-05,
+      "loss": 0.3259,
+      "step": 2769
+    },
+    {
+      "epoch": 0.7135267411533631,
+      "grad_norm": 0.13781967857656147,
+      "learning_rate": 2.2990317644710274e-05,
+      "loss": 0.3662,
+      "step": 2770
+    },
+    {
+      "epoch": 0.7135267411533631,
+      "eval_PRM Accuracy": 0.8018867924528302,
+      "eval_PRM F1": 0.8711656441717791,
+      "eval_PRM F1 AUC": 0.7320586694604505,
+      "eval_PRM F1 AUC (fixed)": 0.8548978522786799,
+      "eval_PRM F1 Neg": 0.5714285714285714,
+      "eval_PRM NPV": 0.5384615384615384,
+      "eval_PRM Precision": 0.8875,
+      "eval_PRM Recall": 0.8554216867469879,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.34255149960517883,
+      "eval_runtime": 17.0691,
+      "eval_samples_per_second": 2.578,
+      "eval_steps_per_second": 0.176,
+      "step": 2770
+    },
+    {
+      "epoch": 0.713784332034646,
+      "grad_norm": 0.12840077547265583,
+      "learning_rate": 2.2952484625908827e-05,
+      "loss": 0.3207,
+      "step": 2771
+    },
+    {
+      "epoch": 0.7140419229159288,
+      "grad_norm": 0.13723383099874353,
+      "learning_rate": 2.2914673486230954e-05,
+      "loss": 0.4079,
+      "step": 2772
+    },
+    {
+      "epoch": 0.7142995137972116,
+      "grad_norm": 0.14335730554159018,
+      "learning_rate": 2.287688425626262e-05,
+      "loss": 0.3296,
+      "step": 2773
+    },
+    {
+      "epoch": 0.7145571046784944,
+      "grad_norm": 0.14139486566500542,
+      "learning_rate": 2.2839116966572105e-05,
+      "loss": 0.2804,
+      "step": 2774
+    },
+    {
+      "epoch": 0.7148146955597772,
+      "grad_norm": 0.16000352992328196,
+      "learning_rate": 2.280137164770989e-05,
+      "loss": 0.364,
+      "step": 2775
+    },
+    {
+      "epoch": 0.7148146955597772,
+      "eval_PRM Accuracy": 0.7924528301886793,
+      "eval_PRM F1": 0.8658536585365854,
+      "eval_PRM F1 AUC": 0.710319539025668,
+      "eval_PRM F1 AUC (fixed)": 0.85489785227868,
+      "eval_PRM F1 Neg": 0.5416666666666666,
+      "eval_PRM NPV": 0.52,
+      "eval_PRM Precision": 0.8765432098765432,
+      "eval_PRM Recall": 0.8554216867469879,
+      "eval_PRM Specificty": 0.5652173913043478,
+      "eval_loss": 0.34237393736839294,
+      "eval_runtime": 16.8733,
+      "eval_samples_per_second": 2.608,
+      "eval_steps_per_second": 0.178,
+      "step": 2775
+    },
+    {
+      "epoch": 0.71507228644106,
+      "grad_norm": 0.1426100036868392,
+      "learning_rate": 2.276364833020868e-05,
+      "loss": 0.3776,
+      "step": 2776
+    },
+    {
+      "epoch": 0.7153298773223428,
+      "grad_norm": 0.15906307930400226,
+      "learning_rate": 2.272594704458344e-05,
+      "loss": 0.3864,
+      "step": 2777
+    },
+    {
+      "epoch": 0.7155874682036256,
+      "grad_norm": 0.17996034084740364,
+      "learning_rate": 2.2688267821331278e-05,
+      "loss": 0.4076,
+      "step": 2778
+    },
+    {
+      "epoch": 0.7158450590849084,
+      "grad_norm": 0.1897824054477158,
+      "learning_rate": 2.2650610690931467e-05,
+      "loss": 0.4065,
+      "step": 2779
+    },
+    {
+      "epoch": 0.7161026499661912,
+      "grad_norm": 0.13218113662802577,
+      "learning_rate": 2.2612975683845372e-05,
+      "loss": 0.2961,
+      "step": 2780
+    },
+    {
+      "epoch": 0.7161026499661912,
+      "eval_PRM Accuracy": 0.7924528301886793,
+      "eval_PRM F1": 0.8658536585365854,
+      "eval_PRM F1 AUC": 0.710319539025668,
+      "eval_PRM F1 AUC (fixed)": 0.8522786799371398,
+      "eval_PRM F1 Neg": 0.5416666666666666,
+      "eval_PRM NPV": 0.52,
+      "eval_PRM Precision": 0.8765432098765432,
+      "eval_PRM Recall": 0.8554216867469879,
+      "eval_PRM Specificty": 0.5652173913043478,
+      "eval_loss": 0.3450372815132141,
+      "eval_runtime": 17.694,
+      "eval_samples_per_second": 2.487,
+      "eval_steps_per_second": 0.17,
+      "step": 2780
+    },
+    {
+      "epoch": 0.716360240847474,
+      "grad_norm": 0.1569939201335511,
+      "learning_rate": 2.2575362830516506e-05,
+      "loss": 0.3165,
+      "step": 2781
+    },
+    {
+      "epoch": 0.7166178317287568,
+      "grad_norm": 0.18097788321208738,
+      "learning_rate": 2.253777216137047e-05,
+      "loss": 0.4363,
+      "step": 2782
+    },
+    {
+      "epoch": 0.7168754226100396,
+      "grad_norm": 0.13620805598169097,
+      "learning_rate": 2.2500203706814856e-05,
+      "loss": 0.3661,
+      "step": 2783
+    },
+    {
+      "epoch": 0.7171330134913224,
+      "grad_norm": 0.10753286752324899,
+      "learning_rate": 2.246265749723936e-05,
+      "loss": 0.2783,
+      "step": 2784
+    },
+    {
+      "epoch": 0.7173906043726052,
+      "grad_norm": 0.1300316161236073,
+      "learning_rate": 2.2425133563015594e-05,
+      "loss": 0.2907,
+      "step": 2785
+    },
+    {
+      "epoch": 0.7173906043726052,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8862275449101796,
+      "eval_PRM F1 AUC": 0.7283918281822945,
+      "eval_PRM F1 AUC (fixed)": 0.854112100576218,
+      "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": 0.3551136255264282,
+      "eval_runtime": 17.4436,
+      "eval_samples_per_second": 2.522,
+      "eval_steps_per_second": 0.172,
+      "step": 2785
+    },
+    {
+      "epoch": 0.717648195253888,
+      "grad_norm": 0.1701872693075267,
+      "learning_rate": 2.238763193449729e-05,
+      "loss": 0.4144,
+      "step": 2786
+    },
+    {
+      "epoch": 0.7179057861351709,
+      "grad_norm": 0.1760300746742691,
+      "learning_rate": 2.2350152642019983e-05,
+      "loss": 0.3941,
+      "step": 2787
+    },
+    {
+      "epoch": 0.7181633770164536,
+      "grad_norm": 0.14581543549786966,
+      "learning_rate": 2.2312695715901266e-05,
+      "loss": 0.3625,
+      "step": 2788
+    },
+    {
+      "epoch": 0.7184209678977365,
+      "grad_norm": 0.24930305966943844,
+      "learning_rate": 2.2275261186440537e-05,
+      "loss": 0.4298,
+      "step": 2789
+    },
+    {
+      "epoch": 0.7186785587790192,
+      "grad_norm": 0.18957999994987207,
+      "learning_rate": 2.2237849083919142e-05,
+      "loss": 0.3462,
+      "step": 2790
+    },
+    {
+      "epoch": 0.7186785587790192,
+      "eval_PRM Accuracy": 0.7924528301886793,
+      "eval_PRM F1": 0.8658536585365854,
+      "eval_PRM F1 AUC": 0.710319539025668,
+      "eval_PRM F1 AUC (fixed)": 0.851754845468832,
+      "eval_PRM F1 Neg": 0.5416666666666666,
+      "eval_PRM NPV": 0.52,
+      "eval_PRM Precision": 0.8765432098765432,
+      "eval_PRM Recall": 0.8554216867469879,
+      "eval_PRM Specificty": 0.5652173913043478,
+      "eval_loss": 0.3500532805919647,
+      "eval_runtime": 17.4986,
+      "eval_samples_per_second": 2.514,
+      "eval_steps_per_second": 0.171,
+      "step": 2790
+    },
+    {
+      "epoch": 0.7189361496603021,
+      "grad_norm": 0.15553216172084486,
+      "learning_rate": 2.2200459438600297e-05,
+      "loss": 0.3787,
+      "step": 2791
+    },
+    {
+      "epoch": 0.7191937405415848,
+      "grad_norm": 0.1609548296303529,
+      "learning_rate": 2.216309228072897e-05,
+      "loss": 0.3803,
+      "step": 2792
+    },
+    {
+      "epoch": 0.7194513314228677,
+      "grad_norm": 0.21372436553220225,
+      "learning_rate": 2.2125747640532023e-05,
+      "loss": 0.3382,
+      "step": 2793
+    },
+    {
+      "epoch": 0.7197089223041504,
+      "grad_norm": 0.13384594375374534,
+      "learning_rate": 2.2088425548218065e-05,
+      "loss": 0.2806,
+      "step": 2794
+    },
+    {
+      "epoch": 0.7199665131854333,
+      "grad_norm": 0.14277732385378297,
+      "learning_rate": 2.2051126033977494e-05,
+      "loss": 0.34,
+      "step": 2795
+    },
+    {
+      "epoch": 0.7199665131854333,
+      "eval_PRM Accuracy": 0.7924528301886793,
+      "eval_PRM F1": 0.8641975308641975,
+      "eval_PRM F1 AUC": 0.7260345730749084,
+      "eval_PRM F1 AUC (fixed)": 0.8496595075955997,
+      "eval_PRM F1 Neg": 0.56,
+      "eval_PRM NPV": 0.5185185185185185,
+      "eval_PRM Precision": 0.8860759493670886,
+      "eval_PRM Recall": 0.8433734939759037,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3463689684867859,
+      "eval_runtime": 16.9121,
+      "eval_samples_per_second": 2.602,
+      "eval_steps_per_second": 0.177,
+      "step": 2795
+    },
+    {
+      "epoch": 0.720224104066716,
+      "grad_norm": 0.1341500531756553,
+      "learning_rate": 2.20138491279824e-05,
+      "loss": 0.3279,
+      "step": 2796
+    },
+    {
+      "epoch": 0.7204816949479989,
+      "grad_norm": 0.16899171892652634,
+      "learning_rate": 2.1976594860386597e-05,
+      "loss": 0.3483,
+      "step": 2797
+    },
+    {
+      "epoch": 0.7207392858292816,
+      "grad_norm": 0.14477035937310337,
+      "learning_rate": 2.1939363261325606e-05,
+      "loss": 0.2884,
+      "step": 2798
+    },
+    {
+      "epoch": 0.7209968767105644,
+      "grad_norm": 0.13349709589620268,
+      "learning_rate": 2.190215436091661e-05,
+      "loss": 0.3415,
+      "step": 2799
+    },
+    {
+      "epoch": 0.7212544675918473,
+      "grad_norm": 0.25768602565818954,
+      "learning_rate": 2.1864968189258434e-05,
+      "loss": 0.4377,
+      "step": 2800
+    },
+    {
+      "epoch": 0.7212544675918473,
+      "eval_PRM Accuracy": 0.7830188679245284,
+      "eval_PRM F1": 0.8571428571428571,
+      "eval_PRM F1 AUC": 0.7200104766893662,
+      "eval_PRM F1 AUC (fixed)": 0.8483499214248299,
+      "eval_PRM F1 Neg": 0.5490196078431373,
+      "eval_PRM NPV": 0.5,
+      "eval_PRM Precision": 0.8846153846153846,
+      "eval_PRM Recall": 0.8313253012048193,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3491654694080353,
+      "eval_runtime": 17.1243,
+      "eval_samples_per_second": 2.569,
+      "eval_steps_per_second": 0.175,
+      "step": 2800
+    },
+    {
+      "epoch": 0.72151205847313,
+      "grad_norm": 0.18235096954259772,
+      "learning_rate": 2.1827804776431476e-05,
+      "loss": 0.3697,
+      "step": 2801
+    },
+    {
+      "epoch": 0.7217696493544129,
+      "grad_norm": 0.22471544209079214,
+      "learning_rate": 2.1790664152497768e-05,
+      "loss": 0.3636,
+      "step": 2802
+    },
+    {
+      "epoch": 0.7220272402356956,
+      "grad_norm": 0.17327612172954443,
+      "learning_rate": 2.1753546347500918e-05,
+      "loss": 0.3078,
+      "step": 2803
+    },
+    {
+      "epoch": 0.7222848311169785,
+      "grad_norm": 0.24146631371380453,
+      "learning_rate": 2.1716451391466008e-05,
+      "loss": 0.4625,
+      "step": 2804
+    },
+    {
+      "epoch": 0.7225424219982612,
+      "grad_norm": 0.13098403629381128,
+      "learning_rate": 2.167937931439972e-05,
+      "loss": 0.3208,
+      "step": 2805
+    },
+    {
+      "epoch": 0.7225424219982612,
+      "eval_PRM Accuracy": 0.7830188679245284,
+      "eval_PRM F1": 0.8571428571428571,
+      "eval_PRM F1 AUC": 0.7200104766893662,
+      "eval_PRM F1 AUC (fixed)": 0.8480880041906758,
+      "eval_PRM F1 Neg": 0.5490196078431373,
+      "eval_PRM NPV": 0.5,
+      "eval_PRM Precision": 0.8846153846153846,
+      "eval_PRM Recall": 0.8313253012048193,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.34938743710517883,
+      "eval_runtime": 17.1567,
+      "eval_samples_per_second": 2.565,
+      "eval_steps_per_second": 0.175,
+      "step": 2805
+    },
+    {
+      "epoch": 0.7228000128795441,
+      "grad_norm": 0.1753570422091817,
+      "learning_rate": 2.1642330146290142e-05,
+      "loss": 0.4,
+      "step": 2806
+    },
+    {
+      "epoch": 0.7230576037608268,
+      "grad_norm": 0.132697904686546,
+      "learning_rate": 2.160530391710694e-05,
+      "loss": 0.3107,
+      "step": 2807
+    },
+    {
+      "epoch": 0.7233151946421097,
+      "grad_norm": 0.12347173853975807,
+      "learning_rate": 2.1568300656801126e-05,
+      "loss": 0.2349,
+      "step": 2808
+    },
+    {
+      "epoch": 0.7235727855233924,
+      "grad_norm": 0.17619796384247116,
+      "learning_rate": 2.153132039530516e-05,
+      "loss": 0.2991,
+      "step": 2809
+    },
+    {
+      "epoch": 0.7238303764046753,
+      "grad_norm": 0.19431505394055815,
+      "learning_rate": 2.1494363162532915e-05,
+      "loss": 0.376,
+      "step": 2810
+    },
+    {
+      "epoch": 0.7238303764046753,
+      "eval_PRM Accuracy": 0.8018867924528302,
+      "eval_PRM F1": 0.8711656441717791,
+      "eval_PRM F1 AUC": 0.7320586694604505,
+      "eval_PRM F1 AUC (fixed)": 0.8504452592980618,
+      "eval_PRM F1 Neg": 0.5714285714285714,
+      "eval_PRM NPV": 0.5384615384615384,
+      "eval_PRM Precision": 0.8875,
+      "eval_PRM Recall": 0.8554216867469879,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3519176244735718,
+      "eval_runtime": 17.0949,
+      "eval_samples_per_second": 2.574,
+      "eval_steps_per_second": 0.175,
+      "step": 2810
+    },
+    {
+      "epoch": 0.724087967285958,
+      "grad_norm": 0.1456938494015266,
+      "learning_rate": 2.1457428988379635e-05,
+      "loss": 0.3923,
+      "step": 2811
+    },
+    {
+      "epoch": 0.7243455581672409,
+      "grad_norm": 0.28893525407443255,
+      "learning_rate": 2.1420517902721914e-05,
+      "loss": 0.4272,
+      "step": 2812
+    },
+    {
+      "epoch": 0.7246031490485236,
+      "grad_norm": 0.16875241353218418,
+      "learning_rate": 2.1383629935417633e-05,
+      "loss": 0.3865,
+      "step": 2813
+    },
+    {
+      "epoch": 0.7248607399298065,
+      "grad_norm": 0.16407317749283473,
+      "learning_rate": 2.1346765116306007e-05,
+      "loss": 0.405,
+      "step": 2814
+    },
+    {
+      "epoch": 0.7251183308110893,
+      "grad_norm": 0.1699188422149344,
+      "learning_rate": 2.1309923475207537e-05,
+      "loss": 0.3776,
+      "step": 2815
+    },
+    {
+      "epoch": 0.7251183308110893,
+      "eval_PRM Accuracy": 0.7830188679245284,
+      "eval_PRM F1": 0.8571428571428571,
+      "eval_PRM F1 AUC": 0.7200104766893662,
+      "eval_PRM F1 AUC (fixed)": 0.8457307490832897,
+      "eval_PRM F1 Neg": 0.5490196078431373,
+      "eval_PRM NPV": 0.5,
+      "eval_PRM Precision": 0.8846153846153846,
+      "eval_PRM Recall": 0.8313253012048193,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35196200013160706,
+      "eval_runtime": 16.9334,
+      "eval_samples_per_second": 2.598,
+      "eval_steps_per_second": 0.177,
+      "step": 2815
+    },
+    {
+      "epoch": 0.7253759216923721,
+      "grad_norm": 0.20051875201203073,
+      "learning_rate": 2.127310504192393e-05,
+      "loss": 0.3865,
+      "step": 2816
+    },
+    {
+      "epoch": 0.7256335125736549,
+      "grad_norm": 0.16568167623786675,
+      "learning_rate": 2.1236309846238166e-05,
+      "loss": 0.3963,
+      "step": 2817
+    },
+    {
+      "epoch": 0.7258911034549377,
+      "grad_norm": 0.19477369321158514,
+      "learning_rate": 2.1199537917914386e-05,
+      "loss": 0.4678,
+      "step": 2818
+    },
+    {
+      "epoch": 0.7261486943362205,
+      "grad_norm": 0.2569997922112408,
+      "learning_rate": 2.116278928669794e-05,
+      "loss": 0.2914,
+      "step": 2819
+    },
+    {
+      "epoch": 0.7264062852175033,
+      "grad_norm": 0.16061324251902118,
+      "learning_rate": 2.112606398231532e-05,
+      "loss": 0.3937,
+      "step": 2820
+    },
+    {
+      "epoch": 0.7264062852175033,
+      "eval_PRM Accuracy": 0.7830188679245284,
+      "eval_PRM F1": 0.8571428571428571,
+      "eval_PRM F1 AUC": 0.7200104766893662,
+      "eval_PRM F1 AUC (fixed)": 0.8486118386589837,
+      "eval_PRM F1 Neg": 0.5490196078431373,
+      "eval_PRM NPV": 0.5,
+      "eval_PRM Precision": 0.8846153846153846,
+      "eval_PRM Recall": 0.8313253012048193,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35342684388160706,
+      "eval_runtime": 16.9883,
+      "eval_samples_per_second": 2.59,
+      "eval_steps_per_second": 0.177,
+      "step": 2820
+    },
+    {
+      "epoch": 0.7266638760987861,
+      "grad_norm": 0.17647205144652872,
+      "learning_rate": 2.1089362034474174e-05,
+      "loss": 0.3539,
+      "step": 2821
+    },
+    {
+      "epoch": 0.7269214669800689,
+      "grad_norm": 0.17270817125147134,
+      "learning_rate": 2.1052683472863206e-05,
+      "loss": 0.3466,
+      "step": 2822
+    },
+    {
+      "epoch": 0.7271790578613517,
+      "grad_norm": 0.13480061391904238,
+      "learning_rate": 2.101602832715224e-05,
+      "loss": 0.2718,
+      "step": 2823
+    },
+    {
+      "epoch": 0.7274366487426345,
+      "grad_norm": 0.20601105689222124,
+      "learning_rate": 2.0979396626992172e-05,
+      "loss": 0.4099,
+      "step": 2824
+    },
+    {
+      "epoch": 0.7276942396239173,
+      "grad_norm": 0.13614223332667116,
+      "learning_rate": 2.0942788402014867e-05,
+      "loss": 0.2952,
+      "step": 2825
+    },
+    {
+      "epoch": 0.7276942396239173,
+      "eval_PRM Accuracy": 0.7830188679245284,
+      "eval_PRM F1": 0.8571428571428571,
+      "eval_PRM F1 AUC": 0.7200104766893662,
+      "eval_PRM F1 AUC (fixed)": 0.8412781561026716,
+      "eval_PRM F1 Neg": 0.5490196078431373,
+      "eval_PRM NPV": 0.5,
+      "eval_PRM Precision": 0.8846153846153846,
+      "eval_PRM Recall": 0.8313253012048193,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.35378196835517883,
+      "eval_runtime": 16.9682,
+      "eval_samples_per_second": 2.593,
+      "eval_steps_per_second": 0.177,
+      "step": 2825
+    },
+    {
+      "epoch": 0.7279518305052001,
+      "grad_norm": 0.21917858956071629,
+      "learning_rate": 2.090620368183329e-05,
+      "loss": 0.4518,
+      "step": 2826
+    },
+    {
+      "epoch": 0.7282094213864829,
+      "grad_norm": 0.15729039477419035,
+      "learning_rate": 2.0869642496041287e-05,
+      "loss": 0.3691,
+      "step": 2827
+    },
+    {
+      "epoch": 0.7284670122677657,
+      "grad_norm": 0.1579740569695876,
+      "learning_rate": 2.08331048742138e-05,
+      "loss": 0.4156,
+      "step": 2828
+    },
+    {
+      "epoch": 0.7287246031490485,
+      "grad_norm": 0.2746904571365112,
+      "learning_rate": 2.07965908459066e-05,
+      "loss": 0.4188,
+      "step": 2829
+    },
+    {
+      "epoch": 0.7289821940303314,
+      "grad_norm": 0.1393606165959669,
+      "learning_rate": 2.0760100440656387e-05,
+      "loss": 0.3145,
+      "step": 2830
+    },
+    {
+      "epoch": 0.7289821940303314,
+      "eval_PRM Accuracy": 0.7830188679245284,
+      "eval_PRM F1": 0.8571428571428571,
+      "eval_PRM F1 AUC": 0.7200104766893662,
+      "eval_PRM F1 AUC (fixed)": 0.8491356731272917,
+      "eval_PRM F1 Neg": 0.5490196078431373,
+      "eval_PRM NPV": 0.5,
+      "eval_PRM Precision": 0.8846153846153846,
+      "eval_PRM Recall": 0.8313253012048193,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3520951569080353,
+      "eval_runtime": 16.8114,
+      "eval_samples_per_second": 2.617,
+      "eval_steps_per_second": 0.178,
+      "step": 2830
+    },
+    {
+      "epoch": 0.7292397849116141,
+      "grad_norm": 0.12932326109029305,
+      "learning_rate": 2.0723633687980796e-05,
+      "loss": 0.3047,
+      "step": 2831
+    },
+    {
+      "epoch": 0.729497375792897,
+      "grad_norm": 0.1498285785627019,
+      "learning_rate": 2.068719061737831e-05,
+      "loss": 0.3614,
+      "step": 2832
+    },
+    {
+      "epoch": 0.7297549666741797,
+      "grad_norm": 0.20868169377215004,
+      "learning_rate": 2.065077125832826e-05,
+      "loss": 0.3363,
+      "step": 2833
+    },
+    {
+      "epoch": 0.7300125575554626,
+      "grad_norm": 0.14402969023476137,
+      "learning_rate": 2.061437564029076e-05,
+      "loss": 0.3519,
+      "step": 2834
+    },
+    {
+      "epoch": 0.7302701484367453,
+      "grad_norm": 0.15265662731074459,
+      "learning_rate": 2.0578003792706768e-05,
+      "loss": 0.2935,
+      "step": 2835
+    },
+    {
+      "epoch": 0.7302701484367453,
+      "eval_PRM Accuracy": 0.7830188679245284,
+      "eval_PRM F1": 0.8571428571428571,
+      "eval_PRM F1 AUC": 0.7200104766893662,
+      "eval_PRM F1 AUC (fixed)": 0.8488737558931377,
+      "eval_PRM F1 Neg": 0.5490196078431373,
+      "eval_PRM NPV": 0.5,
+      "eval_PRM Precision": 0.8846153846153846,
+      "eval_PRM Recall": 0.8313253012048193,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3531605005264282,
+      "eval_runtime": 17.5934,
+      "eval_samples_per_second": 2.501,
+      "eval_steps_per_second": 0.171,
+      "step": 2835
+    },
+    {
+      "epoch": 0.7305277393180282,
+      "grad_norm": 0.24287864454517327,
+      "learning_rate": 2.0541655744998e-05,
+      "loss": 0.4744,
+      "step": 2836
+    },
+    {
+      "epoch": 0.7307853301993109,
+      "grad_norm": 0.15874048526571136,
+      "learning_rate": 2.0505331526566895e-05,
+      "loss": 0.3077,
+      "step": 2837
+    },
+    {
+      "epoch": 0.7310429210805938,
+      "grad_norm": 0.1463427369358526,
+      "learning_rate": 2.0469031166796655e-05,
+      "loss": 0.3317,
+      "step": 2838
+    },
+    {
+      "epoch": 0.7313005119618765,
+      "grad_norm": 0.18444488811203785,
+      "learning_rate": 2.0432754695051136e-05,
+      "loss": 0.3929,
+      "step": 2839
+    },
+    {
+      "epoch": 0.7315581028431594,
+      "grad_norm": 0.20367463774789532,
+      "learning_rate": 2.0396502140674912e-05,
+      "loss": 0.4534,
+      "step": 2840
+    },
+    {
+      "epoch": 0.7315581028431594,
+      "eval_PRM Accuracy": 0.8018867924528302,
+      "eval_PRM F1": 0.8711656441717791,
+      "eval_PRM F1 AUC": 0.7320586694604505,
+      "eval_PRM F1 AUC (fixed)": 0.8501833420639078,
+      "eval_PRM F1 Neg": 0.5714285714285714,
+      "eval_PRM NPV": 0.5384615384615384,
+      "eval_PRM Precision": 0.8875,
+      "eval_PRM Recall": 0.8554216867469879,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3519176244735718,
+      "eval_runtime": 16.8662,
+      "eval_samples_per_second": 2.609,
+      "eval_steps_per_second": 0.178,
+      "step": 2840
+    },
+    {
+      "epoch": 0.7318156937244421,
+      "grad_norm": 0.1991717128847312,
+      "learning_rate": 2.0360273532993197e-05,
+      "loss": 0.3871,
+      "step": 2841
+    },
+    {
+      "epoch": 0.732073284605725,
+      "grad_norm": 0.20576862098770296,
+      "learning_rate": 2.0324068901311845e-05,
+      "loss": 0.3311,
+      "step": 2842
+    },
+    {
+      "epoch": 0.7323308754870078,
+      "grad_norm": 0.16023628152659833,
+      "learning_rate": 2.028788827491729e-05,
+      "loss": 0.3438,
+      "step": 2843
+    },
+    {
+      "epoch": 0.7325884663682906,
+      "grad_norm": 0.16413587615962985,
+      "learning_rate": 2.0251731683076513e-05,
+      "loss": 0.2958,
+      "step": 2844
+    },
+    {
+      "epoch": 0.7328460572495734,
+      "grad_norm": 0.1841223663022876,
+      "learning_rate": 2.021559915503719e-05,
+      "loss": 0.3785,
+      "step": 2845
+    },
+    {
+      "epoch": 0.7328460572495734,
+      "eval_PRM Accuracy": 0.8018867924528302,
+      "eval_PRM F1": 0.8727272727272727,
+      "eval_PRM F1 AUC": 0.7163436354112102,
+      "eval_PRM F1 AUC (fixed)": 0.8525405971712938,
+      "eval_PRM F1 Neg": 0.5531914893617021,
+      "eval_PRM NPV": 0.5416666666666666,
+      "eval_PRM Precision": 0.8780487804878049,
+      "eval_PRM Recall": 0.8674698795180723,
+      "eval_PRM Specificty": 0.5652173913043478,
+      "eval_loss": 0.34912109375,
+      "eval_runtime": 17.3349,
+      "eval_samples_per_second": 2.538,
+      "eval_steps_per_second": 0.173,
+      "step": 2845
+    },
+    {
+      "epoch": 0.7331036481308562,
+      "grad_norm": 0.12811717473489762,
+      "learning_rate": 2.0179490720027372e-05,
+      "loss": 0.3372,
+      "step": 2846
+    },
+    {
+      "epoch": 0.733361239012139,
+      "grad_norm": 0.1688778809301542,
+      "learning_rate": 2.014340640725574e-05,
+      "loss": 0.3721,
+      "step": 2847
+    },
+    {
+      "epoch": 0.7336188298934218,
+      "grad_norm": 0.1388623977493131,
+      "learning_rate": 2.0107346245911363e-05,
+      "loss": 0.3548,
+      "step": 2848
+    },
+    {
+      "epoch": 0.7338764207747046,
+      "grad_norm": 0.17128413866710143,
+      "learning_rate": 2.007131026516385e-05,
+      "loss": 0.3845,
+      "step": 2849
+    },
+    {
+      "epoch": 0.7341340116559874,
+      "grad_norm": 0.20677829508976778,
+      "learning_rate": 2.0035298494163242e-05,
+      "loss": 0.5202,
+      "step": 2850
+    },
+    {
+      "epoch": 0.7341340116559874,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8795180722891566,
+      "eval_PRM F1 AUC": 0.7223677317967523,
+      "eval_PRM F1 AUC (fixed)": 0.856469355683604,
+      "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": 0.3509410619735718,
+      "eval_runtime": 16.7754,
+      "eval_samples_per_second": 2.623,
+      "eval_steps_per_second": 0.179,
+      "step": 2850
+    },
+    {
+      "epoch": 0.7343916025372702,
+      "grad_norm": 0.1708544325194527,
+      "learning_rate": 1.9999310962039934e-05,
+      "loss": 0.2936,
+      "step": 2851
+    },
+    {
+      "epoch": 0.734649193418553,
+      "grad_norm": 0.1266410946760741,
+      "learning_rate": 1.996334769790479e-05,
+      "loss": 0.3464,
+      "step": 2852
+    },
+    {
+      "epoch": 0.7349067842998358,
+      "grad_norm": 0.2469503024356025,
+      "learning_rate": 1.992740873084899e-05,
+      "loss": 0.4851,
+      "step": 2853
+    },
+    {
+      "epoch": 0.7351643751811185,
+      "grad_norm": 0.13580119361733087,
+      "learning_rate": 1.989149408994412e-05,
+      "loss": 0.3464,
+      "step": 2854
+    },
+    {
+      "epoch": 0.7354219660624014,
+      "grad_norm": 0.16299674356087096,
+      "learning_rate": 1.9855603804241996e-05,
+      "loss": 0.3548,
+      "step": 2855
+    },
+    {
+      "epoch": 0.7354219660624014,
+      "eval_PRM Accuracy": 0.8018867924528302,
+      "eval_PRM F1": 0.8727272727272727,
+      "eval_PRM F1 AUC": 0.7163436354112102,
+      "eval_PRM F1 AUC (fixed)": 0.85620743844945,
+      "eval_PRM F1 Neg": 0.5531914893617021,
+      "eval_PRM NPV": 0.5416666666666666,
+      "eval_PRM Precision": 0.8780487804878049,
+      "eval_PRM Recall": 0.8674698795180723,
+      "eval_PRM Specificty": 0.5652173913043478,
+      "eval_loss": 0.3465021252632141,
+      "eval_runtime": 18.0216,
+      "eval_samples_per_second": 2.442,
+      "eval_steps_per_second": 0.166,
+      "step": 2855
+    },
+    {
+      "epoch": 0.7356795569436841,
+      "grad_norm": 0.15873327356315253,
+      "learning_rate": 1.9819737902774827e-05,
+      "loss": 0.4211,
+      "step": 2856
+    },
+    {
+      "epoch": 0.735937147824967,
+      "grad_norm": 0.15839506512733864,
+      "learning_rate": 1.9783896414555025e-05,
+      "loss": 0.3314,
+      "step": 2857
+    },
+    {
+      "epoch": 0.7361947387062499,
+      "grad_norm": 0.1452327507498212,
+      "learning_rate": 1.9748079368575295e-05,
+      "loss": 0.3955,
+      "step": 2858
+    },
+    {
+      "epoch": 0.7364523295875326,
+      "grad_norm": 0.15903516455174,
+      "learning_rate": 1.971228679380858e-05,
+      "loss": 0.3542,
+      "step": 2859
+    },
+    {
+      "epoch": 0.7367099204688154,
+      "grad_norm": 0.17939374114155127,
+      "learning_rate": 1.9676518719207977e-05,
+      "loss": 0.4656,
+      "step": 2860
+    },
+    {
+      "epoch": 0.7367099204688154,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8765432098765432,
+      "eval_PRM F1 AUC": 0.7537977998952332,
+      "eval_PRM F1 AUC (fixed)": 0.855945521215296,
+      "eval_PRM F1 Neg": 0.6,
+      "eval_PRM NPV": 0.5555555555555556,
+      "eval_PRM Precision": 0.8987341772151899,
+      "eval_PRM Recall": 0.8554216867469879,
+      "eval_PRM Specificty": 0.6521739130434783,
+      "eval_loss": 0.3450816869735718,
+      "eval_runtime": 17.0479,
+      "eval_samples_per_second": 2.581,
+      "eval_steps_per_second": 0.176,
+      "step": 2860
+    },
+    {
+      "epoch": 0.7369675113500982,
+      "grad_norm": 0.15200576468787627,
+      "learning_rate": 1.964077517370681e-05,
+      "loss": 0.3171,
+      "step": 2861
+    },
+    {
+      "epoch": 0.737225102231381,
+      "grad_norm": 0.21334655667634317,
+      "learning_rate": 1.9605056186218556e-05,
+      "loss": 0.4409,
+      "step": 2862
+    },
+    {
+      "epoch": 0.7374826931126638,
+      "grad_norm": 0.14205795400710666,
+      "learning_rate": 1.9569361785636797e-05,
+      "loss": 0.3282,
+      "step": 2863
+    },
+    {
+      "epoch": 0.7377402839939466,
+      "grad_norm": 0.20040320969796976,
+      "learning_rate": 1.953369200083527e-05,
+      "loss": 0.3636,
+      "step": 2864
+    },
+    {
+      "epoch": 0.7379978748752294,
+      "grad_norm": 0.11111142214894738,
+      "learning_rate": 1.949804686066772e-05,
+      "loss": 0.2554,
+      "step": 2865
+    },
+    {
+      "epoch": 0.7379978748752294,
+      "eval_PRM Accuracy": 0.8018867924528302,
+      "eval_PRM F1": 0.8695652173913043,
+      "eval_PRM F1 AUC": 0.747773703509691,
+      "eval_PRM F1 AUC (fixed)": 0.8520167627029859,
+      "eval_PRM F1 Neg": 0.5882352941176471,
+      "eval_PRM NPV": 0.5357142857142857,
+      "eval_PRM Precision": 0.8974358974358975,
+      "eval_PRM Recall": 0.8433734939759037,
+      "eval_PRM Specificty": 0.6521739130434783,
+      "eval_loss": 0.3450372815132141,
+      "eval_runtime": 17.6178,
+      "eval_samples_per_second": 2.497,
+      "eval_steps_per_second": 0.17,
+      "step": 2865
+    },
+    {
+      "epoch": 0.7382554657565122,
+      "grad_norm": 0.18847941333564858,
+      "learning_rate": 1.946242639396809e-05,
+      "loss": 0.3828,
+      "step": 2866
+    },
+    {
+      "epoch": 0.738513056637795,
+      "grad_norm": 0.18877652381601745,
+      "learning_rate": 1.9426830629550242e-05,
+      "loss": 0.4274,
+      "step": 2867
+    },
+    {
+      "epoch": 0.7387706475190778,
+      "grad_norm": 0.16026963134003505,
+      "learning_rate": 1.939125959620812e-05,
+      "loss": 0.3781,
+      "step": 2868
+    },
+    {
+      "epoch": 0.7390282384003606,
+      "grad_norm": 0.18082191678975715,
+      "learning_rate": 1.935571332271562e-05,
+      "loss": 0.3381,
+      "step": 2869
+    },
+    {
+      "epoch": 0.7392858292816434,
+      "grad_norm": 0.15372085179022296,
+      "learning_rate": 1.9320191837826645e-05,
+      "loss": 0.4042,
+      "step": 2870
+    },
+    {
+      "epoch": 0.7392858292816434,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8765432098765432,
+      "eval_PRM F1 AUC": 0.7537977998952332,
+      "eval_PRM F1 AUC (fixed)": 0.8504452592980618,
+      "eval_PRM F1 Neg": 0.6,
+      "eval_PRM NPV": 0.5555555555555556,
+      "eval_PRM Precision": 0.8987341772151899,
+      "eval_PRM Recall": 0.8554216867469879,
+      "eval_PRM Specificty": 0.6521739130434783,
+      "eval_loss": 0.3495649993419647,
+      "eval_runtime": 16.936,
+      "eval_samples_per_second": 2.598,
+      "eval_steps_per_second": 0.177,
+      "step": 2870
+    },
+    {
+      "epoch": 0.7395434201629262,
+      "grad_norm": 0.14695245780883914,
+      "learning_rate": 1.9284695170275054e-05,
+      "loss": 0.3517,
+      "step": 2871
+    },
+    {
+      "epoch": 0.739801011044209,
+      "grad_norm": 0.12643757593323876,
+      "learning_rate": 1.9249223348774576e-05,
+      "loss": 0.3489,
+      "step": 2872
+    },
+    {
+      "epoch": 0.7400586019254919,
+      "grad_norm": 0.19796817841926875,
+      "learning_rate": 1.921377640201889e-05,
+      "loss": 0.4138,
+      "step": 2873
+    },
+    {
+      "epoch": 0.7403161928067746,
+      "grad_norm": 0.1298000475881407,
+      "learning_rate": 1.917835435868155e-05,
+      "loss": 0.3158,
+      "step": 2874
+    },
+    {
+      "epoch": 0.7405737836880575,
+      "grad_norm": 0.13516710204554838,
+      "learning_rate": 1.914295724741596e-05,
+      "loss": 0.3638,
+      "step": 2875
+    },
+    {
+      "epoch": 0.7405737836880575,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8795180722891566,
+      "eval_PRM F1 AUC": 0.7223677317967523,
+      "eval_PRM F1 AUC (fixed)": 0.8514929282346779,
+      "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": 0.3528497815132141,
+      "eval_runtime": 17.5638,
+      "eval_samples_per_second": 2.505,
+      "eval_steps_per_second": 0.171,
+      "step": 2875
+    },
+    {
+      "epoch": 0.7408313745693402,
+      "grad_norm": 0.17167873035919043,
+      "learning_rate": 1.9107585096855347e-05,
+      "loss": 0.3429,
+      "step": 2876
+    },
+    {
+      "epoch": 0.7410889654506231,
+      "grad_norm": 0.13290459640636187,
+      "learning_rate": 1.9072237935612723e-05,
+      "loss": 0.3138,
+      "step": 2877
+    },
+    {
+      "epoch": 0.7413465563319058,
+      "grad_norm": 0.19189743311010096,
+      "learning_rate": 1.903691579228094e-05,
+      "loss": 0.3939,
+      "step": 2878
+    },
+    {
+      "epoch": 0.7416041472131887,
+      "grad_norm": 0.16895816792952276,
+      "learning_rate": 1.9001618695432587e-05,
+      "loss": 0.42,
+      "step": 2879
+    },
+    {
+      "epoch": 0.7418617380944714,
+      "grad_norm": 0.14064985293965904,
+      "learning_rate": 1.896634667362e-05,
+      "loss": 0.3335,
+      "step": 2880
+    },
+    {
+      "epoch": 0.7418617380944714,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8862275449101796,
+      "eval_PRM F1 AUC": 0.7283918281822945,
+      "eval_PRM F1 AUC (fixed)": 0.8551597695128339,
+      "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": 0.3513849377632141,
+      "eval_runtime": 17.0451,
+      "eval_samples_per_second": 2.581,
+      "eval_steps_per_second": 0.176,
+      "step": 2880
+    },
+    {
+      "epoch": 0.7421193289757543,
+      "grad_norm": 0.18602414961212507,
+      "learning_rate": 1.8931099755375203e-05,
+      "loss": 0.3734,
+      "step": 2881
+    },
+    {
+      "epoch": 0.742376919857037,
+      "grad_norm": 0.15488617309263178,
+      "learning_rate": 1.8895877969209942e-05,
+      "loss": 0.3407,
+      "step": 2882
+    },
+    {
+      "epoch": 0.7426345107383199,
+      "grad_norm": 0.15243021960849687,
+      "learning_rate": 1.8860681343615637e-05,
+      "loss": 0.396,
+      "step": 2883
+    },
+    {
+      "epoch": 0.7428921016196026,
+      "grad_norm": 0.1580456194150447,
+      "learning_rate": 1.8825509907063327e-05,
+      "loss": 0.3891,
+      "step": 2884
+    },
+    {
+      "epoch": 0.7431496925008855,
+      "grad_norm": 0.14722958450716395,
+      "learning_rate": 1.8790363688003715e-05,
+      "loss": 0.3335,
+      "step": 2885
+    },
+    {
+      "epoch": 0.7431496925008855,
+      "eval_PRM Accuracy": 0.7924528301886793,
+      "eval_PRM F1": 0.8658536585365854,
+      "eval_PRM F1 AUC": 0.710319539025668,
+      "eval_PRM F1 AUC (fixed)": 0.8548978522786799,
+      "eval_PRM F1 Neg": 0.5416666666666666,
+      "eval_PRM NPV": 0.52,
+      "eval_PRM Precision": 0.8765432098765432,
+      "eval_PRM Recall": 0.8554216867469879,
+      "eval_PRM Specificty": 0.5652173913043478,
+      "eval_loss": 0.34659090638160706,
+      "eval_runtime": 16.8948,
+      "eval_samples_per_second": 2.604,
+      "eval_steps_per_second": 0.178,
+      "step": 2885
+    },
+    {
+      "epoch": 0.7434072833821682,
+      "grad_norm": 0.14951969006379529,
+      "learning_rate": 1.8755242714867033e-05,
+      "loss": 0.3679,
+      "step": 2886
+    },
+    {
+      "epoch": 0.7436648742634511,
+      "grad_norm": 0.20213164773469453,
+      "learning_rate": 1.872014701606321e-05,
+      "loss": 0.3379,
+      "step": 2887
+    },
+    {
+      "epoch": 0.7439224651447339,
+      "grad_norm": 0.15156321122722566,
+      "learning_rate": 1.8685076619981608e-05,
+      "loss": 0.4043,
+      "step": 2888
+    },
+    {
+      "epoch": 0.7441800560260167,
+      "grad_norm": 0.1691706422290455,
+      "learning_rate": 1.8650031554991204e-05,
+      "loss": 0.397,
+      "step": 2889
+    },
+    {
+      "epoch": 0.7444376469072995,
+      "grad_norm": 0.17246103476737243,
+      "learning_rate": 1.861501184944043e-05,
+      "loss": 0.4437,
+      "step": 2890
+    },
+    {
+      "epoch": 0.7444376469072995,
+      "eval_PRM Accuracy": 0.8018867924528302,
+      "eval_PRM F1": 0.8711656441717791,
+      "eval_PRM F1 AUC": 0.7320586694604505,
+      "eval_PRM F1 AUC (fixed)": 0.8543740178103718,
+      "eval_PRM F1 Neg": 0.5714285714285714,
+      "eval_PRM NPV": 0.5384615384615384,
+      "eval_PRM Precision": 0.8875,
+      "eval_PRM Recall": 0.8554216867469879,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3437943756580353,
+      "eval_runtime": 16.8841,
+      "eval_samples_per_second": 2.606,
+      "eval_steps_per_second": 0.178,
+      "step": 2890
+    },
+    {
+      "epoch": 0.7446952377885823,
+      "grad_norm": 0.1400396128521265,
+      "learning_rate": 1.858001753165724e-05,
+      "loss": 0.3539,
+      "step": 2891
+    },
+    {
+      "epoch": 0.7449528286698651,
+      "grad_norm": 0.17662014355854183,
+      "learning_rate": 1.8545048629949053e-05,
+      "loss": 0.4447,
+      "step": 2892
+    },
+    {
+      "epoch": 0.7452104195511479,
+      "grad_norm": 0.1482406577038468,
+      "learning_rate": 1.8510105172602693e-05,
+      "loss": 0.3367,
+      "step": 2893
+    },
+    {
+      "epoch": 0.7454680104324307,
+      "grad_norm": 0.12641438932045507,
+      "learning_rate": 1.847518718788443e-05,
+      "loss": 0.32,
+      "step": 2894
+    },
+    {
+      "epoch": 0.7457256013137135,
+      "grad_norm": 0.1585391209790409,
+      "learning_rate": 1.844029470403993e-05,
+      "loss": 0.4044,
+      "step": 2895
+    },
+    {
+      "epoch": 0.7457256013137135,
+      "eval_PRM Accuracy": 0.8018867924528302,
+      "eval_PRM F1": 0.8711656441717791,
+      "eval_PRM F1 AUC": 0.7320586694604505,
+      "eval_PRM F1 AUC (fixed)": 0.8522786799371398,
+      "eval_PRM F1 Neg": 0.5714285714285714,
+      "eval_PRM NPV": 0.5384615384615384,
+      "eval_PRM Precision": 0.8875,
+      "eval_PRM Recall": 0.8554216867469879,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3456587493419647,
+      "eval_runtime": 17.5313,
+      "eval_samples_per_second": 2.51,
+      "eval_steps_per_second": 0.171,
+      "step": 2895
+    },
+    {
+      "epoch": 0.7459831921949963,
+      "grad_norm": 0.16251051820303028,
+      "learning_rate": 1.8405427749294234e-05,
+      "loss": 0.3462,
+      "step": 2896
+    },
+    {
+      "epoch": 0.7462407830762791,
+      "grad_norm": 0.17249731070989127,
+      "learning_rate": 1.837058635185172e-05,
+      "loss": 0.3703,
+      "step": 2897
+    },
+    {
+      "epoch": 0.7464983739575619,
+      "grad_norm": 0.1519789464954743,
+      "learning_rate": 1.8335770539896062e-05,
+      "loss": 0.3211,
+      "step": 2898
+    },
+    {
+      "epoch": 0.7467559648388447,
+      "grad_norm": 0.12250229087928584,
+      "learning_rate": 1.8300980341590297e-05,
+      "loss": 0.3416,
+      "step": 2899
+    },
+    {
+      "epoch": 0.7470135557201275,
+      "grad_norm": 0.12434608201649983,
+      "learning_rate": 1.8266215785076708e-05,
+      "loss": 0.3311,
+      "step": 2900
+    },
+    {
+      "epoch": 0.7470135557201275,
+      "eval_PRM Accuracy": 0.8018867924528302,
+      "eval_PRM F1": 0.8711656441717791,
+      "eval_PRM F1 AUC": 0.7320586694604505,
+      "eval_PRM F1 AUC (fixed)": 0.857778941854374,
+      "eval_PRM F1 Neg": 0.5714285714285714,
+      "eval_PRM NPV": 0.5384615384615384,
+      "eval_PRM Precision": 0.8875,
+      "eval_PRM Recall": 0.8554216867469879,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3418412506580353,
+      "eval_runtime": 17.0572,
+      "eval_samples_per_second": 2.58,
+      "eval_steps_per_second": 0.176,
+      "step": 2900
+    },
+    {
+      "epoch": 0.7472711466014104,
+      "grad_norm": 0.1492807344471492,
+      "learning_rate": 1.8231476898476867e-05,
+      "loss": 0.331,
+      "step": 2901
+    },
+    {
+      "epoch": 0.7475287374826931,
+      "grad_norm": 0.1413288815423703,
+      "learning_rate": 1.8196763709891524e-05,
+      "loss": 0.2894,
+      "step": 2902
+    },
+    {
+      "epoch": 0.747786328363976,
+      "grad_norm": 0.1452211515922782,
+      "learning_rate": 1.816207624740069e-05,
+      "loss": 0.3075,
+      "step": 2903
+    },
+    {
+      "epoch": 0.7480439192452587,
+      "grad_norm": 0.1480829176685368,
+      "learning_rate": 1.812741453906357e-05,
+      "loss": 0.3622,
+      "step": 2904
+    },
+    {
+      "epoch": 0.7483015101265416,
+      "grad_norm": 0.15788110558347976,
+      "learning_rate": 1.8092778612918482e-05,
+      "loss": 0.3923,
+      "step": 2905
+    },
+    {
+      "epoch": 0.7483015101265416,
+      "eval_PRM Accuracy": 0.8018867924528302,
+      "eval_PRM F1": 0.8711656441717791,
+      "eval_PRM F1 AUC": 0.7320586694604505,
+      "eval_PRM F1 AUC (fixed)": 0.85620743844945,
+      "eval_PRM F1 Neg": 0.5714285714285714,
+      "eval_PRM NPV": 0.5384615384615384,
+      "eval_PRM Precision": 0.8875,
+      "eval_PRM Recall": 0.8554216867469879,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.34512606263160706,
+      "eval_runtime": 17.7062,
+      "eval_samples_per_second": 2.485,
+      "eval_steps_per_second": 0.169,
+      "step": 2905
+    },
+    {
+      "epoch": 0.7485591010078243,
+      "grad_norm": 0.15126243423791813,
+      "learning_rate": 1.8058168496982963e-05,
+      "loss": 0.3614,
+      "step": 2906
+    },
+    {
+      "epoch": 0.7488166918891072,
+      "grad_norm": 0.16507873655754893,
+      "learning_rate": 1.8023584219253575e-05,
+      "loss": 0.4163,
+      "step": 2907
+    },
+    {
+      "epoch": 0.7490742827703899,
+      "grad_norm": 0.1488832117971817,
+      "learning_rate": 1.798902580770611e-05,
+      "loss": 0.3626,
+      "step": 2908
+    },
+    {
+      "epoch": 0.7493318736516728,
+      "grad_norm": 0.1513545373401777,
+      "learning_rate": 1.795449329029531e-05,
+      "loss": 0.3941,
+      "step": 2909
+    },
+    {
+      "epoch": 0.7495894645329555,
+      "grad_norm": 0.13517131173517444,
+      "learning_rate": 1.791998669495506e-05,
+      "loss": 0.3546,
+      "step": 2910
+    },
+    {
+      "epoch": 0.7495894645329555,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8780487804878049,
+      "eval_PRM F1 AUC": 0.7380827658459927,
+      "eval_PRM F1 AUC (fixed)": 0.8590885280251441,
+      "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": 0.3465021252632141,
+      "eval_runtime": 17.5578,
+      "eval_samples_per_second": 2.506,
+      "eval_steps_per_second": 0.171,
+      "step": 2910
+    },
+    {
+      "epoch": 0.7498470554142384,
+      "grad_norm": 0.22523685544678199,
+      "learning_rate": 1.78855060495982e-05,
+      "loss": 0.4323,
+      "step": 2911
+    },
+    {
+      "epoch": 0.7501046462955211,
+      "grad_norm": 0.1718645357263855,
+      "learning_rate": 1.7851051382116646e-05,
+      "loss": 0.4329,
+      "step": 2912
+    },
+    {
+      "epoch": 0.750362237176804,
+      "grad_norm": 0.16909439145573646,
+      "learning_rate": 1.7816622720381283e-05,
+      "loss": 0.3165,
+      "step": 2913
+    },
+    {
+      "epoch": 0.7506198280580867,
+      "grad_norm": 0.14053595469665883,
+      "learning_rate": 1.778222009224192e-05,
+      "loss": 0.3302,
+      "step": 2914
+    },
+    {
+      "epoch": 0.7508774189393695,
+      "grad_norm": 0.1503230658218862,
+      "learning_rate": 1.7747843525527353e-05,
+      "loss": 0.3257,
+      "step": 2915
+    },
+    {
+      "epoch": 0.7508774189393695,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.854635935044526,
+      "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": 0.34787818789482117,
+      "eval_runtime": 17.3232,
+      "eval_samples_per_second": 2.54,
+      "eval_steps_per_second": 0.173,
+      "step": 2915
+    },
+    {
+      "epoch": 0.7511350098206524,
+      "grad_norm": 0.15720921123220546,
+      "learning_rate": 1.7713493048045294e-05,
+      "loss": 0.3257,
+      "step": 2916
+    },
+    {
+      "epoch": 0.7513926007019351,
+      "grad_norm": 0.13069791043410423,
+      "learning_rate": 1.767916868758231e-05,
+      "loss": 0.3076,
+      "step": 2917
+    },
+    {
+      "epoch": 0.751650191583218,
+      "grad_norm": 0.17679548430754036,
+      "learning_rate": 1.7644870471903897e-05,
+      "loss": 0.4254,
+      "step": 2918
+    },
+    {
+      "epoch": 0.7519077824645007,
+      "grad_norm": 0.1207566534633257,
+      "learning_rate": 1.7610598428754337e-05,
+      "loss": 0.3105,
+      "step": 2919
+    },
+    {
+      "epoch": 0.7521653733457836,
+      "grad_norm": 0.14864696819534934,
+      "learning_rate": 1.7576352585856808e-05,
+      "loss": 0.3577,
+      "step": 2920
+    },
+    {
+      "epoch": 0.7521653733457836,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8795180722891566,
+      "eval_PRM F1 AUC": 0.7223677317967523,
+      "eval_PRM F1 AUC (fixed)": 0.857778941854374,
+      "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": 0.34912109375,
+      "eval_runtime": 17.493,
+      "eval_samples_per_second": 2.515,
+      "eval_steps_per_second": 0.171,
+      "step": 2920
+    },
+    {
+      "epoch": 0.7524229642270663,
+      "grad_norm": 0.16679396112225395,
+      "learning_rate": 1.754213297091325e-05,
+      "loss": 0.4314,
+      "step": 2921
+    },
+    {
+      "epoch": 0.7526805551083492,
+      "grad_norm": 0.19271587669515353,
+      "learning_rate": 1.7507939611604428e-05,
+      "loss": 0.3953,
+      "step": 2922
+    },
+    {
+      "epoch": 0.7529381459896319,
+      "grad_norm": 0.17871937531964524,
+      "learning_rate": 1.747377253558982e-05,
+      "loss": 0.4266,
+      "step": 2923
+    },
+    {
+      "epoch": 0.7531957368709148,
+      "grad_norm": 0.13026572785919774,
+      "learning_rate": 1.743963177050763e-05,
+      "loss": 0.3305,
+      "step": 2924
+    },
+    {
+      "epoch": 0.7534533277521975,
+      "grad_norm": 0.1453669538080503,
+      "learning_rate": 1.7405517343974887e-05,
+      "loss": 0.4153,
+      "step": 2925
+    },
+    {
+      "epoch": 0.7534533277521975,
+      "eval_PRM Accuracy": 0.839622641509434,
+      "eval_PRM F1": 0.8994082840236687,
+      "eval_PRM F1 AUC": 0.7404400209533788,
+      "eval_PRM F1 AUC (fixed)": 0.855683603981142,
+      "eval_PRM F1 Neg": 0.6046511627906976,
+      "eval_PRM NPV": 0.65,
+      "eval_PRM Precision": 0.8837209302325582,
+      "eval_PRM Recall": 0.9156626506024096,
+      "eval_PRM Specificty": 0.5652173913043478,
+      "eval_loss": 0.35000887513160706,
+      "eval_runtime": 17.4626,
+      "eval_samples_per_second": 2.52,
+      "eval_steps_per_second": 0.172,
+      "step": 2925
+    },
+    {
+      "epoch": 0.7537109186334804,
+      "grad_norm": 0.24307759373423093,
+      "learning_rate": 1.7371429283587177e-05,
+      "loss": 0.3801,
+      "step": 2926
+    },
+    {
+      "epoch": 0.7539685095147631,
+      "grad_norm": 0.14212415088413297,
+      "learning_rate": 1.733736761691887e-05,
+      "loss": 0.3175,
+      "step": 2927
+    },
+    {
+      "epoch": 0.754226100396046,
+      "grad_norm": 0.15282471754587118,
+      "learning_rate": 1.730333237152289e-05,
+      "loss": 0.3233,
+      "step": 2928
+    },
+    {
+      "epoch": 0.7544836912773287,
+      "grad_norm": 0.13897934498451875,
+      "learning_rate": 1.7269323574930864e-05,
+      "loss": 0.3659,
+      "step": 2929
+    },
+    {
+      "epoch": 0.7547412821586116,
+      "grad_norm": 0.1695791763278988,
+      "learning_rate": 1.7235341254653005e-05,
+      "loss": 0.3957,
+      "step": 2930
+    },
+    {
+      "epoch": 0.7547412821586116,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8564693556836039,
+      "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": 0.3519176244735718,
+      "eval_runtime": 17.0676,
+      "eval_samples_per_second": 2.578,
+      "eval_steps_per_second": 0.176,
+      "step": 2930
+    },
+    {
+      "epoch": 0.7549988730398944,
+      "grad_norm": 0.16179982830368525,
+      "learning_rate": 1.720138543817807e-05,
+      "loss": 0.3185,
+      "step": 2931
+    },
+    {
+      "epoch": 0.7552564639211772,
+      "grad_norm": 0.23922589104682979,
+      "learning_rate": 1.7167456152973423e-05,
+      "loss": 0.3178,
+      "step": 2932
+    },
+    {
+      "epoch": 0.75551405480246,
+      "grad_norm": 0.20969924399766007,
+      "learning_rate": 1.7133553426484944e-05,
+      "loss": 0.4275,
+      "step": 2933
+    },
+    {
+      "epoch": 0.7557716456837428,
+      "grad_norm": 0.16089740417640336,
+      "learning_rate": 1.7099677286137066e-05,
+      "loss": 0.3567,
+      "step": 2934
+    },
+    {
+      "epoch": 0.7560292365650256,
+      "grad_norm": 0.12886208418372316,
+      "learning_rate": 1.7065827759332635e-05,
+      "loss": 0.3124,
+      "step": 2935
+    },
+    {
+      "epoch": 0.7560292365650256,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8780487804878049,
+      "eval_PRM F1 AUC": 0.7380827658459927,
+      "eval_PRM F1 AUC (fixed)": 0.85751702462022,
+      "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": 0.3492986559867859,
+      "eval_runtime": 17.5164,
+      "eval_samples_per_second": 2.512,
+      "eval_steps_per_second": 0.171,
+      "step": 2935
+    },
+    {
+      "epoch": 0.7562868274463084,
+      "grad_norm": 0.1337285576937397,
+      "learning_rate": 1.7032004873453067e-05,
+      "loss": 0.3412,
+      "step": 2936
+    },
+    {
+      "epoch": 0.7565444183275912,
+      "grad_norm": 0.1664150507331768,
+      "learning_rate": 1.6998208655858137e-05,
+      "loss": 0.4291,
+      "step": 2937
+    },
+    {
+      "epoch": 0.756802009208874,
+      "grad_norm": 0.13580790660922049,
+      "learning_rate": 1.6964439133886113e-05,
+      "loss": 0.3988,
+      "step": 2938
+    },
+    {
+      "epoch": 0.7570596000901568,
+      "grad_norm": 0.1404404085199257,
+      "learning_rate": 1.6930696334853662e-05,
+      "loss": 0.345,
+      "step": 2939
+    },
+    {
+      "epoch": 0.7573171909714396,
+      "grad_norm": 0.15238213854304544,
+      "learning_rate": 1.689698028605578e-05,
+      "loss": 0.3218,
+      "step": 2940
+    },
+    {
+      "epoch": 0.7573171909714396,
+      "eval_PRM Accuracy": 0.7924528301886793,
+      "eval_PRM F1": 0.8641975308641975,
+      "eval_PRM F1 AUC": 0.7260345730749084,
+      "eval_PRM F1 AUC (fixed)": 0.8596123624934521,
+      "eval_PRM F1 Neg": 0.56,
+      "eval_PRM NPV": 0.5185185185185185,
+      "eval_PRM Precision": 0.8860759493670886,
+      "eval_PRM Recall": 0.8433734939759037,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3463689684867859,
+      "eval_runtime": 16.9174,
+      "eval_samples_per_second": 2.601,
+      "eval_steps_per_second": 0.177,
+      "step": 2940
+    },
+    {
+      "epoch": 0.7575747818527224,
+      "grad_norm": 0.13425930199780564,
+      "learning_rate": 1.6863291014765892e-05,
+      "loss": 0.3337,
+      "step": 2941
+    },
+    {
+      "epoch": 0.7578323727340052,
+      "grad_norm": 0.16054514975701115,
+      "learning_rate": 1.6829628548235714e-05,
+      "loss": 0.3058,
+      "step": 2942
+    },
+    {
+      "epoch": 0.758089963615288,
+      "grad_norm": 0.16866158814780352,
+      "learning_rate": 1.6795992913695336e-05,
+      "loss": 0.3933,
+      "step": 2943
+    },
+    {
+      "epoch": 0.7583475544965709,
+      "grad_norm": 0.13142791020502617,
+      "learning_rate": 1.6762384138353078e-05,
+      "loss": 0.318,
+      "step": 2944
+    },
+    {
+      "epoch": 0.7586051453778536,
+      "grad_norm": 0.15822763947870488,
+      "learning_rate": 1.6728802249395526e-05,
+      "loss": 0.3451,
+      "step": 2945
+    },
+    {
+      "epoch": 0.7586051453778536,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8848484848484849,
+      "eval_PRM F1 AUC": 0.7441068622315348,
+      "eval_PRM F1 AUC (fixed)": 0.8632792037716082,
+      "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": 0.34716796875,
+      "eval_runtime": 16.8788,
+      "eval_samples_per_second": 2.607,
+      "eval_steps_per_second": 0.178,
+      "step": 2945
+    },
+    {
+      "epoch": 0.7588627362591365,
+      "grad_norm": 0.17066267423933648,
+      "learning_rate": 1.6695247273987624e-05,
+      "loss": 0.3425,
+      "step": 2946
+    },
+    {
+      "epoch": 0.7591203271404192,
+      "grad_norm": 0.11733629413361088,
+      "learning_rate": 1.666171923927242e-05,
+      "loss": 0.2829,
+      "step": 2947
+    },
+    {
+      "epoch": 0.7593779180217021,
+      "grad_norm": 0.1925757723152739,
+      "learning_rate": 1.6628218172371256e-05,
+      "loss": 0.3487,
+      "step": 2948
+    },
+    {
+      "epoch": 0.7596355089029848,
+      "grad_norm": 0.11046305102940092,
+      "learning_rate": 1.65947441003836e-05,
+      "loss": 0.2704,
+      "step": 2949
+    },
+    {
+      "epoch": 0.7598930997842677,
+      "grad_norm": 0.12447076120217489,
+      "learning_rate": 1.6561297050387115e-05,
+      "loss": 0.357,
+      "step": 2950
+    },
+    {
+      "epoch": 0.7598930997842677,
+      "eval_PRM Accuracy": 0.8207547169811321,
+      "eval_PRM F1": 0.8862275449101796,
+      "eval_PRM F1 AUC": 0.7283918281822945,
+      "eval_PRM F1 AUC (fixed)": 0.8593504452592979,
+      "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": 0.3507634997367859,
+      "eval_runtime": 17.1561,
+      "eval_samples_per_second": 2.565,
+      "eval_steps_per_second": 0.175,
+      "step": 2950
+    },
+    {
+      "epoch": 0.7601506906655504,
+      "grad_norm": 0.20224682982646164,
+      "learning_rate": 1.6527877049437622e-05,
+      "loss": 0.442,
+      "step": 2951
+    },
+    {
+      "epoch": 0.7604082815468333,
+      "grad_norm": 0.14177204700939802,
+      "learning_rate": 1.649448412456901e-05,
+      "loss": 0.2664,
+      "step": 2952
+    },
+    {
+      "epoch": 0.760665872428116,
+      "grad_norm": 0.2889674168079724,
+      "learning_rate": 1.646111830279331e-05,
+      "loss": 0.5855,
+      "step": 2953
+    },
+    {
+      "epoch": 0.7609234633093989,
+      "grad_norm": 0.16607495086402554,
+      "learning_rate": 1.642777961110061e-05,
+      "loss": 0.3882,
+      "step": 2954
+    },
+    {
+      "epoch": 0.7611810541906816,
+      "grad_norm": 0.18332211113271998,
+      "learning_rate": 1.6394468076459075e-05,
+      "loss": 0.377,
+      "step": 2955
+    },
+    {
+      "epoch": 0.7611810541906816,
+      "eval_PRM Accuracy": 0.8113207547169812,
+      "eval_PRM F1": 0.8780487804878049,
+      "eval_PRM F1 AUC": 0.7380827658459927,
+      "eval_PRM F1 AUC (fixed)": 0.8564693556836039,
+      "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": 0.3504527807235718,
+      "eval_runtime": 16.9539,
+      "eval_samples_per_second": 2.595,
+      "eval_steps_per_second": 0.177,
+      "step": 2955
+    },
+    {
+      "epoch": 0.7614386450719645,
+      "grad_norm": 0.13803947369702252,
+      "learning_rate": 1.6361183725814854e-05,
+      "loss": 0.3689,
+      "step": 2956
+    },
+    {
+      "epoch": 0.7616962359532472,
+      "grad_norm": 0.14152335309676822,
+      "learning_rate": 1.6327926586092164e-05,
+      "loss": 0.3651,
+      "step": 2957
+    },
+    {
+      "epoch": 0.7619538268345301,
+      "grad_norm": 0.14016725907975244,
+      "learning_rate": 1.6294696684193154e-05,
+      "loss": 0.3171,
+      "step": 2958
+    },
+    {
+      "epoch": 0.7622114177158129,
+      "grad_norm": 0.1701719017758532,
+      "learning_rate": 1.626149404699799e-05,
+      "loss": 0.3681,
+      "step": 2959
+    },
+    {
+      "epoch": 0.7624690085970957,
+      "grad_norm": 0.14654085957327564,
+      "learning_rate": 1.6228318701364766e-05,
+      "loss": 0.3471,
+      "step": 2960
+    },
+    {
+      "epoch": 0.7624690085970957,
+      "eval_PRM Accuracy": 0.8018867924528302,
+      "eval_PRM F1": 0.8711656441717791,
+      "eval_PRM F1 AUC": 0.7320586694604505,
+      "eval_PRM F1 AUC (fixed)": 0.8522786799371399,
+      "eval_PRM F1 Neg": 0.5714285714285714,
+      "eval_PRM NPV": 0.5384615384615384,
+      "eval_PRM Precision": 0.8875,
+      "eval_PRM Recall": 0.8554216867469879,
+      "eval_PRM Specificty": 0.6086956521739131,
+      "eval_loss": 0.3520951569080353,
+      "eval_runtime": 17.1812,
+      "eval_samples_per_second": 2.561,
+      "eval_steps_per_second": 0.175,
+      "step": 2960
+    },
+    {
+      "epoch": 0.7627265994783785,
+      "grad_norm": 0.19656535107566206,
+      "learning_rate": 1.619517067412947e-05,
+      "loss": 0.3658,
+      "step": 2961
+    },
+    {
+      "epoch": 0.7629841903596613,
+      "grad_norm": 0.14743344260199182,
+      "learning_rate": 1.6162049992106028e-05,
+      "loss": 0.402,
+      "step": 2962
+    },
+    {
+      "epoch": 0.7632417812409441,
+      "grad_norm": 0.15003486832519877,
+      "learning_rate": 1.6128956682086244e-05,
+      "loss": 0.3789,
+      "step": 2963
+    },
+    {
+      "epoch": 0.7634993721222268,
+      "grad_norm": 0.1478983781365589,
+      "learning_rate": 1.6095890770839785e-05,
+      "loss": 0.3257,
+      "step": 2964
+    },
+    {
+      "epoch": 0.7637569630035097,
+      "grad_norm": 0.16434245775381556,
+      "learning_rate": 1.6062852285114123e-05,
+      "loss": 0.3357,
+      "step": 2965
+    },
+    {
+      "epoch": 0.7637569630035097,
+      "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": 0.35653409361839294,
+      "eval_runtime": 17.0825,
+      "eval_samples_per_second": 2.576,
+      "eval_steps_per_second": 0.176,
+      "step": 2965
+    },
+    {
+      "epoch": 0.7640145538847924,
+      "grad_norm": 0.18186633279718095,
+      "learning_rate": 1.602984125163453e-05,
+      "loss": 0.4663,
+      "step": 2966
+    },
+    {
+      "epoch": 0.7642721447660753,
+      "grad_norm": 0.16844634436853015,
+      "learning_rate": 1.5996857697104173e-05,
+      "loss": 0.4044,
+      "step": 2967
+    },
+    {
+      "epoch": 0.764529735647358,
+      "grad_norm": 0.1545485980685033,
+      "learning_rate": 1.5963901648203873e-05,
+      "loss": 0.4153,
+      "step": 2968
+    },
+    {
+      "epoch": 0.7647873265286409,
+      "grad_norm": 0.2063714841277203,
+      "learning_rate": 1.5930973131592287e-05,
+      "loss": 0.4315,
+      "step": 2969
+    },
+    {
+      "epoch": 0.7650449174099236,
+      "grad_norm": 0.1705379775897913,
+      "learning_rate": 1.5898072173905726e-05,
+      "loss": 0.3686,
+      "step": 2970
+    },
+    {
+      "epoch": 0.7650449174099236,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.8928571428571429,
+      "eval_PRM F1 AUC": 0.7344159245678366,
+      "eval_PRM F1 AUC (fixed)": 0.8567312729177581,
+      "eval_PRM F1 Neg": 0.5909090909090909,
+      "eval_PRM NPV": 0.6190476190476191,
+      "eval_PRM Precision": 0.8823529411764706,
+      "eval_PRM Recall": 0.9036144578313253,
+      "eval_PRM Specificty": 0.5652173913043478,
+      "eval_loss": 0.35684481263160706,
+      "eval_runtime": 16.5503,
+      "eval_samples_per_second": 2.659,
+      "eval_steps_per_second": 0.181,
+      "step": 2970
+    },
+    {
+      "epoch": 0.7653025082912065,
+      "grad_norm": 0.16356938844768326,
+      "learning_rate": 1.586519880175827e-05,
+      "loss": 0.3312,
+      "step": 2971
+    },
+    {
+      "epoch": 0.7655600991724892,
+      "grad_norm": 0.15365701888615851,
+      "learning_rate": 1.583235304174167e-05,
+      "loss": 0.3351,
+      "step": 2972
+    },
+    {
+      "epoch": 0.7658176900537721,
+      "grad_norm": 0.12856143044646068,
+      "learning_rate": 1.5799534920425306e-05,
+      "loss": 0.3165,
+      "step": 2973
+    },
+    {
+      "epoch": 0.766075280935055,
+      "grad_norm": 0.16976094460049398,
+      "learning_rate": 1.576674446435624e-05,
+      "loss": 0.3903,
+      "step": 2974
+    },
+    {
+      "epoch": 0.7663328718163377,
+      "grad_norm": 0.1349218547438422,
+      "learning_rate": 1.573398170005915e-05,
+      "loss": 0.3744,
+      "step": 2975
+    },
+    {
+      "epoch": 0.7663328718163377,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.8928571428571429,
+      "eval_PRM F1 AUC": 0.7344159245678366,
+      "eval_PRM F1 AUC (fixed)": 0.8548978522786799,
+      "eval_PRM F1 Neg": 0.5909090909090909,
+      "eval_PRM NPV": 0.6190476190476191,
+      "eval_PRM Precision": 0.8823529411764706,
+      "eval_PRM Recall": 0.9036144578313253,
+      "eval_PRM Specificty": 0.5652173913043478,
+      "eval_loss": 0.3609730005264282,
+      "eval_runtime": 17.4479,
+      "eval_samples_per_second": 2.522,
+      "eval_steps_per_second": 0.172,
+      "step": 2975
+    },
+    {
+      "epoch": 0.7665904626976205,
+      "grad_norm": 0.14586490633803456,
+      "learning_rate": 1.570124665403632e-05,
+      "loss": 0.2993,
+      "step": 2976
+    },
+    {
+      "epoch": 0.7668480535789033,
+      "grad_norm": 0.17185579787156047,
+      "learning_rate": 1.566853935276758e-05,
+      "loss": 0.46,
+      "step": 2977
+    },
+    {
+      "epoch": 0.7671056444601861,
+      "grad_norm": 0.15102550584338897,
+      "learning_rate": 1.563585982271032e-05,
+      "loss": 0.3124,
+      "step": 2978
+    },
+    {
+      "epoch": 0.7673632353414689,
+      "grad_norm": 0.1813132581237937,
+      "learning_rate": 1.5603208090299498e-05,
+      "loss": 0.4488,
+      "step": 2979
+    },
+    {
+      "epoch": 0.7676208262227517,
+      "grad_norm": 0.17197077484984621,
+      "learning_rate": 1.557058418194757e-05,
+      "loss": 0.4041,
+      "step": 2980
+    },
+    {
+      "epoch": 0.7676208262227517,
+      "eval_PRM Accuracy": 0.8490566037735849,
+      "eval_PRM F1": 0.9058823529411765,
+      "eval_PRM F1 AUC": 0.746464117338921,
+      "eval_PRM F1 AUC (fixed)": 0.8538501833420639,
+      "eval_PRM F1 Neg": 0.6190476190476191,
+      "eval_PRM NPV": 0.6842105263157895,
+      "eval_PRM Precision": 0.8850574712643678,
+      "eval_PRM Recall": 0.927710843373494,
+      "eval_PRM Specificty": 0.5652173913043478,
+      "eval_loss": 0.3643909692764282,
+      "eval_runtime": 17.3241,
+      "eval_samples_per_second": 2.54,
+      "eval_steps_per_second": 0.173,
+      "step": 2980
+    },
+    {
+      "epoch": 0.7678784171040345,
+      "grad_norm": 0.148938786911208,
+      "learning_rate": 1.5537988124044495e-05,
+      "loss": 0.346,
+      "step": 2981
+    },
+    {
+      "epoch": 0.7681360079853173,
+      "grad_norm": 0.15400556027034898,
+      "learning_rate": 1.5505419942957668e-05,
+      "loss": 0.4183,
+      "step": 2982
+    },
+    {
+      "epoch": 0.7683935988666001,
+      "grad_norm": 0.15300890855965438,
+      "learning_rate": 1.5472879665031965e-05,
+      "loss": 0.3504,
+      "step": 2983
+    },
+    {
+      "epoch": 0.7686511897478829,
+      "grad_norm": 0.13426689113485515,
+      "learning_rate": 1.544036731658971e-05,
+      "loss": 0.3121,
+      "step": 2984
+    },
+    {
+      "epoch": 0.7689087806291657,
+      "grad_norm": 0.15191938007208422,
+      "learning_rate": 1.5407882923930565e-05,
+      "loss": 0.3667,
+      "step": 2985
+    },
+    {
+      "epoch": 0.7689087806291657,
+      "eval_PRM Accuracy": 0.8490566037735849,
+      "eval_PRM F1": 0.9058823529411765,
+      "eval_PRM F1 AUC": 0.746464117338921,
+      "eval_PRM F1 AUC (fixed)": 0.8530644316396018,
+      "eval_PRM F1 Neg": 0.6190476190476191,
+      "eval_PRM NPV": 0.6842105263157895,
+      "eval_PRM Precision": 0.8850574712643678,
+      "eval_PRM Recall": 0.927710843373494,
+      "eval_PRM Specificty": 0.5652173913043478,
+      "eval_loss": 0.3705610930919647,
+      "eval_runtime": 16.981,
+      "eval_samples_per_second": 2.591,
+      "eval_steps_per_second": 0.177,
+      "step": 2985
+    },
+    {
+      "epoch": 0.7691663715104485,
+      "grad_norm": 0.13208861504575023,
+      "learning_rate": 1.537542651333167e-05,
+      "loss": 0.3071,
+      "step": 2986
+    },
+    {
+      "epoch": 0.7694239623917314,
+      "grad_norm": 0.12707672366812425,
+      "learning_rate": 1.534299811104742e-05,
+      "loss": 0.3703,
+      "step": 2987
+    },
+    {
+      "epoch": 0.7696815532730141,
+      "grad_norm": 0.12697393465493456,
+      "learning_rate": 1.5310597743309684e-05,
+      "loss": 0.3754,
+      "step": 2988
+    },
+    {
+      "epoch": 0.769939144154297,
+      "grad_norm": 0.12172270298945981,
+      "learning_rate": 1.5278225436327536e-05,
+      "loss": 0.316,
+      "step": 2989
+    },
+    {
+      "epoch": 0.7701967350355797,
+      "grad_norm": 0.18743615861942842,
+      "learning_rate": 1.5245881216287433e-05,
+      "loss": 0.4215,
+      "step": 2990
+    },
+    {
+      "epoch": 0.7701967350355797,
+      "eval_PRM Accuracy": 0.839622641509434,
+      "eval_PRM F1": 0.8994082840236687,
+      "eval_PRM F1 AUC": 0.7404400209533788,
+      "eval_PRM F1 AUC (fixed)": 0.8486118386589837,
+      "eval_PRM F1 Neg": 0.6046511627906976,
+      "eval_PRM NPV": 0.65,
+      "eval_PRM Precision": 0.8837209302325582,
+      "eval_PRM Recall": 0.9156626506024096,
+      "eval_PRM Specificty": 0.5652173913043478,
+      "eval_loss": 0.3681196868419647,
+      "eval_runtime": 17.2366,
+      "eval_samples_per_second": 2.553,
+      "eval_steps_per_second": 0.174,
+      "step": 2990
+    },
+    {
+      "epoch": 0.7704543259168626,
+      "grad_norm": 0.11547746948552717,
+      "learning_rate": 1.5213565109353045e-05,
+      "loss": 0.3056,
+      "step": 2991
+    },
+    {
+      "epoch": 0.7707119167981453,
+      "grad_norm": 0.12374691774652735,
+      "learning_rate": 1.5181277141665356e-05,
+      "loss": 0.2785,
+      "step": 2992
+    },
+    {
+      "epoch": 0.7709695076794282,
+      "grad_norm": 0.13162751974187664,
+      "learning_rate": 1.5149017339342574e-05,
+      "loss": 0.3711,
+      "step": 2993
+    },
+    {
+      "epoch": 0.7712270985607109,
+      "grad_norm": 0.2034467324859911,
+      "learning_rate": 1.5116785728480093e-05,
+      "loss": 0.3958,
+      "step": 2994
+    },
+    {
+      "epoch": 0.7714846894419938,
+      "grad_norm": 0.15343478310052172,
+      "learning_rate": 1.5084582335150538e-05,
+      "loss": 0.3587,
+      "step": 2995
+    },
+    {
+      "epoch": 0.7714846894419938,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.8928571428571429,
+      "eval_PRM F1 AUC": 0.7344159245678366,
+      "eval_PRM F1 AUC (fixed)": 0.8478260869565218,
+      "eval_PRM F1 Neg": 0.5909090909090909,
+      "eval_PRM NPV": 0.6190476190476191,
+      "eval_PRM Precision": 0.8823529411764706,
+      "eval_PRM Recall": 0.9036144578313253,
+      "eval_PRM Specificty": 0.5652173913043478,
+      "eval_loss": 0.3704723119735718,
+      "eval_runtime": 16.8955,
+      "eval_samples_per_second": 2.604,
+      "eval_steps_per_second": 0.178,
+      "step": 2995
+    },
+    {
+      "epoch": 0.7717422803232765,
+      "grad_norm": 0.1080886012047265,
+      "learning_rate": 1.5052407185403694e-05,
+      "loss": 0.2992,
+      "step": 2996
+    },
+    {
+      "epoch": 0.7719998712045594,
+      "grad_norm": 0.12905574745651321,
+      "learning_rate": 1.502026030526652e-05,
+      "loss": 0.3368,
+      "step": 2997
+    },
+    {
+      "epoch": 0.7722574620858421,
+      "grad_norm": 0.16972772518993512,
+      "learning_rate": 1.4988141720743076e-05,
+      "loss": 0.3682,
+      "step": 2998
+    },
+    {
+      "epoch": 0.772515052967125,
+      "grad_norm": 0.17634437263973804,
+      "learning_rate": 1.4956051457814523e-05,
+      "loss": 0.3871,
+      "step": 2999
+    },
+    {
+      "epoch": 0.7727726438484077,
+      "grad_norm": 0.18345313042461878,
+      "learning_rate": 1.4923989542439159e-05,
+      "loss": 0.4054,
+      "step": 3000
+    },
+    {
+      "epoch": 0.7727726438484077,
+      "eval_PRM Accuracy": 0.8301886792452831,
+      "eval_PRM F1": 0.8928571428571429,
+      "eval_PRM F1 AUC": 0.7344159245678366,
+      "eval_PRM F1 AUC (fixed)": 0.8486118386589838,
+      "eval_PRM F1 Neg": 0.5909090909090909,
+      "eval_PRM NPV": 0.6190476190476191,
+      "eval_PRM Precision": 0.8823529411764706,
+      "eval_PRM Recall": 0.9036144578313253,
+      "eval_PRM Specificty": 0.5652173913043478,
+      "eval_loss": 0.36376953125,
+      "eval_runtime": 17.6273,
+      "eval_samples_per_second": 2.496,
+      "eval_steps_per_second": 0.17,
+      "step": 3000
     }
   ],
   "logging_steps": 1,
@@ -26043,7 +31243,7 @@
       "attributes": {}
     }
   },
-  "total_flos": 5453105592795136.0,
+  "total_flos": 6548628801716224.0,
   "train_batch_size": 4,
   "trial_name": null,
   "trial_params": null