diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -13,4354 +13,4354 @@
       "grad_norm": 0.0,
       "learning_rate": 3.690036900369004e-07,
       "loss": 2.8171,
-      "mean_token_accuracy": 0.7238644688644686,
-      "mean_token_accuracy_utility": 0.6261577028529871,
+      "mean_token_accuracy": 0.7239194139194136,
+      "mean_token_accuracy_utility": 0.6262431729384572,
       "step": 5
     },
     {
       "epoch": 0.003700619853825516,
       "grad_norm": 0.0,
       "learning_rate": 7.380073800738008e-07,
-      "loss": 2.8459,
-      "mean_token_accuracy": 0.7136263736263736,
-      "mean_token_accuracy_utility": 0.624126984126984,
+      "loss": 2.8457,
+      "mean_token_accuracy": 0.7140170940170939,
+      "mean_token_accuracy_utility": 0.6237240537240536,
       "step": 10
     },
     {
       "epoch": 0.005550929780738274,
       "grad_norm": 0.0,
       "learning_rate": 1.1070110701107011e-06,
-      "loss": 2.8883,
-      "mean_token_accuracy": 0.7147313797313795,
-      "mean_token_accuracy_utility": 0.6203296703296701,
+      "loss": 2.8885,
+      "mean_token_accuracy": 0.7149389499389497,
+      "mean_token_accuracy_utility": 0.6202136752136751,
       "step": 15
     },
     {
       "epoch": 0.007401239707651032,
       "grad_norm": 0.0,
       "learning_rate": 1.4760147601476015e-06,
-      "loss": 2.8436,
-      "mean_token_accuracy": 0.7276739926739927,
-      "mean_token_accuracy_utility": 0.6185164835164834,
+      "loss": 2.8437,
+      "mean_token_accuracy": 0.7275946275946275,
+      "mean_token_accuracy_utility": 0.6184371184371182,
       "step": 20
     },
     {
       "epoch": 0.00925154963456379,
       "grad_norm": 0.0,
       "learning_rate": 1.845018450184502e-06,
-      "loss": 2.8527,
-      "mean_token_accuracy": 0.7133943833943832,
-      "mean_token_accuracy_utility": 0.6248778998778997,
+      "loss": 2.8526,
+      "mean_token_accuracy": 0.7134065934065933,
+      "mean_token_accuracy_utility": 0.6250488400488398,
       "step": 25
     },
     {
       "epoch": 0.011101859561476548,
       "grad_norm": 0.0,
       "learning_rate": 2.2140221402214023e-06,
-      "loss": 2.8887,
-      "mean_token_accuracy": 0.7092368742368741,
-      "mean_token_accuracy_utility": 0.6195115995115993,
+      "loss": 2.8883,
+      "mean_token_accuracy": 0.7093162393162392,
+      "mean_token_accuracy_utility": 0.6195787545787543,
       "step": 30
     },
     {
       "epoch": 0.012952169488389306,
       "grad_norm": 0.0,
       "learning_rate": 2.5830258302583027e-06,
-      "loss": 2.8154,
-      "mean_token_accuracy": 0.7137606837606836,
-      "mean_token_accuracy_utility": 0.6317032967032965,
+      "loss": 2.8149,
+      "mean_token_accuracy": 0.7138400488400486,
+      "mean_token_accuracy_utility": 0.6315384615384614,
       "step": 35
     },
     {
       "epoch": 0.014802479415302064,
       "grad_norm": 0.0,
       "learning_rate": 2.952029520295203e-06,
-      "loss": 2.7905,
-      "mean_token_accuracy": 0.7237728937728936,
-      "mean_token_accuracy_utility": 0.6272232038394496,
+      "loss": 2.7901,
+      "mean_token_accuracy": 0.7237912087912087,
+      "mean_token_accuracy_utility": 0.6271476917607619,
       "step": 40
     },
     {
       "epoch": 0.01665278934221482,
       "grad_norm": 0.0,
       "learning_rate": 3.3210332103321034e-06,
-      "loss": 2.7976,
-      "mean_token_accuracy": 0.7118559218559218,
-      "mean_token_accuracy_utility": 0.6385042735042733,
+      "loss": 2.7968,
+      "mean_token_accuracy": 0.7120390720390721,
+      "mean_token_accuracy_utility": 0.6386080586080584,
       "step": 45
     },
     {
       "epoch": 0.01850309926912758,
       "grad_norm": 0.0,
       "learning_rate": 3.690036900369004e-06,
-      "loss": 2.7852,
-      "mean_token_accuracy": 0.7181446331446332,
-      "mean_token_accuracy_utility": 0.6288766788766786,
+      "loss": 2.7843,
+      "mean_token_accuracy": 0.7184282384282384,
+      "mean_token_accuracy_utility": 0.6289621489621487,
       "step": 50
     },
     {
       "epoch": 0.020353409196040336,
       "grad_norm": 0.0,
       "learning_rate": 4.059040590405905e-06,
-      "loss": 2.806,
-      "mean_token_accuracy": 0.7129304029304027,
-      "mean_token_accuracy_utility": 0.6368506493506493,
+      "loss": 2.8048,
+      "mean_token_accuracy": 0.7131623931623929,
+      "mean_token_accuracy_utility": 0.6366239316239315,
       "step": 55
     },
     {
       "epoch": 0.022203719122953096,
       "grad_norm": 0.0,
       "learning_rate": 4.428044280442805e-06,
-      "loss": 2.8408,
-      "mean_token_accuracy": 0.7153540903540904,
-      "mean_token_accuracy_utility": 0.6185653235653235,
+      "loss": 2.8392,
+      "mean_token_accuracy": 0.7157875457875458,
+      "mean_token_accuracy_utility": 0.6183150183150181,
       "step": 60
     },
     {
       "epoch": 0.024054029049865852,
       "grad_norm": 0.0,
       "learning_rate": 4.797047970479705e-06,
-      "loss": 2.7525,
-      "mean_token_accuracy": 0.7125763125763125,
-      "mean_token_accuracy_utility": 0.6417167249055199,
+      "loss": 2.7504,
+      "mean_token_accuracy": 0.7130158730158729,
+      "mean_token_accuracy_utility": 0.641578052713893,
       "step": 65
     },
     {
       "epoch": 0.025904338976778612,
       "grad_norm": 0.0,
       "learning_rate": 5.166051660516605e-06,
-      "loss": 2.7102,
-      "mean_token_accuracy": 0.7395054945054944,
-      "mean_token_accuracy_utility": 0.6291819291819291,
+      "loss": 2.7078,
+      "mean_token_accuracy": 0.7402625152625152,
+      "mean_token_accuracy_utility": 0.6292185592185591,
       "step": 70
     },
     {
       "epoch": 0.027754648903691368,
       "grad_norm": 0.0,
       "learning_rate": 5.535055350553506e-06,
-      "loss": 2.8015,
-      "mean_token_accuracy": 0.716007326007326,
-      "mean_token_accuracy_utility": 0.6239133089133088,
+      "loss": 2.7982,
+      "mean_token_accuracy": 0.7168376068376066,
+      "mean_token_accuracy_utility": 0.6239621489621487,
       "step": 75
     },
     {
       "epoch": 0.029604958830604128,
       "grad_norm": 0.0,
       "learning_rate": 5.904059040590406e-06,
-      "loss": 2.8497,
-      "mean_token_accuracy": 0.712173382173382,
-      "mean_token_accuracy_utility": 0.6209645909645908,
+      "loss": 2.8456,
+      "mean_token_accuracy": 0.7128998778998776,
+      "mean_token_accuracy_utility": 0.621184371184371,
       "step": 80
     },
     {
       "epoch": 0.031455268757516884,
       "grad_norm": 0.0,
       "learning_rate": 6.273062730627307e-06,
-      "loss": 2.7578,
-      "mean_token_accuracy": 0.713083028083028,
-      "mean_token_accuracy_utility": 0.6346995293263948,
+      "loss": 2.7529,
+      "mean_token_accuracy": 0.7137667887667887,
+      "mean_token_accuracy_utility": 0.6346273363103683,
       "step": 85
     },
     {
       "epoch": 0.03330557868442964,
       "grad_norm": 0.0,
       "learning_rate": 6.642066420664207e-06,
-      "loss": 2.7594,
-      "mean_token_accuracy": 0.7263553113553114,
-      "mean_token_accuracy_utility": 0.6264529914529914,
+      "loss": 2.7538,
+      "mean_token_accuracy": 0.7273748473748473,
+      "mean_token_accuracy_utility": 0.6264102564102563,
       "step": 90
     },
     {
       "epoch": 0.0351558886113424,
       "grad_norm": 0.0,
       "learning_rate": 7.011070110701108e-06,
-      "loss": 2.8301,
-      "mean_token_accuracy": 0.7217643467643466,
-      "mean_token_accuracy_utility": 0.6160476047604758,
+      "loss": 2.824,
+      "mean_token_accuracy": 0.7227289377289374,
+      "mean_token_accuracy_utility": 0.6159964952539208,
       "step": 95
     },
     {
       "epoch": 0.03700619853825516,
       "grad_norm": 0.0,
       "learning_rate": 7.380073800738008e-06,
-      "loss": 2.7515,
-      "mean_token_accuracy": 0.719163614163614,
-      "mean_token_accuracy_utility": 0.6315018315018313,
+      "loss": 2.7443,
+      "mean_token_accuracy": 0.7202014652014651,
+      "mean_token_accuracy_utility": 0.6314590964590961,
       "step": 100
     },
     {
       "epoch": 0.03700619853825516,
-      "eval_mean_token_accuracy": 0.7064531329237212,
-      "eval_reasoning_loss": 1.113277554512024,
-      "eval_reasoning_runtime": 28.2024,
-      "eval_reasoning_samples_per_second": 4.574,
-      "eval_reasoning_steps_per_second": 1.17,
+      "eval_mean_token_accuracy": 0.7076255988020693,
+      "eval_reasoning_loss": 1.1029913425445557,
+      "eval_reasoning_runtime": 28.3037,
+      "eval_reasoning_samples_per_second": 4.558,
+      "eval_reasoning_steps_per_second": 1.166,
       "step": 100
     },
     {
       "epoch": 0.03700619853825516,
-      "eval_mean_token_accuracy": 0.6317870052928675,
-      "eval_utility_loss": 1.6742433309555054,
-      "eval_utility_runtime": 10.3553,
-      "eval_utility_samples_per_second": 4.732,
-      "eval_utility_steps_per_second": 1.255,
+      "eval_mean_token_accuracy": 0.6319696491987027,
+      "eval_utility_loss": 1.6765165328979492,
+      "eval_utility_runtime": 10.3825,
+      "eval_utility_samples_per_second": 4.719,
+      "eval_utility_steps_per_second": 1.252,
       "step": 100
     },
     {
       "epoch": 0.038856508465167916,
       "grad_norm": 0.0,
       "learning_rate": 7.749077490774908e-06,
-      "loss": 2.7223,
-      "mean_token_accuracy": 0.7272588522588522,
-      "mean_token_accuracy_utility": 0.639188034188034,
+      "loss": 2.7142,
+      "mean_token_accuracy": 0.7288400488400489,
+      "mean_token_accuracy_utility": 0.6390781440781439,
       "step": 105
     },
     {
       "epoch": 0.04070681839208067,
       "grad_norm": 0.0,
       "learning_rate": 8.11808118081181e-06,
-      "loss": 2.7834,
-      "mean_token_accuracy": 0.7155372405372407,
-      "mean_token_accuracy_utility": 0.6321306471306469,
+      "loss": 2.7745,
+      "mean_token_accuracy": 0.716935286935287,
+      "mean_token_accuracy_utility": 0.6320879120879119,
       "step": 110
     },
     {
       "epoch": 0.04255712831899343,
       "grad_norm": 0.0,
       "learning_rate": 8.48708487084871e-06,
-      "loss": 2.6893,
-      "mean_token_accuracy": 0.7254761904761903,
-      "mean_token_accuracy_utility": 0.6381868131868129,
+      "loss": 2.6804,
+      "mean_token_accuracy": 0.7269108669108667,
+      "mean_token_accuracy_utility": 0.6378632478632477,
       "step": 115
     },
     {
       "epoch": 0.04440743824590619,
       "grad_norm": 0.0,
       "learning_rate": 8.85608856088561e-06,
-      "loss": 2.7541,
-      "mean_token_accuracy": 0.7166178266178264,
-      "mean_token_accuracy_utility": 0.6306471306471304,
+      "loss": 2.744,
+      "mean_token_accuracy": 0.718266178266178,
+      "mean_token_accuracy_utility": 0.6305860805860803,
       "step": 120
     },
     {
       "epoch": 0.04625774817281895,
       "grad_norm": 0.0,
       "learning_rate": 9.22509225092251e-06,
-      "loss": 2.7483,
-      "mean_token_accuracy": 0.72028083028083,
-      "mean_token_accuracy_utility": 0.6265323565323564,
+      "loss": 2.7369,
+      "mean_token_accuracy": 0.7223687423687422,
+      "mean_token_accuracy_utility": 0.6265995115995114,
       "step": 125
     },
     {
       "epoch": 0.048108058099731704,
       "grad_norm": 0.0,
       "learning_rate": 9.59409594095941e-06,
-      "loss": 2.6766,
-      "mean_token_accuracy": 0.7317155067155066,
-      "mean_token_accuracy_utility": 0.6383637928467196,
+      "loss": 2.6643,
+      "mean_token_accuracy": 0.7337484737484736,
+      "mean_token_accuracy_utility": 0.6377910598886207,
       "step": 130
     },
     {
       "epoch": 0.04995836802664446,
       "grad_norm": 0.0,
       "learning_rate": 9.963099630996312e-06,
-      "loss": 2.7168,
-      "mean_token_accuracy": 0.721324786324786,
-      "mean_token_accuracy_utility": 0.6296947496947496,
+      "loss": 2.7034,
+      "mean_token_accuracy": 0.7237423687423685,
+      "mean_token_accuracy_utility": 0.6289316239316238,
       "step": 135
     },
     {
       "epoch": 0.051808677953557224,
       "grad_norm": 0.0,
       "learning_rate": 1.033210332103321e-05,
-      "loss": 2.6831,
-      "mean_token_accuracy": 0.7197924297924295,
-      "mean_token_accuracy_utility": 0.6337728937728936,
+      "loss": 2.6693,
+      "mean_token_accuracy": 0.7219658119658119,
+      "mean_token_accuracy_utility": 0.6333821733821732,
       "step": 140
     },
     {
       "epoch": 0.05365898788046998,
       "grad_norm": 0.0,
       "learning_rate": 1.0701107011070112e-05,
-      "loss": 2.6514,
-      "mean_token_accuracy": 0.72496336996337,
-      "mean_token_accuracy_utility": 0.6427228327228326,
+      "loss": 2.6362,
+      "mean_token_accuracy": 0.7277655677655677,
+      "mean_token_accuracy_utility": 0.6422588522588522,
       "step": 145
     },
     {
       "epoch": 0.055509297807382736,
       "grad_norm": 0.0,
       "learning_rate": 1.1070110701107012e-05,
-      "loss": 2.6592,
-      "mean_token_accuracy": 0.7232600732600731,
-      "mean_token_accuracy_utility": 0.6361111111111108,
+      "loss": 2.6456,
+      "mean_token_accuracy": 0.7257692307692306,
+      "mean_token_accuracy_utility": 0.6358363858363856,
       "step": 150
     },
     {
       "epoch": 0.05735960773429549,
       "grad_norm": 0.0,
       "learning_rate": 1.1439114391143913e-05,
-      "loss": 2.6102,
-      "mean_token_accuracy": 0.7183760683760684,
-      "mean_token_accuracy_utility": 0.6478937728937727,
+      "loss": 2.593,
+      "mean_token_accuracy": 0.7219474969474969,
+      "mean_token_accuracy_utility": 0.6473992673992672,
       "step": 155
     },
     {
       "epoch": 0.059209917661208256,
       "grad_norm": 0.0,
       "learning_rate": 1.1808118081180812e-05,
-      "loss": 2.7045,
-      "mean_token_accuracy": 0.7227533577533576,
-      "mean_token_accuracy_utility": 0.6307081807081806,
+      "loss": 2.6883,
+      "mean_token_accuracy": 0.7254761904761903,
+      "mean_token_accuracy_utility": 0.630152625152625,
       "step": 160
     },
     {
       "epoch": 0.06106022758812101,
       "grad_norm": 0.0,
       "learning_rate": 1.2177121771217713e-05,
-      "loss": 2.6903,
-      "mean_token_accuracy": 0.7196642246642245,
-      "mean_token_accuracy_utility": 0.6294566544566543,
+      "loss": 2.6726,
+      "mean_token_accuracy": 0.7234432234432232,
+      "mean_token_accuracy_utility": 0.629053724053724,
       "step": 165
     },
     {
       "epoch": 0.06291053751503377,
       "grad_norm": 0.0,
       "learning_rate": 1.2546125461254614e-05,
-      "loss": 2.6181,
-      "mean_token_accuracy": 0.7212637362637363,
-      "mean_token_accuracy_utility": 0.6450854700854699,
+      "loss": 2.6005,
+      "mean_token_accuracy": 0.7244688644688644,
+      "mean_token_accuracy_utility": 0.6447557997557996,
       "step": 170
     },
     {
       "epoch": 0.06476084744194653,
       "grad_norm": 0.0,
       "learning_rate": 1.2915129151291515e-05,
-      "loss": 2.6901,
-      "mean_token_accuracy": 0.7224786324786325,
-      "mean_token_accuracy_utility": 0.6315384615384613,
+      "loss": 2.6718,
+      "mean_token_accuracy": 0.7267887667887669,
+      "mean_token_accuracy_utility": 0.631373626373626,
       "step": 175
     },
     {
       "epoch": 0.06661115736885928,
       "grad_norm": 0.0,
       "learning_rate": 1.3284132841328414e-05,
-      "loss": 2.6409,
-      "mean_token_accuracy": 0.7293406593406593,
-      "mean_token_accuracy_utility": 0.6380158730158728,
+      "loss": 2.6207,
+      "mean_token_accuracy": 0.7332417582417582,
+      "mean_token_accuracy_utility": 0.6377655677655676,
       "step": 180
     },
     {
       "epoch": 0.06846146729577204,
       "grad_norm": 0.0,
       "learning_rate": 1.3653136531365315e-05,
-      "loss": 2.7572,
-      "mean_token_accuracy": 0.728919413919414,
-      "mean_token_accuracy_utility": 0.6156349206349204,
+      "loss": 2.7362,
+      "mean_token_accuracy": 0.733266178266178,
+      "mean_token_accuracy_utility": 0.6155982905982904,
       "step": 185
     },
     {
       "epoch": 0.0703117772226848,
       "grad_norm": 0.0,
       "learning_rate": 1.4022140221402215e-05,
-      "loss": 2.6127,
-      "mean_token_accuracy": 0.736068376068376,
-      "mean_token_accuracy_utility": 0.6355555555555553,
+      "loss": 2.5944,
+      "mean_token_accuracy": 0.7400488400488399,
+      "mean_token_accuracy_utility": 0.6355921855921853,
       "step": 190
     },
     {
       "epoch": 0.07216208714959756,
       "grad_norm": 0.0,
       "learning_rate": 1.4391143911439116e-05,
-      "loss": 2.6937,
-      "mean_token_accuracy": 0.722942612942613,
-      "mean_token_accuracy_utility": 0.6204256023717922,
+      "loss": 2.6748,
+      "mean_token_accuracy": 0.7267521367521368,
+      "mean_token_accuracy_utility": 0.6197368037391602,
       "step": 195
     },
     {
       "epoch": 0.07401239707651032,
       "grad_norm": 0.0,
       "learning_rate": 1.4760147601476015e-05,
-      "loss": 2.6286,
-      "mean_token_accuracy": 0.722973137973138,
-      "mean_token_accuracy_utility": 0.6358180708180706,
+      "loss": 2.61,
+      "mean_token_accuracy": 0.727106227106227,
+      "mean_token_accuracy_utility": 0.6355555555555553,
       "step": 200
     },
     {
       "epoch": 0.07401239707651032,
-      "eval_mean_token_accuracy": 0.718920948332713,
-      "eval_reasoning_loss": 1.0296032428741455,
-      "eval_reasoning_runtime": 27.5106,
-      "eval_reasoning_samples_per_second": 4.689,
-      "eval_reasoning_steps_per_second": 1.2,
+      "eval_mean_token_accuracy": 0.7227414198002431,
+      "eval_reasoning_loss": 1.0048792362213135,
+      "eval_reasoning_runtime": 27.583,
+      "eval_reasoning_samples_per_second": 4.677,
+      "eval_reasoning_steps_per_second": 1.196,
       "step": 200
     },
     {
       "epoch": 0.07401239707651032,
-      "eval_mean_token_accuracy": 0.6398229007036556,
-      "eval_utility_loss": 1.5840386152267456,
-      "eval_utility_runtime": 10.3585,
-      "eval_utility_samples_per_second": 4.73,
-      "eval_utility_steps_per_second": 1.255,
+      "eval_mean_token_accuracy": 0.639254665520036,
+      "eval_utility_loss": 1.588428258895874,
+      "eval_utility_runtime": 10.3884,
+      "eval_utility_samples_per_second": 4.717,
+      "eval_utility_steps_per_second": 1.251,
       "step": 200
     },
     {
       "epoch": 0.07586270700342307,
       "grad_norm": 0.0,
       "learning_rate": 1.5129151291512916e-05,
-      "loss": 2.6703,
-      "mean_token_accuracy": 0.729035409035409,
-      "mean_token_accuracy_utility": 0.626019536019536,
+      "loss": 2.6508,
+      "mean_token_accuracy": 0.7328449328449329,
+      "mean_token_accuracy_utility": 0.6252930402930402,
       "step": 205
     },
     {
       "epoch": 0.07771301693033583,
       "grad_norm": 0.0,
       "learning_rate": 1.5498154981549817e-05,
-      "loss": 2.5598,
-      "mean_token_accuracy": 0.7310378510378508,
-      "mean_token_accuracy_utility": 0.639139194139194,
+      "loss": 2.5398,
+      "mean_token_accuracy": 0.7354456654456654,
+      "mean_token_accuracy_utility": 0.6389377289377287,
       "step": 210
     },
     {
       "epoch": 0.0795633268572486,
       "grad_norm": 0.0,
       "learning_rate": 1.5867158671586716e-05,
-      "loss": 2.5814,
-      "mean_token_accuracy": 0.7347374847374846,
-      "mean_token_accuracy_utility": 0.631284433723458,
+      "loss": 2.5611,
+      "mean_token_accuracy": 0.7391697191697191,
+      "mean_token_accuracy_utility": 0.6310683045951337,
       "step": 215
     },
     {
       "epoch": 0.08141363678416134,
       "grad_norm": 0.0,
       "learning_rate": 1.623616236162362e-05,
-      "loss": 2.713,
-      "mean_token_accuracy": 0.7138278388278386,
-      "mean_token_accuracy_utility": 0.6262381274348818,
+      "loss": 2.6949,
+      "mean_token_accuracy": 0.7175824175824174,
+      "mean_token_accuracy_utility": 0.6249915669185444,
       "step": 220
     },
     {
       "epoch": 0.08326394671107411,
       "grad_norm": 0.0,
       "learning_rate": 1.6605166051660518e-05,
-      "loss": 2.5933,
-      "mean_token_accuracy": 0.7267155067155067,
-      "mean_token_accuracy_utility": 0.6369108669108667,
+      "loss": 2.5742,
+      "mean_token_accuracy": 0.7299816849816848,
+      "mean_token_accuracy_utility": 0.6346153846153845,
       "step": 225
     },
     {
       "epoch": 0.08511425663798686,
       "grad_norm": 0.0,
       "learning_rate": 1.697416974169742e-05,
-      "loss": 2.5396,
-      "mean_token_accuracy": 0.7369963369963369,
-      "mean_token_accuracy_utility": 0.6387179487179486,
+      "loss": 2.5213,
+      "mean_token_accuracy": 0.7410561660561661,
+      "mean_token_accuracy_utility": 0.6372649572649571,
       "step": 230
     },
     {
       "epoch": 0.08696456656489962,
       "grad_norm": 0.0,
       "learning_rate": 1.734317343173432e-05,
-      "loss": 2.6289,
-      "mean_token_accuracy": 0.7330158730158729,
-      "mean_token_accuracy_utility": 0.6313308913308913,
+      "loss": 2.6093,
+      "mean_token_accuracy": 0.7373076923076922,
+      "mean_token_accuracy_utility": 0.6303052503052502,
       "step": 235
     },
     {
       "epoch": 0.08881487649181238,
       "grad_norm": 0.0,
       "learning_rate": 1.771217712177122e-05,
-      "loss": 2.4309,
-      "mean_token_accuracy": 0.7366117216117216,
-      "mean_token_accuracy_utility": 0.6589804639804638,
+      "loss": 2.4115,
+      "mean_token_accuracy": 0.7407753357753357,
+      "mean_token_accuracy_utility": 0.6580341880341879,
       "step": 240
     },
     {
       "epoch": 0.09066518641872513,
       "grad_norm": 0.0,
       "learning_rate": 1.8081180811808117e-05,
-      "loss": 2.6513,
-      "mean_token_accuracy": 0.7233821733821733,
-      "mean_token_accuracy_utility": 0.6351221001221,
+      "loss": 2.6302,
+      "mean_token_accuracy": 0.7275274725274723,
+      "mean_token_accuracy_utility": 0.634120879120879,
       "step": 245
     },
     {
       "epoch": 0.0925154963456379,
       "grad_norm": 0.0,
       "learning_rate": 1.845018450184502e-05,
-      "loss": 2.5592,
-      "mean_token_accuracy": 0.7294139194139193,
-      "mean_token_accuracy_utility": 0.641233211233211,
+      "loss": 2.5418,
+      "mean_token_accuracy": 0.7332234432234432,
+      "mean_token_accuracy_utility": 0.6402747252747251,
       "step": 250
     },
     {
       "epoch": 0.09436580627255066,
       "grad_norm": 0.0,
       "learning_rate": 1.8819188191881922e-05,
-      "loss": 2.5514,
-      "mean_token_accuracy": 0.7368803418803418,
-      "mean_token_accuracy_utility": 0.6387193224816986,
+      "loss": 2.5339,
+      "mean_token_accuracy": 0.7414102564102565,
+      "mean_token_accuracy_utility": 0.6379966623035929,
       "step": 255
     },
     {
       "epoch": 0.09621611619946341,
       "grad_norm": 0.0,
       "learning_rate": 1.918819188191882e-05,
-      "loss": 2.4914,
-      "mean_token_accuracy": 0.7332234432234431,
-      "mean_token_accuracy_utility": 0.6520573870573869,
+      "loss": 2.4701,
+      "mean_token_accuracy": 0.7384126984126983,
+      "mean_token_accuracy_utility": 0.6513186813186812,
       "step": 260
     },
     {
       "epoch": 0.09806642612637617,
       "grad_norm": 0.0,
       "learning_rate": 1.955719557195572e-05,
-      "loss": 2.4729,
-      "mean_token_accuracy": 0.7326862026862024,
-      "mean_token_accuracy_utility": 0.6482967032967032,
+      "loss": 2.4531,
+      "mean_token_accuracy": 0.7377960927960927,
+      "mean_token_accuracy_utility": 0.6476739926739926,
       "step": 265
     },
     {
       "epoch": 0.09991673605328892,
       "grad_norm": 0.0,
       "learning_rate": 1.9926199261992623e-05,
-      "loss": 2.6749,
-      "mean_token_accuracy": 0.7192246642246639,
-      "mean_token_accuracy_utility": 0.6245054945054943,
+      "loss": 2.6544,
+      "mean_token_accuracy": 0.7239804639804639,
+      "mean_token_accuracy_utility": 0.6240781440781439,
       "step": 270
     },
     {
       "epoch": 0.10176704598020168,
       "grad_norm": 0.0,
       "learning_rate": 1.9999866396188624e-05,
-      "loss": 2.5045,
-      "mean_token_accuracy": 0.7275824175824174,
-      "mean_token_accuracy_utility": 0.6468864468864466,
+      "loss": 2.4835,
+      "mean_token_accuracy": 0.7331379731379729,
+      "mean_token_accuracy_utility": 0.646233211233211,
       "step": 275
     },
     {
       "epoch": 0.10361735590711445,
       "grad_norm": 0.0,
       "learning_rate": 1.9999323636823398e-05,
-      "loss": 2.5939,
-      "mean_token_accuracy": 0.7384065934065933,
-      "mean_token_accuracy_utility": 0.6306532356532355,
+      "loss": 2.5739,
+      "mean_token_accuracy": 0.7436813186813186,
+      "mean_token_accuracy_utility": 0.6297741147741147,
       "step": 280
     },
     {
       "epoch": 0.1054676658340272,
       "grad_norm": 0.0,
       "learning_rate": 1.9998363394309497e-05,
-      "loss": 2.4263,
-      "mean_token_accuracy": 0.7558669108669107,
-      "mean_token_accuracy_utility": 0.6415750915750914,
+      "loss": 2.4084,
+      "mean_token_accuracy": 0.7600305250305249,
+      "mean_token_accuracy_utility": 0.6405799755799755,
       "step": 285
     },
     {
       "epoch": 0.10731797576093996,
       "grad_norm": 0.0,
       "learning_rate": 1.9996985708738146e-05,
-      "loss": 2.3333,
-      "mean_token_accuracy": 0.7557326007326008,
-      "mean_token_accuracy_utility": 0.6545421245421243,
+      "loss": 2.3149,
+      "mean_token_accuracy": 0.7608547008547009,
+      "mean_token_accuracy_utility": 0.6535653235653234,
       "step": 290
     },
     {
       "epoch": 0.10916828568785271,
       "grad_norm": 0.0,
       "learning_rate": 1.999519063762928e-05,
-      "loss": 2.4943,
-      "mean_token_accuracy": 0.735940170940171,
-      "mean_token_accuracy_utility": 0.6419299996067473,
+      "loss": 2.473,
+      "mean_token_accuracy": 0.7412271062271062,
+      "mean_token_accuracy_utility": 0.641165806786388,
       "step": 295
     },
     {
       "epoch": 0.11101859561476547,
       "grad_norm": 0.0,
       "learning_rate": 1.9992978255929168e-05,
-      "loss": 2.4449,
-      "mean_token_accuracy": 0.7292185592185593,
-      "mean_token_accuracy_utility": 0.6530525030525027,
+      "loss": 2.4255,
+      "mean_token_accuracy": 0.7351587301587303,
+      "mean_token_accuracy_utility": 0.6521794871794869,
       "step": 300
     },
     {
       "epoch": 0.11101859561476547,
-      "eval_mean_token_accuracy": 0.7285399785399784,
-      "eval_reasoning_loss": 0.9666416049003601,
-      "eval_reasoning_runtime": 27.5197,
-      "eval_reasoning_samples_per_second": 4.688,
-      "eval_reasoning_steps_per_second": 1.199,
+      "eval_mean_token_accuracy": 0.7336922336922334,
+      "eval_reasoning_loss": 0.9379008412361145,
+      "eval_reasoning_runtime": 27.5957,
+      "eval_reasoning_samples_per_second": 4.675,
+      "eval_reasoning_steps_per_second": 1.196,
       "step": 300
     },
     {
       "epoch": 0.11101859561476547,
-      "eval_mean_token_accuracy": 0.6496320411390472,
-      "eval_utility_loss": 1.493607759475708,
-      "eval_utility_runtime": 10.3556,
-      "eval_utility_samples_per_second": 4.732,
-      "eval_utility_steps_per_second": 1.255,
+      "eval_mean_token_accuracy": 0.6485167645876826,
+      "eval_utility_loss": 1.5009737014770508,
+      "eval_utility_runtime": 10.3867,
+      "eval_utility_samples_per_second": 4.718,
+      "eval_utility_steps_per_second": 1.252,
       "step": 300
     },
     {
       "epoch": 0.11286890554167824,
       "grad_norm": 0.0,
       "learning_rate": 1.999034865600726e-05,
-      "loss": 2.5426,
-      "mean_token_accuracy": 0.7203357753357752,
-      "mean_token_accuracy_utility": 0.6469291819291817,
+      "loss": 2.5232,
+      "mean_token_accuracy": 0.7248962148962149,
+      "mean_token_accuracy_utility": 0.6458669108669107,
       "step": 305
     },
     {
       "epoch": 0.11471921546859098,
       "grad_norm": 0.0,
       "learning_rate": 1.9987301947652354e-05,
-      "loss": 2.4146,
-      "mean_token_accuracy": 0.7401404151404151,
-      "mean_token_accuracy_utility": 0.6476007326007325,
+      "loss": 2.3944,
+      "mean_token_accuracy": 0.7449389499389499,
+      "mean_token_accuracy_utility": 0.6469169719169717,
       "step": 310
     },
     {
       "epoch": 0.11656952539550375,
       "grad_norm": 0.0,
       "learning_rate": 1.998383825806799e-05,
-      "loss": 2.4807,
-      "mean_token_accuracy": 0.7396886446886447,
-      "mean_token_accuracy_utility": 0.6409523809523808,
+      "loss": 2.4591,
+      "mean_token_accuracy": 0.7448962148962149,
+      "mean_token_accuracy_utility": 0.6400244200244198,
       "step": 315
     },
     {
       "epoch": 0.11841983532241651,
       "grad_norm": 0.0,
       "learning_rate": 1.9979957731867143e-05,
-      "loss": 2.4696,
-      "mean_token_accuracy": 0.7448534798534802,
-      "mean_token_accuracy_utility": 0.6369536019536017,
+      "loss": 2.4502,
+      "mean_token_accuracy": 0.7496642246642248,
+      "mean_token_accuracy_utility": 0.6358424908424907,
       "step": 320
     },
     {
       "epoch": 0.12027014524932926,
       "grad_norm": 0.0,
       "learning_rate": 1.9975660531066215e-05,
-      "loss": 2.4087,
-      "mean_token_accuracy": 0.7227960927960928,
-      "mean_token_accuracy_utility": 0.6600549450549449,
+      "loss": 2.384,
+      "mean_token_accuracy": 0.7281501831501832,
+      "mean_token_accuracy_utility": 0.6594261294261293,
       "step": 325
     },
     {
       "epoch": 0.12212045517624202,
       "grad_norm": 0.0,
       "learning_rate": 1.9970946835078227e-05,
-      "loss": 2.4029,
-      "mean_token_accuracy": 0.7436385836385837,
-      "mean_token_accuracy_utility": 0.6440476190476189,
+      "loss": 2.3798,
+      "mean_token_accuracy": 0.749108669108669,
+      "mean_token_accuracy_utility": 0.6437179487179485,
       "step": 330
     },
     {
       "epoch": 0.12397076510315477,
       "grad_norm": 0.0,
       "learning_rate": 1.9965816840705355e-05,
-      "loss": 2.4082,
-      "mean_token_accuracy": 0.7511599511599513,
-      "mean_token_accuracy_utility": 0.641318681318681,
+      "loss": 2.386,
+      "mean_token_accuracy": 0.7557081807081809,
+      "mean_token_accuracy_utility": 0.6405860805860802,
       "step": 335
     },
     {
       "epoch": 0.12582107503006754,
       "grad_norm": 0.0,
       "learning_rate": 1.9960270762130705e-05,
-      "loss": 2.4537,
-      "mean_token_accuracy": 0.739529245374983,
-      "mean_token_accuracy_utility": 0.6395299145299143,
+      "loss": 2.4295,
+      "mean_token_accuracy": 0.7452674223986809,
+      "mean_token_accuracy_utility": 0.6385531135531133,
       "step": 340
     },
     {
       "epoch": 0.1276713849569803,
       "grad_norm": 0.0,
       "learning_rate": 1.9954308830909372e-05,
-      "loss": 2.4737,
-      "mean_token_accuracy": 0.7475702075702078,
-      "mean_token_accuracy_utility": 0.6308913308913306,
+      "loss": 2.4495,
+      "mean_token_accuracy": 0.7528510378510379,
+      "mean_token_accuracy_utility": 0.6302869352869351,
       "step": 345
     },
     {
       "epoch": 0.12952169488389306,
       "grad_norm": 0.0,
       "learning_rate": 1.9947931295958778e-05,
-      "loss": 2.4592,
-      "mean_token_accuracy": 0.7438766788766786,
-      "mean_token_accuracy_utility": 0.6352686202686201,
+      "loss": 2.4366,
+      "mean_token_accuracy": 0.7487973137973137,
+      "mean_token_accuracy_utility": 0.6340476190476189,
       "step": 350
     },
     {
       "epoch": 0.1313720048108058,
       "grad_norm": 0.0,
       "learning_rate": 1.9941138423548266e-05,
-      "loss": 2.3309,
-      "mean_token_accuracy": 0.7435409035409036,
-      "mean_token_accuracy_utility": 0.6567948717948717,
+      "loss": 2.3102,
+      "mean_token_accuracy": 0.747899877899878,
+      "mean_token_accuracy_utility": 0.6557692307692305,
       "step": 355
     },
     {
       "epoch": 0.13322231473771856,
       "grad_norm": 0.0,
       "learning_rate": 1.9933930497287996e-05,
-      "loss": 2.4314,
-      "mean_token_accuracy": 0.7449999999999999,
-      "mean_token_accuracy_utility": 0.6419047619047618,
+      "loss": 2.4088,
+      "mean_token_accuracy": 0.7495787545787546,
+      "mean_token_accuracy_utility": 0.6408363858363858,
       "step": 360
     },
     {
       "epoch": 0.13507262466463132,
       "grad_norm": 0.0,
       "learning_rate": 1.9926307818117098e-05,
-      "loss": 2.3883,
-      "mean_token_accuracy": 0.7443101343101342,
-      "mean_token_accuracy_utility": 0.6529365079365077,
+      "loss": 2.3656,
+      "mean_token_accuracy": 0.7492735042735041,
+      "mean_token_accuracy_utility": 0.6517887667887666,
       "step": 365
     },
     {
       "epoch": 0.1369229345915441,
       "grad_norm": 0.0,
       "learning_rate": 1.9918270704291104e-05,
-      "loss": 2.4265,
-      "mean_token_accuracy": 0.7486813186813185,
-      "mean_token_accuracy_utility": 0.6425579975579975,
+      "loss": 2.4022,
+      "mean_token_accuracy": 0.7535958485958487,
+      "mean_token_accuracy_utility": 0.6411599511599511,
       "step": 370
     },
     {
       "epoch": 0.13877324451845685,
       "grad_norm": 0.0,
       "learning_rate": 1.9909819491368677e-05,
-      "loss": 2.3925,
-      "mean_token_accuracy": 0.765775335775336,
-      "mean_token_accuracy_utility": 0.6356166056166054,
+      "loss": 2.3728,
+      "mean_token_accuracy": 0.7704700854700857,
+      "mean_token_accuracy_utility": 0.6338766788766786,
       "step": 375
     },
     {
       "epoch": 0.1406235544453696,
       "grad_norm": 0.0,
       "learning_rate": 1.990095453219757e-05,
-      "loss": 2.2384,
-      "mean_token_accuracy": 0.7504395604395603,
-      "mean_token_accuracy_utility": 0.6720085470085468,
+      "loss": 2.2161,
+      "mean_token_accuracy": 0.7549755799755798,
+      "mean_token_accuracy_utility": 0.6697680097680097,
       "step": 380
     },
     {
       "epoch": 0.14247386437228235,
       "grad_norm": 0.0,
       "learning_rate": 1.989167619689993e-05,
-      "loss": 2.4041,
-      "mean_token_accuracy": 0.7417155067155065,
-      "mean_token_accuracy_utility": 0.6489844924442513,
+      "loss": 2.3799,
+      "mean_token_accuracy": 0.7464529914529912,
+      "mean_token_accuracy_utility": 0.6475193290780943,
       "step": 385
     },
     {
       "epoch": 0.1443241742991951,
       "grad_norm": 0.0,
       "learning_rate": 1.988198487285682e-05,
-      "loss": 2.4011,
-      "mean_token_accuracy": 0.7290476190476188,
-      "mean_token_accuracy_utility": 0.6638705738705737,
+      "loss": 2.3778,
+      "mean_token_accuracy": 0.734139194139194,
+      "mean_token_accuracy_utility": 0.6625641025641024,
       "step": 390
     },
     {
       "epoch": 0.14617448422610788,
       "grad_norm": 0.0,
       "learning_rate": 1.9871880964692055e-05,
-      "loss": 2.3785,
-      "mean_token_accuracy": 0.7425152625152626,
-      "mean_token_accuracy_utility": 0.6560256410256409,
+      "loss": 2.3582,
+      "mean_token_accuracy": 0.7468986568986568,
+      "mean_token_accuracy_utility": 0.6548290598290597,
       "step": 395
     },
     {
       "epoch": 0.14802479415302064,
       "grad_norm": 0.0,
       "learning_rate": 1.9861364894255306e-05,
-      "loss": 2.4065,
-      "mean_token_accuracy": 0.7506654456654456,
-      "mean_token_accuracy_utility": 0.6461233211233209,
+      "loss": 2.3836,
+      "mean_token_accuracy": 0.7554456654456656,
+      "mean_token_accuracy_utility": 0.6450427350427349,
       "step": 400
     },
     {
       "epoch": 0.14802479415302064,
-      "eval_mean_token_accuracy": 0.7347636024106613,
-      "eval_reasoning_loss": 0.929710328578949,
-      "eval_reasoning_runtime": 27.5306,
-      "eval_reasoning_samples_per_second": 4.686,
-      "eval_reasoning_steps_per_second": 1.199,
+      "eval_mean_token_accuracy": 0.7394219070689659,
+      "eval_reasoning_loss": 0.9023851156234741,
+      "eval_reasoning_runtime": 27.6097,
+      "eval_reasoning_samples_per_second": 4.672,
+      "eval_reasoning_steps_per_second": 1.195,
       "step": 400
     },
     {
       "epoch": 0.14802479415302064,
-      "eval_mean_token_accuracy": 0.6547456515143104,
-      "eval_utility_loss": 1.4680131673812866,
-      "eval_utility_runtime": 10.3646,
-      "eval_utility_samples_per_second": 4.728,
-      "eval_utility_steps_per_second": 1.254,
+      "eval_mean_token_accuracy": 0.6537331841735617,
+      "eval_utility_loss": 1.4724605083465576,
+      "eval_utility_runtime": 10.3959,
+      "eval_utility_samples_per_second": 4.713,
+      "eval_utility_steps_per_second": 1.25,
       "step": 400
     },
     {
       "epoch": 0.1498751040799334,
       "grad_norm": 0.0,
       "learning_rate": 1.985043710060449e-05,
-      "loss": 2.295,
-      "mean_token_accuracy": 0.7385836385836384,
-      "mean_token_accuracy_utility": 0.669181929181929,
+      "loss": 2.2729,
+      "mean_token_accuracy": 0.7431929181929182,
+      "mean_token_accuracy_utility": 0.668241758241758,
       "step": 405
     },
     {
       "epoch": 0.15172541400684614,
       "grad_norm": 0.0,
       "learning_rate": 1.9839098039987435e-05,
-      "loss": 2.3503,
-      "mean_token_accuracy": 0.752808302808303,
-      "mean_token_accuracy_utility": 0.6557020757020755,
+      "loss": 2.3298,
+      "mean_token_accuracy": 0.7576007326007327,
+      "mean_token_accuracy_utility": 0.6548290598290596,
       "step": 410
     },
     {
       "epoch": 0.1535757239337589,
       "grad_norm": 0.0,
       "learning_rate": 1.9827348185822834e-05,
-      "loss": 2.4131,
-      "mean_token_accuracy": 0.7482539682539682,
-      "mean_token_accuracy_utility": 0.6453724053724051,
+      "loss": 2.3909,
+      "mean_token_accuracy": 0.7528815628815628,
+      "mean_token_accuracy_utility": 0.644279609279609,
       "step": 415
     },
     {
       "epoch": 0.15542603386067166,
       "grad_norm": 0.0,
       "learning_rate": 1.981518802868048e-05,
-      "loss": 2.3045,
-      "mean_token_accuracy": 0.743003663003663,
-      "mean_token_accuracy_utility": 0.6717277167277166,
+      "loss": 2.2837,
+      "mean_token_accuracy": 0.7474725274725273,
+      "mean_token_accuracy_utility": 0.6707020757020756,
       "step": 420
     },
     {
       "epoch": 0.15727634378758443,
       "grad_norm": 0.0,
       "learning_rate": 1.9802618076260784e-05,
-      "loss": 2.3354,
-      "mean_token_accuracy": 0.7554273504273501,
-      "mean_token_accuracy_utility": 0.6568620268620267,
+      "loss": 2.3147,
+      "mean_token_accuracy": 0.760103785103785,
+      "mean_token_accuracy_utility": 0.6559096459096457,
       "step": 425
     },
     {
       "epoch": 0.1591266537144972,
       "grad_norm": 0.0,
       "learning_rate": 1.9789638853373563e-05,
-      "loss": 2.2433,
-      "mean_token_accuracy": 0.7508058608058608,
-      "mean_token_accuracy_utility": 0.672173382173382,
+      "loss": 2.2234,
+      "mean_token_accuracy": 0.7556288156288156,
+      "mean_token_accuracy_utility": 0.6708791208791207,
       "step": 430
     },
     {
       "epoch": 0.16097696364140993,
       "grad_norm": 0.0,
       "learning_rate": 1.9776250901916168e-05,
-      "loss": 2.4012,
-      "mean_token_accuracy": 0.7284615384615384,
-      "mean_token_accuracy_utility": 0.660286935286935,
+      "loss": 2.3813,
+      "mean_token_accuracy": 0.7326312576312576,
+      "mean_token_accuracy_utility": 0.6588644688644687,
       "step": 435
     },
     {
       "epoch": 0.1628272735683227,
       "grad_norm": 0.0,
       "learning_rate": 1.9762454780850807e-05,
-      "loss": 2.3365,
-      "mean_token_accuracy": 0.7485958485958486,
-      "mean_token_accuracy_utility": 0.6566422466422466,
+      "loss": 2.3164,
+      "mean_token_accuracy": 0.753034188034188,
+      "mean_token_accuracy_utility": 0.6559890109890109,
       "step": 440
     },
     {
       "epoch": 0.16467758349523545,
       "grad_norm": 0.0,
       "learning_rate": 1.9748251066181247e-05,
-      "loss": 2.3806,
-      "mean_token_accuracy": 0.7539194139194139,
-      "mean_token_accuracy_utility": 0.6452197802197801,
+      "loss": 2.3602,
+      "mean_token_accuracy": 0.7580219780219781,
+      "mean_token_accuracy_utility": 0.644285714285714,
       "step": 445
     },
     {
       "epoch": 0.16652789342214822,
       "grad_norm": 0.0,
       "learning_rate": 1.973364035092875e-05,
-      "loss": 2.469,
-      "mean_token_accuracy": 0.7352258852258851,
-      "mean_token_accuracy_utility": 0.6419169719169717,
+      "loss": 2.4477,
+      "mean_token_accuracy": 0.7395238095238094,
+      "mean_token_accuracy_utility": 0.640286935286935,
       "step": 450
     },
     {
       "epoch": 0.16837820334906098,
       "grad_norm": 0.0,
       "learning_rate": 1.971862324510732e-05,
-      "loss": 2.4203,
-      "mean_token_accuracy": 0.7307570207570208,
-      "mean_token_accuracy_utility": 0.6586691086691084,
+      "loss": 2.4002,
+      "mean_token_accuracy": 0.7348778998778999,
+      "mean_token_accuracy_utility": 0.6574847374847373,
       "step": 455
     },
     {
       "epoch": 0.17022851327597371,
       "grad_norm": 0.0,
       "learning_rate": 1.9703200375698223e-05,
-      "loss": 2.4337,
-      "mean_token_accuracy": 0.7374481074481073,
-      "mean_token_accuracy_utility": 0.6470146520146518,
+      "loss": 2.4145,
+      "mean_token_accuracy": 0.7414835164835166,
+      "mean_token_accuracy_utility": 0.6460439560439558,
       "step": 460
     },
     {
       "epoch": 0.17207882320288648,
       "grad_norm": 0.0,
       "learning_rate": 1.968737238662382e-05,
-      "loss": 2.2523,
-      "mean_token_accuracy": 0.7598351648351651,
-      "mean_token_accuracy_utility": 0.6671721430229887,
+      "loss": 2.2344,
+      "mean_token_accuracy": 0.763913308913309,
+      "mean_token_accuracy_utility": 0.6664038838603163,
       "step": 465
     },
     {
       "epoch": 0.17392913312979924,
       "grad_norm": 0.0,
       "learning_rate": 1.9671139938720678e-05,
-      "loss": 2.3539,
-      "mean_token_accuracy": 0.7533760683760683,
-      "mean_token_accuracy_utility": 0.650152625152625,
+      "loss": 2.3368,
+      "mean_token_accuracy": 0.7572344322344323,
+      "mean_token_accuracy_utility": 0.6493467643467642,
       "step": 470
     },
     {
       "epoch": 0.175779443056712,
       "grad_norm": 0.0,
       "learning_rate": 1.9654503709711984e-05,
-      "loss": 2.3861,
-      "mean_token_accuracy": 0.7417887667887667,
-      "mean_token_accuracy_utility": 0.6546825396825395,
+      "loss": 2.3668,
+      "mean_token_accuracy": 0.7465262515262514,
+      "mean_token_accuracy_utility": 0.6533150183150181,
       "step": 475
     },
     {
       "epoch": 0.17762975298362477,
       "grad_norm": 0.0,
       "learning_rate": 1.963746439417924e-05,
-      "loss": 2.3862,
-      "mean_token_accuracy": 0.7587789987789988,
-      "mean_token_accuracy_utility": 0.649145299145299,
+      "loss": 2.3685,
+      "mean_token_accuracy": 0.7627167277167277,
+      "mean_token_accuracy_utility": 0.648186813186813,
       "step": 480
     },
     {
       "epoch": 0.1794800629105375,
       "grad_norm": 0.0,
       "learning_rate": 1.962002270353328e-05,
-      "loss": 2.3155,
-      "mean_token_accuracy": 0.7590781440781442,
-      "mean_token_accuracy_utility": 0.6514896214896214,
+      "loss": 2.2974,
+      "mean_token_accuracy": 0.7633577533577535,
+      "mean_token_accuracy_utility": 0.6504822954822954,
       "step": 485
     },
     {
       "epoch": 0.18133037283745027,
       "grad_norm": 0.0,
       "learning_rate": 1.960217936598454e-05,
-      "loss": 2.2053,
-      "mean_token_accuracy": 0.7601343101343103,
-      "mean_token_accuracy_utility": 0.6665018315018313,
+      "loss": 2.1864,
+      "mean_token_accuracy": 0.7645604395604397,
+      "mean_token_accuracy_utility": 0.6655799755799754,
       "step": 490
     },
     {
       "epoch": 0.18318068276436303,
       "grad_norm": 0.0,
       "learning_rate": 1.958393512651269e-05,
-      "loss": 2.3005,
-      "mean_token_accuracy": 0.7678937728937731,
-      "mean_token_accuracy_utility": 0.6465238319376145,
+      "loss": 2.2836,
+      "mean_token_accuracy": 0.7717216117216118,
+      "mean_token_accuracy_utility": 0.6457377044662629,
       "step": 495
     },
     {
       "epoch": 0.1850309926912758,
       "grad_norm": 0.0,
       "learning_rate": 1.956529074683551e-05,
-      "loss": 2.3661,
-      "mean_token_accuracy": 0.7472710622710622,
-      "mean_token_accuracy_utility": 0.6517277167277165,
+      "loss": 2.3474,
+      "mean_token_accuracy": 0.7521062271062272,
+      "mean_token_accuracy_utility": 0.6508852258852256,
       "step": 500
     },
     {
       "epoch": 0.1850309926912758,
-      "eval_mean_token_accuracy": 0.738359570712512,
-      "eval_reasoning_loss": 0.9078736305236816,
-      "eval_reasoning_runtime": 27.5335,
-      "eval_reasoning_samples_per_second": 4.685,
-      "eval_reasoning_steps_per_second": 1.199,
+      "eval_mean_token_accuracy": 0.7429813106283695,
+      "eval_reasoning_loss": 0.8834718465805054,
+      "eval_reasoning_runtime": 27.6075,
+      "eval_reasoning_samples_per_second": 4.673,
+      "eval_reasoning_steps_per_second": 1.195,
       "step": 500
     },
     {
       "epoch": 0.1850309926912758,
-      "eval_mean_token_accuracy": 0.6575623340850677,
-      "eval_utility_loss": 1.4581704139709473,
-      "eval_utility_runtime": 10.3652,
-      "eval_utility_samples_per_second": 4.727,
-      "eval_utility_steps_per_second": 1.254,
+      "eval_mean_token_accuracy": 0.6567363164589358,
+      "eval_utility_loss": 1.4627714157104492,
+      "eval_utility_runtime": 10.3844,
+      "eval_utility_samples_per_second": 4.719,
+      "eval_utility_steps_per_second": 1.252,
       "step": 500
     },
     {
       "epoch": 0.18688130261818856,
       "grad_norm": 0.0,
       "learning_rate": 1.9546247005377065e-05,
-      "loss": 2.2533,
-      "mean_token_accuracy": 0.7527411477411476,
-      "mean_token_accuracy_utility": 0.668150183150183,
+      "loss": 2.2352,
+      "mean_token_accuracy": 0.756813186813187,
+      "mean_token_accuracy_utility": 0.6669963369963369,
       "step": 505
     },
     {
       "epoch": 0.18873161254510132,
       "grad_norm": 0.0,
       "learning_rate": 1.952680469723526e-05,
-      "loss": 2.3449,
-      "mean_token_accuracy": 0.7547069597069598,
-      "mean_token_accuracy_utility": 0.6521123321123319,
+      "loss": 2.3279,
+      "mean_token_accuracy": 0.7585531135531137,
+      "mean_token_accuracy_utility": 0.6511660561660559,
       "step": 510
     },
     {
       "epoch": 0.19058192247201405,
       "grad_norm": 0.0,
       "learning_rate": 1.9506964634148597e-05,
-      "loss": 2.3947,
-      "mean_token_accuracy": 0.7467216117216117,
-      "mean_token_accuracy_utility": 0.6490415140415139,
+      "loss": 2.3766,
+      "mean_token_accuracy": 0.7520940170940171,
+      "mean_token_accuracy_utility": 0.6479731379731378,
       "step": 515
     },
     {
       "epoch": 0.19243223239892682,
       "grad_norm": 0.0,
       "learning_rate": 1.9486727644462306e-05,
-      "loss": 2.3414,
-      "mean_token_accuracy": 0.7484798534798537,
-      "mean_token_accuracy_utility": 0.660140415140415,
+      "loss": 2.3234,
+      "mean_token_accuracy": 0.7526007326007327,
+      "mean_token_accuracy_utility": 0.6591147741147739,
       "step": 520
     },
     {
       "epoch": 0.19428254232583958,
       "grad_norm": 0.0,
       "learning_rate": 1.9466094573093744e-05,
-      "loss": 2.3219,
-      "mean_token_accuracy": 0.7676556776556775,
-      "mean_token_accuracy_utility": 0.6476739926739925,
+      "loss": 2.3047,
+      "mean_token_accuracy": 0.7714224664224663,
+      "mean_token_accuracy_utility": 0.6463980463980462,
       "step": 525
     },
     {
       "epoch": 0.19613285225275234,
       "grad_norm": 0.0,
       "learning_rate": 1.9445066281497144e-05,
-      "loss": 2.2747,
-      "mean_token_accuracy": 0.7598840048840049,
-      "mean_token_accuracy_utility": 0.6588461538461535,
+      "loss": 2.2582,
+      "mean_token_accuracy": 0.7641330891330891,
+      "mean_token_accuracy_utility": 0.6576984126984124,
       "step": 530
     },
     {
       "epoch": 0.1979831621796651,
       "grad_norm": 0.0,
       "learning_rate": 1.9423643647627625e-05,
-      "loss": 2.357,
-      "mean_token_accuracy": 0.7461843711843713,
-      "mean_token_accuracy_utility": 0.6574321718502176,
+      "loss": 2.3405,
+      "mean_token_accuracy": 0.7503357753357756,
+      "mean_token_accuracy_utility": 0.6566135322528808,
       "step": 535
     },
     {
       "epoch": 0.19983347210657784,
       "grad_norm": 0.0,
       "learning_rate": 1.940182756590454e-05,
-      "loss": 2.2512,
-      "mean_token_accuracy": 0.7514163614163614,
-      "mean_token_accuracy_utility": 0.6684371184371182,
+      "loss": 2.233,
+      "mean_token_accuracy": 0.7558669108669109,
+      "mean_token_accuracy_utility": 0.6672344322344321,
       "step": 540
     },
     {
       "epoch": 0.2016837820334906,
       "grad_norm": 0.0,
       "learning_rate": 1.9379618947174155e-05,
-      "loss": 2.3319,
-      "mean_token_accuracy": 0.7552625152625151,
-      "mean_token_accuracy_utility": 0.6598595848595846,
+      "loss": 2.3151,
+      "mean_token_accuracy": 0.7593589743589744,
+      "mean_token_accuracy_utility": 0.6585042735042733,
       "step": 545
     },
     {
       "epoch": 0.20353409196040337,
       "grad_norm": 0.0,
       "learning_rate": 1.935701871867158e-05,
-      "loss": 2.291,
-      "mean_token_accuracy": 0.75495115995116,
-      "mean_token_accuracy_utility": 0.6590781440781439,
+      "loss": 2.274,
+      "mean_token_accuracy": 0.7591636141636143,
+      "mean_token_accuracy_utility": 0.6581379731379731,
       "step": 550
     },
     {
       "epoch": 0.20538440188731613,
       "grad_norm": 0.0,
       "learning_rate": 1.9334027823982103e-05,
-      "loss": 2.2772,
-      "mean_token_accuracy": 0.7548708378708378,
-      "mean_token_accuracy_utility": 0.6618620268620267,
+      "loss": 2.2597,
+      "mean_token_accuracy": 0.7592370392370392,
+      "mean_token_accuracy_utility": 0.6610134310134308,
       "step": 555
     },
     {
       "epoch": 0.2072347118142289,
       "grad_norm": 0.0,
       "learning_rate": 1.9310647223001752e-05,
-      "loss": 2.3329,
-      "mean_token_accuracy": 0.7393772893772891,
-      "mean_token_accuracy_utility": 0.6618315018315016,
+      "loss": 2.3149,
+      "mean_token_accuracy": 0.7438705738705738,
+      "mean_token_accuracy_utility": 0.6606349206349205,
       "step": 560
     },
     {
       "epoch": 0.20908502174114163,
       "grad_norm": 0.0,
       "learning_rate": 1.9286877891897244e-05,
-      "loss": 2.3174,
-      "mean_token_accuracy": 0.7682295482295484,
-      "mean_token_accuracy_utility": 0.6484798534798533,
+      "loss": 2.2997,
+      "mean_token_accuracy": 0.772918192918193,
+      "mean_token_accuracy_utility": 0.6476617826617825,
       "step": 565
     },
     {
       "epoch": 0.2109353316680544,
       "grad_norm": 0.0,
       "learning_rate": 1.9262720823065217e-05,
-      "loss": 2.369,
-      "mean_token_accuracy": 0.7386507936507936,
-      "mean_token_accuracy_utility": 0.6569474969474969,
+      "loss": 2.3511,
+      "mean_token_accuracy": 0.7427411477411479,
+      "mean_token_accuracy_utility": 0.6558791208791207,
       "step": 570
     },
     {
       "epoch": 0.21278564159496716,
       "grad_norm": 0.0,
       "learning_rate": 1.923817702509081e-05,
-      "loss": 2.3106,
-      "mean_token_accuracy": 0.7577960927960927,
-      "mean_token_accuracy_utility": 0.6536935286935285,
+      "loss": 2.2941,
+      "mean_token_accuracy": 0.7615811965811967,
+      "mean_token_accuracy_utility": 0.6527533577533575,
       "step": 575
     },
     {
       "epoch": 0.21463595152187992,
       "grad_norm": 0.0,
       "learning_rate": 1.9213247522705532e-05,
-      "loss": 2.2852,
-      "mean_token_accuracy": 0.7480158730158729,
-      "mean_token_accuracy_utility": 0.6678815628815626,
+      "loss": 2.2676,
+      "mean_token_accuracy": 0.7521001221001221,
+      "mean_token_accuracy_utility": 0.6669413919413917,
       "step": 580
     },
     {
       "epoch": 0.21648626144879268,
       "grad_norm": 0.0,
       "learning_rate": 1.9187933356744504e-05,
-      "loss": 2.3053,
-      "mean_token_accuracy": 0.757960927960928,
-      "mean_token_accuracy_utility": 0.6589560439560438,
+      "loss": 2.288,
+      "mean_token_accuracy": 0.7616117216117216,
+      "mean_token_accuracy_utility": 0.658131868131868,
       "step": 585
     },
     {
       "epoch": 0.21833657137570542,
       "grad_norm": 0.0,
       "learning_rate": 1.9162235584102973e-05,
-      "loss": 2.2777,
-      "mean_token_accuracy": 0.7517826617826617,
-      "mean_token_accuracy_utility": 0.6597557997557997,
+      "loss": 2.2605,
+      "mean_token_accuracy": 0.756092796092796,
+      "mean_token_accuracy_utility": 0.6588949938949937,
       "step": 590
     },
     {
       "epoch": 0.22018688130261818,
       "grad_norm": 0.0,
       "learning_rate": 1.9136155277692215e-05,
-      "loss": 2.2694,
-      "mean_token_accuracy": 0.7567155067155068,
-      "mean_token_accuracy_utility": 0.6574481074481072,
+      "loss": 2.2529,
+      "mean_token_accuracy": 0.7612332112332113,
+      "mean_token_accuracy_utility": 0.6567094017094015,
       "step": 595
     },
     {
       "epoch": 0.22203719122953094,
       "grad_norm": 0.0,
       "learning_rate": 1.9109693526394722e-05,
-      "loss": 2.3196,
-      "mean_token_accuracy": 0.7531746031746032,
-      "mean_token_accuracy_utility": 0.6512393162393162,
+      "loss": 2.3027,
+      "mean_token_accuracy": 0.7576923076923078,
+      "mean_token_accuracy_utility": 0.6496886446886446,
       "step": 600
     },
     {
       "epoch": 0.22203719122953094,
-      "eval_mean_token_accuracy": 0.7407111407111407,
-      "eval_reasoning_loss": 0.8948262929916382,
-      "eval_reasoning_runtime": 27.5389,
-      "eval_reasoning_samples_per_second": 4.684,
-      "eval_reasoning_steps_per_second": 1.198,
+      "eval_mean_token_accuracy": 0.7450977127447715,
+      "eval_reasoning_loss": 0.8715410828590393,
+      "eval_reasoning_runtime": 27.5734,
+      "eval_reasoning_samples_per_second": 4.678,
+      "eval_reasoning_steps_per_second": 1.197,
       "step": 600
     },
     {
       "epoch": 0.22203719122953094,
-      "eval_mean_token_accuracy": 0.6583714699814212,
-      "eval_utility_loss": 1.4508330821990967,
-      "eval_utility_runtime": 10.3718,
-      "eval_utility_samples_per_second": 4.724,
-      "eval_utility_steps_per_second": 1.253,
+      "eval_mean_token_accuracy": 0.657560937457992,
+      "eval_utility_loss": 1.4556632041931152,
+      "eval_utility_runtime": 10.3854,
+      "eval_utility_samples_per_second": 4.718,
+      "eval_utility_steps_per_second": 1.252,
       "step": 600
     },
     {
       "epoch": 0.2238875011564437,
       "grad_norm": 0.0,
       "learning_rate": 1.9082851435018743e-05,
-      "loss": 2.251,
-      "mean_token_accuracy": 0.7498168498168496,
-      "mean_token_accuracy_utility": 0.6715262515262512,
+      "loss": 2.2353,
+      "mean_token_accuracy": 0.7537179487179487,
+      "mean_token_accuracy_utility": 0.6706898656898654,
       "step": 605
     },
     {
       "epoch": 0.22573781108335647,
       "grad_norm": 0.0,
       "learning_rate": 1.905563012425216e-05,
-      "loss": 2.3848,
-      "mean_token_accuracy": 0.7583943833943836,
-      "mean_token_accuracy_utility": 0.640323565323565,
+      "loss": 2.3683,
+      "mean_token_accuracy": 0.7624236874236876,
+      "mean_token_accuracy_utility": 0.6390598290598288,
       "step": 610
     },
     {
       "epoch": 0.2275881210102692,
       "grad_norm": 0.0,
       "learning_rate": 1.9028030730615696e-05,
-      "loss": 2.4284,
-      "mean_token_accuracy": 0.7476312576312576,
-      "mean_token_accuracy_utility": 0.6428998778998778,
+      "loss": 2.4114,
+      "mean_token_accuracy": 0.7514652014652013,
+      "mean_token_accuracy_utility": 0.6419169719169718,
       "step": 615
     },
     {
       "epoch": 0.22943843093718197,
       "grad_norm": 0.0,
       "learning_rate": 1.9000054406415467e-05,
-      "loss": 2.3465,
-      "mean_token_accuracy": 0.7581501831501832,
-      "mean_token_accuracy_utility": 0.6499572649572648,
+      "loss": 2.3291,
+      "mean_token_accuracy": 0.7625518925518926,
+      "mean_token_accuracy_utility": 0.6493467643467642,
       "step": 620
     },
     {
       "epoch": 0.23128874086409473,
       "grad_norm": 0.0,
       "learning_rate": 1.897170231969486e-05,
-      "loss": 2.319,
-      "mean_token_accuracy": 0.7520757020757023,
-      "mean_token_accuracy_utility": 0.6596459096459095,
+      "loss": 2.3022,
+      "mean_token_accuracy": 0.7567582417582417,
+      "mean_token_accuracy_utility": 0.6586996336996336,
       "step": 625
     },
     {
       "epoch": 0.2331390507910075,
       "grad_norm": 0.0,
       "learning_rate": 1.8942975654185788e-05,
-      "loss": 2.2645,
-      "mean_token_accuracy": 0.7573931623931623,
-      "mean_token_accuracy_utility": 0.6639377289377288,
+      "loss": 2.2479,
+      "mean_token_accuracy": 0.7615750915750916,
+      "mean_token_accuracy_utility": 0.6628449328449326,
       "step": 630
     },
     {
       "epoch": 0.23498936071792026,
       "grad_norm": 0.0,
       "learning_rate": 1.8913875609259246e-05,
-      "loss": 2.2555,
-      "mean_token_accuracy": 0.7589010989010989,
-      "mean_token_accuracy_utility": 0.6612759462759461,
+      "loss": 2.239,
+      "mean_token_accuracy": 0.762985347985348,
+      "mean_token_accuracy_utility": 0.660201465201465,
       "step": 635
     },
     {
       "epoch": 0.23683967064483302,
       "grad_norm": 0.0,
       "learning_rate": 1.8884403399875252e-05,
-      "loss": 2.3082,
-      "mean_token_accuracy": 0.7495482295482296,
-      "mean_token_accuracy_utility": 0.6644369519369518,
+      "loss": 2.2921,
+      "mean_token_accuracy": 0.7534981684981685,
+      "mean_token_accuracy_utility": 0.6633397158397156,
       "step": 640
     },
     {
       "epoch": 0.23868998057174576,
       "grad_norm": 0.0,
       "learning_rate": 1.8854560256532098e-05,
-      "loss": 2.195,
-      "mean_token_accuracy": 0.7702639360639361,
-      "mean_token_accuracy_utility": 0.6676068376068374,
+      "loss": 2.1802,
+      "mean_token_accuracy": 0.7746005439005439,
+      "mean_token_accuracy_utility": 0.6661355311355309,
       "step": 645
     },
     {
       "epoch": 0.24054029049865852,
       "grad_norm": 0.0,
       "learning_rate": 1.8824347425215016e-05,
-      "loss": 2.2965,
-      "mean_token_accuracy": 0.7428205128205129,
-      "mean_token_accuracy_utility": 0.6654334554334552,
+      "loss": 2.28,
+      "mean_token_accuracy": 0.7473931623931624,
+      "mean_token_accuracy_utility": 0.6643589743589742,
       "step": 650
     },
     {
       "epoch": 0.24239060042557128,
       "grad_norm": 0.0,
       "learning_rate": 1.8793766167344115e-05,
-      "loss": 2.2355,
-      "mean_token_accuracy": 0.7575946275946277,
-      "mean_token_accuracy_utility": 0.664120879120879,
+      "loss": 2.2201,
+      "mean_token_accuracy": 0.7614652014652017,
+      "mean_token_accuracy_utility": 0.663278388278388,
       "step": 655
     },
     {
       "epoch": 0.24424091035248405,
       "grad_norm": 0.0,
       "learning_rate": 1.8762817759721735e-05,
-      "loss": 2.3772,
-      "mean_token_accuracy": 0.7555494505494504,
-      "mean_token_accuracy_utility": 0.6438698932507355,
+      "loss": 2.3618,
+      "mean_token_accuracy": 0.7595604395604396,
+      "mean_token_accuracy_utility": 0.642784652528631,
       "step": 660
     },
     {
       "epoch": 0.2460912202793968,
       "grad_norm": 0.0,
       "learning_rate": 1.8731503494479132e-05,
-      "loss": 2.3524,
-      "mean_token_accuracy": 0.7487118437118438,
-      "mean_token_accuracy_utility": 0.6513003663003661,
+      "loss": 2.3363,
+      "mean_token_accuracy": 0.7528144078144079,
+      "mean_token_accuracy_utility": 0.6502686202686201,
       "step": 665
     },
     {
       "epoch": 0.24794153020630955,
       "grad_norm": 0.0,
       "learning_rate": 1.869982467902255e-05,
-      "loss": 2.3416,
-      "mean_token_accuracy": 0.7600793650793654,
-      "mean_token_accuracy_utility": 0.6469474969474969,
+      "loss": 2.3257,
+      "mean_token_accuracy": 0.7649084249084251,
+      "mean_token_accuracy_utility": 0.6463186813186812,
       "step": 670
     },
     {
       "epoch": 0.2497918401332223,
       "grad_norm": 0.0,
       "learning_rate": 1.8667782635978597e-05,
-      "loss": 2.3856,
-      "mean_token_accuracy": 0.7438217338217339,
-      "mean_token_accuracy_utility": 0.6483627569452309,
+      "loss": 2.3685,
+      "mean_token_accuracy": 0.7478510378510377,
+      "mean_token_accuracy_utility": 0.6475006045380433,
       "step": 675
     },
     {
       "epoch": 0.2516421500601351,
       "grad_norm": 0.0,
       "learning_rate": 1.8635378703139066e-05,
-      "loss": 2.2094,
-      "mean_token_accuracy": 0.7662332112332113,
-      "mean_token_accuracy_utility": 0.6636324786324784,
+      "loss": 2.194,
+      "mean_token_accuracy": 0.7697680097680099,
+      "mean_token_accuracy_utility": 0.6627838827838826,
       "step": 680
     },
     {
       "epoch": 0.25349245998704784,
       "grad_norm": 0.0,
       "learning_rate": 1.8602614233405047e-05,
-      "loss": 2.3027,
-      "mean_token_accuracy": 0.7563125763125763,
-      "mean_token_accuracy_utility": 0.6546938624620284,
+      "loss": 2.2882,
+      "mean_token_accuracy": 0.7603907203907204,
+      "mean_token_accuracy_utility": 0.6536777063766682,
       "step": 685
     },
     {
       "epoch": 0.2553427699139606,
       "grad_norm": 0.0,
       "learning_rate": 1.8569490594730474e-05,
-      "loss": 2.2929,
-      "mean_token_accuracy": 0.7523748473748475,
-      "mean_token_accuracy_utility": 0.6646886446886444,
+      "loss": 2.2787,
+      "mean_token_accuracy": 0.756727716727717,
+      "mean_token_accuracy_utility": 0.6629975579975578,
       "step": 690
     },
     {
       "epoch": 0.25719307984087336,
       "grad_norm": 0.0,
       "learning_rate": 1.853600917006497e-05,
-      "loss": 2.1942,
-      "mean_token_accuracy": 0.7684676434676435,
-      "mean_token_accuracy_utility": 0.6686324786324784,
+      "loss": 2.1788,
+      "mean_token_accuracy": 0.7727533577533577,
+      "mean_token_accuracy_utility": 0.6673382173382172,
       "step": 695
     },
     {
       "epoch": 0.2590433897677861,
       "grad_norm": 0.0,
       "learning_rate": 1.8502171357296144e-05,
-      "loss": 2.2224,
-      "mean_token_accuracy": 0.767881562881563,
-      "mean_token_accuracy_utility": 0.663241758241758,
+      "loss": 2.2062,
+      "mean_token_accuracy": 0.7722771672771673,
+      "mean_token_accuracy_utility": 0.6620879120879118,
       "step": 700
     },
     {
       "epoch": 0.2590433897677861,
-      "eval_mean_token_accuracy": 0.7428485457897223,
-      "eval_reasoning_loss": 0.8856377005577087,
-      "eval_reasoning_runtime": 27.5494,
-      "eval_reasoning_samples_per_second": 4.682,
-      "eval_reasoning_steps_per_second": 1.198,
+      "eval_mean_token_accuracy": 0.7465592142062727,
+      "eval_reasoning_loss": 0.8631607294082642,
+      "eval_reasoning_runtime": 27.6056,
+      "eval_reasoning_samples_per_second": 4.673,
+      "eval_reasoning_steps_per_second": 1.195,
       "step": 700
     },
     {
       "epoch": 0.2590433897677861,
-      "eval_mean_token_accuracy": 0.6589529032909075,
-      "eval_utility_loss": 1.4456983804702759,
-      "eval_utility_runtime": 10.3688,
-      "eval_utility_samples_per_second": 4.726,
-      "eval_utility_steps_per_second": 1.254,
+      "eval_mean_token_accuracy": 0.6583264113003888,
+      "eval_utility_loss": 1.4505559206008911,
+      "eval_utility_runtime": 10.3821,
+      "eval_utility_samples_per_second": 4.72,
+      "eval_utility_steps_per_second": 1.252,
       "step": 700
     },
     {
       "epoch": 0.2608936996946989,
       "grad_norm": 0.0,
       "learning_rate": 1.8467978569191216e-05,
-      "loss": 2.3189,
-      "mean_token_accuracy": 0.7533760683760683,
-      "mean_token_accuracy_utility": 0.6550854700854699,
+      "loss": 2.3034,
+      "mean_token_accuracy": 0.7577228327228327,
+      "mean_token_accuracy_utility": 0.6543467643467642,
       "step": 705
     },
     {
       "epoch": 0.2627440096216116,
       "grad_norm": 0.0,
       "learning_rate": 1.8433432233338027e-05,
-      "loss": 2.2326,
-      "mean_token_accuracy": 0.7642063492063493,
-      "mean_token_accuracy_utility": 0.668144078144078,
+      "loss": 2.2187,
+      "mean_token_accuracy": 0.7680036630036631,
+      "mean_token_accuracy_utility": 0.6670268620268619,
       "step": 710
     },
     {
       "epoch": 0.26459431954852436,
       "grad_norm": 0.0,
       "learning_rate": 1.8398533792085436e-05,
-      "loss": 2.2872,
-      "mean_token_accuracy": 0.7543467643467645,
-      "mean_token_accuracy_utility": 0.6596719288677007,
+      "loss": 2.2727,
+      "mean_token_accuracy": 0.7587118437118436,
+      "mean_token_accuracy_utility": 0.6588137827246281,
       "step": 715
     },
     {
       "epoch": 0.2664446294754371,
       "grad_norm": 0.0,
       "learning_rate": 1.8363284702483106e-05,
-      "loss": 2.3583,
-      "mean_token_accuracy": 0.7484188034188035,
-      "mean_token_accuracy_utility": 0.6488644688644687,
+      "loss": 2.3429,
+      "mean_token_accuracy": 0.7524358974358976,
+      "mean_token_accuracy_utility": 0.6479242979242977,
       "step": 720
     },
     {
       "epoch": 0.2682949394023499,
       "grad_norm": 0.0,
       "learning_rate": 1.832768643622067e-05,
-      "loss": 2.3036,
-      "mean_token_accuracy": 0.7634249084249085,
-      "mean_token_accuracy_utility": 0.6494078144078143,
+      "loss": 2.288,
+      "mean_token_accuracy": 0.767954822954823,
+      "mean_token_accuracy_utility": 0.6477777777777776,
       "step": 725
     },
     {
       "epoch": 0.27014524932926265,
       "grad_norm": 0.0,
       "learning_rate": 1.8291740479566286e-05,
-      "loss": 2.2047,
-      "mean_token_accuracy": 0.7613614163614165,
-      "mean_token_accuracy_utility": 0.6745238095238094,
+      "loss": 2.1905,
+      "mean_token_accuracy": 0.7653296703296706,
+      "mean_token_accuracy_utility": 0.673241758241758,
       "step": 730
     },
     {
       "epoch": 0.2719955592561754,
       "grad_norm": 0.0,
       "learning_rate": 1.825544833330457e-05,
-      "loss": 2.2285,
-      "mean_token_accuracy": 0.7584493284493286,
-      "mean_token_accuracy_utility": 0.6668681318681318,
+      "loss": 2.2141,
+      "mean_token_accuracy": 0.7619291819291819,
+      "mean_token_accuracy_utility": 0.6653601953601952,
       "step": 735
     },
     {
       "epoch": 0.2738458691830882,
       "grad_norm": 0.0,
       "learning_rate": 1.8218811512673958e-05,
-      "loss": 2.3097,
-      "mean_token_accuracy": 0.7449938949938948,
-      "mean_token_accuracy_utility": 0.6655921855921854,
+      "loss": 2.2931,
+      "mean_token_accuracy": 0.7493772893772893,
+      "mean_token_accuracy_utility": 0.6644627594627592,
       "step": 740
     },
     {
       "epoch": 0.27569617911000094,
       "grad_norm": 0.0,
       "learning_rate": 1.818183154730344e-05,
-      "loss": 2.297,
-      "mean_token_accuracy": 0.7546153846153848,
-      "mean_token_accuracy_utility": 0.6573137973137971,
+      "loss": 2.2829,
+      "mean_token_accuracy": 0.7584737484737485,
+      "mean_token_accuracy_utility": 0.656147741147741,
       "step": 745
     },
     {
       "epoch": 0.2775464890369137,
       "grad_norm": 0.0,
       "learning_rate": 1.8144509981148675e-05,
-      "loss": 2.2833,
-      "mean_token_accuracy": 0.7582661782661783,
-      "mean_token_accuracy_utility": 0.6567948717948716,
+      "loss": 2.268,
+      "mean_token_accuracy": 0.7628571428571429,
+      "mean_token_accuracy_utility": 0.6560134310134308,
       "step": 750
     },
     {
       "epoch": 0.27939679896382646,
       "grad_norm": 0.0,
       "learning_rate": 1.810684837242755e-05,
-      "loss": 2.3887,
-      "mean_token_accuracy": 0.7531379731379733,
-      "mean_token_accuracy_utility": 0.6400845289486262,
+      "loss": 2.3739,
+      "mean_token_accuracy": 0.7572039072039073,
+      "mean_token_accuracy_utility": 0.6386652450546972,
       "step": 755
     },
     {
       "epoch": 0.2812471088907392,
       "grad_norm": 0.0,
       "learning_rate": 1.8068848293555118e-05,
-      "loss": 2.2767,
-      "mean_token_accuracy": 0.7527289377289378,
-      "mean_token_accuracy_utility": 0.664188034188034,
+      "loss": 2.2608,
+      "mean_token_accuracy": 0.756929181929182,
+      "mean_token_accuracy_utility": 0.663211233211233,
       "step": 760
     },
     {
       "epoch": 0.28309741881765194,
       "grad_norm": 0.0,
       "learning_rate": 1.8030511331077945e-05,
-      "loss": 2.243,
-      "mean_token_accuracy": 0.7564163614163617,
-      "mean_token_accuracy_utility": 0.6687545787545787,
+      "loss": 2.2277,
+      "mean_token_accuracy": 0.7604639804639807,
+      "mean_token_accuracy_utility": 0.6671001221001219,
       "step": 765
     },
     {
       "epoch": 0.2849477287445647,
       "grad_norm": 0.0,
       "learning_rate": 1.799183908560787e-05,
-      "loss": 2.249,
-      "mean_token_accuracy": 0.74507326007326,
-      "mean_token_accuracy_utility": 0.6748412698412696,
+      "loss": 2.2332,
+      "mean_token_accuracy": 0.7492002442002441,
+      "mean_token_accuracy_utility": 0.6738705738705736,
       "step": 770
     },
     {
       "epoch": 0.28679803867147746,
       "grad_norm": 0.0,
       "learning_rate": 1.795283317175518e-05,
-      "loss": 2.254,
-      "mean_token_accuracy": 0.7642429792429793,
-      "mean_token_accuracy_utility": 0.6592185592185591,
+      "loss": 2.2399,
+      "mean_token_accuracy": 0.7680708180708183,
+      "mean_token_accuracy_utility": 0.6578998778998778,
       "step": 775
     },
     {
       "epoch": 0.2886483485983902,
       "grad_norm": 0.0,
       "learning_rate": 1.7913495218061202e-05,
-      "loss": 2.3365,
-      "mean_token_accuracy": 0.7552075702075701,
-      "mean_token_accuracy_utility": 0.6522954822954821,
+      "loss": 2.3213,
+      "mean_token_accuracy": 0.7592124542124542,
+      "mean_token_accuracy_utility": 0.651227106227106,
       "step": 780
     },
     {
       "epoch": 0.290498658525303,
       "grad_norm": 0.0,
       "learning_rate": 1.787382686693029e-05,
-      "loss": 2.3283,
-      "mean_token_accuracy": 0.7592735042735044,
-      "mean_token_accuracy_utility": 0.6512454212454211,
+      "loss": 2.3142,
+      "mean_token_accuracy": 0.7630830280830283,
+      "mean_token_accuracy_utility": 0.6499023199023197,
       "step": 785
     },
     {
       "epoch": 0.29234896845221575,
       "grad_norm": 0.0,
       "learning_rate": 1.783382977456128e-05,
-      "loss": 2.1887,
-      "mean_token_accuracy": 0.7624603174603175,
-      "mean_token_accuracy_utility": 0.6711477411477409,
+      "loss": 2.1726,
+      "mean_token_accuracy": 0.767124542124542,
+      "mean_token_accuracy_utility": 0.6697741147741145,
       "step": 790
     },
     {
       "epoch": 0.2941992783791285,
       "grad_norm": 0.0,
       "learning_rate": 1.779350561087833e-05,
-      "loss": 2.334,
-      "mean_token_accuracy": 0.7477594627594628,
-      "mean_token_accuracy_utility": 0.6596886446886445,
+      "loss": 2.3189,
+      "mean_token_accuracy": 0.7519230769230771,
+      "mean_token_accuracy_utility": 0.6584554334554333,
       "step": 795
     },
     {
       "epoch": 0.2960495883060413,
       "grad_norm": 0.0,
       "learning_rate": 1.775285605946119e-05,
-      "loss": 2.2044,
-      "mean_token_accuracy": 0.7661660561660563,
-      "mean_token_accuracy_utility": 0.6687423687423686,
+      "loss": 2.1899,
+      "mean_token_accuracy": 0.7701709401709402,
+      "mean_token_accuracy_utility": 0.6677594627594626,
       "step": 800
     },
     {
       "epoch": 0.2960495883060413,
-      "eval_mean_token_accuracy": 0.7440769470181238,
-      "eval_reasoning_loss": 0.878727376461029,
-      "eval_reasoning_runtime": 27.535,
-      "eval_reasoning_samples_per_second": 4.685,
-      "eval_reasoning_steps_per_second": 1.198,
+      "eval_mean_token_accuracy": 0.748052600993778,
+      "eval_reasoning_loss": 0.8567540645599365,
+      "eval_reasoning_runtime": 27.6145,
+      "eval_reasoning_samples_per_second": 4.671,
+      "eval_reasoning_steps_per_second": 1.195,
       "step": 800
     },
     {
       "epoch": 0.2960495883060413,
-      "eval_mean_token_accuracy": 0.6597089848162198,
-      "eval_utility_loss": 1.4420039653778076,
-      "eval_utility_runtime": 10.3623,
-      "eval_utility_samples_per_second": 4.729,
-      "eval_utility_steps_per_second": 1.255,
+      "eval_mean_token_accuracy": 0.6589697850206857,
+      "eval_utility_loss": 1.446785569190979,
+      "eval_utility_runtime": 10.3934,
+      "eval_utility_samples_per_second": 4.715,
+      "eval_utility_steps_per_second": 1.251,
       "step": 800
     },
     {
       "epoch": 0.29789989823295404,
       "grad_norm": 0.0,
       "learning_rate": 1.7711882817474922e-05,
-      "loss": 2.2575,
-      "mean_token_accuracy": 0.7619963369963371,
-      "mean_token_accuracy_utility": 0.6597863247863246,
+      "loss": 2.2427,
+      "mean_token_accuracy": 0.7661965811965814,
+      "mean_token_accuracy_utility": 0.6587423687423686,
       "step": 805
     },
     {
       "epoch": 0.2997502081598668,
       "grad_norm": 0.0,
       "learning_rate": 1.7670587595599034e-05,
-      "loss": 2.2518,
-      "mean_token_accuracy": 0.7612148962148962,
-      "mean_token_accuracy_utility": 0.6689926739926737,
+      "loss": 2.237,
+      "mean_token_accuracy": 0.7652136752136751,
+      "mean_token_accuracy_utility": 0.6678754578754577,
       "step": 810
     },
     {
       "epoch": 0.3016005180867795,
       "grad_norm": 0.0,
       "learning_rate": 1.762897211795607e-05,
-      "loss": 2.3275,
-      "mean_token_accuracy": 0.7471489621489619,
-      "mean_token_accuracy_utility": 0.6527228327228326,
+      "loss": 2.3125,
+      "mean_token_accuracy": 0.7510927960927958,
+      "mean_token_accuracy_utility": 0.6515384615384614,
       "step": 815
     },
     {
       "epoch": 0.3034508280136923,
       "grad_norm": 0.0,
       "learning_rate": 1.758703812203961e-05,
-      "loss": 2.3389,
-      "mean_token_accuracy": 0.7462087912087911,
-      "mean_token_accuracy_utility": 0.6563736263736263,
+      "loss": 2.3245,
+      "mean_token_accuracy": 0.7505921855921857,
+      "mean_token_accuracy_utility": 0.655091575091575,
       "step": 820
     },
     {
       "epoch": 0.30530113794060504,
       "grad_norm": 0.0,
       "learning_rate": 1.7544787358641735e-05,
-      "loss": 2.3797,
-      "mean_token_accuracy": 0.7358180708180708,
-      "mean_token_accuracy_utility": 0.6568376068376067,
+      "loss": 2.3636,
+      "mean_token_accuracy": 0.7400732600732601,
+      "mean_token_accuracy_utility": 0.6555433455433455,
       "step": 825
     },
     {
       "epoch": 0.3071514478675178,
       "grad_norm": 0.0,
       "learning_rate": 1.7502221591779932e-05,
-      "loss": 2.3064,
-      "mean_token_accuracy": 0.7368253968253969,
-      "mean_token_accuracy_utility": 0.6669184912790087,
+      "loss": 2.2907,
+      "mean_token_accuracy": 0.7407326007326008,
+      "mean_token_accuracy_utility": 0.6659709201087471,
       "step": 830
     },
     {
       "epoch": 0.30900175779443056,
       "grad_norm": 0.0,
       "learning_rate": 1.7459342598623438e-05,
-      "loss": 2.2693,
-      "mean_token_accuracy": 0.757032967032967,
-      "mean_token_accuracy_utility": 0.6663431013431012,
+      "loss": 2.2552,
+      "mean_token_accuracy": 0.7608547008547009,
+      "mean_token_accuracy_utility": 0.665067155067155,
       "step": 835
     },
     {
       "epoch": 0.31085206772134333,
       "grad_norm": 0.0,
       "learning_rate": 1.741615216941905e-05,
-      "loss": 2.2096,
-      "mean_token_accuracy": 0.7686080586080588,
-      "mean_token_accuracy_utility": 0.6668070818070817,
+      "loss": 2.1951,
+      "mean_token_accuracy": 0.7726678876678877,
+      "mean_token_accuracy_utility": 0.6657081807081806,
       "step": 840
     },
     {
       "epoch": 0.3127023776482561,
       "grad_norm": 0.0,
       "learning_rate": 1.7372652107416364e-05,
-      "loss": 2.331,
-      "mean_token_accuracy": 0.7513980463980465,
-      "mean_token_accuracy_utility": 0.6557081807081806,
+      "loss": 2.3163,
+      "mean_token_accuracy": 0.7556227106227107,
+      "mean_token_accuracy_utility": 0.6542857142857141,
       "step": 845
     },
     {
       "epoch": 0.31455268757516885,
       "grad_norm": 0.0,
       "learning_rate": 1.7328844228792513e-05,
-      "loss": 2.2306,
-      "mean_token_accuracy": 0.775750915750916,
-      "mean_token_accuracy_utility": 0.658205128205128,
+      "loss": 2.2178,
+      "mean_token_accuracy": 0.7791819291819294,
+      "mean_token_accuracy_utility": 0.6570268620268618,
       "step": 850
     },
     {
       "epoch": 0.3164029975020816,
       "grad_norm": 0.0,
       "learning_rate": 1.7284730362576308e-05,
-      "loss": 2.3024,
-      "mean_token_accuracy": 0.7541941391941392,
-      "mean_token_accuracy_utility": 0.6573931623931623,
+      "loss": 2.2881,
+      "mean_token_accuracy": 0.7584004884004885,
+      "mean_token_accuracy_utility": 0.6562759462759462,
       "step": 855
     },
     {
       "epoch": 0.3182533074289944,
       "grad_norm": 0.0,
       "learning_rate": 1.7240312350571905e-05,
-      "loss": 2.3921,
-      "mean_token_accuracy": 0.7605372405372405,
-      "mean_token_accuracy_utility": 0.6378510378510376,
+      "loss": 2.3779,
+      "mean_token_accuracy": 0.7647435897435897,
+      "mean_token_accuracy_utility": 0.6367338217338215,
       "step": 860
     },
     {
       "epoch": 0.3201036173559071,
       "grad_norm": 0.0,
       "learning_rate": 1.719559204728188e-05,
-      "loss": 2.2628,
-      "mean_token_accuracy": 0.7685470085470086,
-      "mean_token_accuracy_utility": 0.6533638583638582,
+      "loss": 2.2498,
+      "mean_token_accuracy": 0.7724908424908425,
+      "mean_token_accuracy_utility": 0.6519902319902319,
       "step": 865
     },
     {
       "epoch": 0.32195392728281985,
       "grad_norm": 0.0,
       "learning_rate": 1.715057131982983e-05,
-      "loss": 2.3505,
-      "mean_token_accuracy": 0.7585531135531134,
-      "mean_token_accuracy_utility": 0.6450488400488399,
+      "loss": 2.3367,
+      "mean_token_accuracy": 0.7627716727716728,
+      "mean_token_accuracy_utility": 0.6438278388278387,
       "step": 870
     },
     {
       "epoch": 0.3238042372097326,
       "grad_norm": 0.0,
       "learning_rate": 1.710525204788239e-05,
-      "loss": 2.2158,
-      "mean_token_accuracy": 0.7654639804639807,
-      "mean_token_accuracy_utility": 0.6651789747299784,
+      "loss": 2.2019,
+      "mean_token_accuracy": 0.7693711843711845,
+      "mean_token_accuracy_utility": 0.6638257038975974,
       "step": 875
     },
     {
       "epoch": 0.3256545471366454,
       "grad_norm": 0.0,
       "learning_rate": 1.7059636123570767e-05,
-      "loss": 2.3128,
-      "mean_token_accuracy": 0.7595848595848598,
-      "mean_token_accuracy_utility": 0.6506471306471304,
+      "loss": 2.2998,
+      "mean_token_accuracy": 0.7635714285714288,
+      "mean_token_accuracy_utility": 0.6495604395604394,
       "step": 880
     },
     {
       "epoch": 0.32750485706355814,
       "grad_norm": 0.0,
       "learning_rate": 1.7013725451411757e-05,
-      "loss": 2.2406,
-      "mean_token_accuracy": 0.7721367521367521,
-      "mean_token_accuracy_utility": 0.6600976800976799,
+      "loss": 2.2267,
+      "mean_token_accuracy": 0.7764102564102565,
+      "mean_token_accuracy_utility": 0.6588339438339436,
       "step": 885
     },
     {
       "epoch": 0.3293551669904709,
       "grad_norm": 0.0,
       "learning_rate": 1.696752194822819e-05,
-      "loss": 2.3062,
-      "mean_token_accuracy": 0.7693223443223444,
-      "mean_token_accuracy_utility": 0.6452503052503051,
+      "loss": 2.2917,
+      "mean_token_accuracy": 0.7735225885225886,
+      "mean_token_accuracy_utility": 0.6443101343101342,
       "step": 890
     },
     {
       "epoch": 0.33120547691738367,
       "grad_norm": 0.0,
       "learning_rate": 1.692102754306895e-05,
-      "loss": 2.2957,
-      "mean_token_accuracy": 0.7690659340659343,
-      "mean_token_accuracy_utility": 0.6449694749694748,
+      "loss": 2.2817,
+      "mean_token_accuracy": 0.7732295482295485,
+      "mean_token_accuracy_utility": 0.6436324786324785,
       "step": 895
     },
     {
       "epoch": 0.33305578684429643,
       "grad_norm": 0.0,
       "learning_rate": 1.6874244177128395e-05,
-      "loss": 2.1944,
-      "mean_token_accuracy": 0.7681257631257632,
-      "mean_token_accuracy_utility": 0.667228327228327,
+      "loss": 2.1806,
+      "mean_token_accuracy": 0.7719047619047619,
+      "mean_token_accuracy_utility": 0.6656471306471304,
       "step": 900
     },
     {
       "epoch": 0.33305578684429643,
-      "eval_mean_token_accuracy": 0.7449353478765242,
-      "eval_reasoning_loss": 0.8736643195152283,
-      "eval_reasoning_runtime": 27.5475,
-      "eval_reasoning_samples_per_second": 4.683,
-      "eval_reasoning_steps_per_second": 1.198,
+      "eval_mean_token_accuracy": 0.7491019873372817,
+      "eval_reasoning_loss": 0.8521426916122437,
+      "eval_reasoning_runtime": 27.5729,
+      "eval_reasoning_samples_per_second": 4.679,
+      "eval_reasoning_steps_per_second": 1.197,
       "step": 900
     },
     {
       "epoch": 0.33305578684429643,
-      "eval_mean_token_accuracy": 0.6603316709979563,
-      "eval_utility_loss": 1.4390980005264282,
-      "eval_utility_runtime": 10.3648,
-      "eval_utility_samples_per_second": 4.728,
-      "eval_utility_steps_per_second": 1.254,
+      "eval_mean_token_accuracy": 0.6594830629288736,
+      "eval_utility_loss": 1.4440678358078003,
+      "eval_utility_runtime": 10.3828,
+      "eval_utility_samples_per_second": 4.719,
+      "eval_utility_steps_per_second": 1.252,
       "step": 900
     },
     {
       "epoch": 0.3349060967712092,
       "grad_norm": 0.0,
       "learning_rate": 1.6827173803665328e-05,
-      "loss": 2.3407,
-      "mean_token_accuracy": 0.7432905982905982,
-      "mean_token_accuracy_utility": 0.6655982905982903,
+      "loss": 2.3267,
+      "mean_token_accuracy": 0.7473992673992672,
+      "mean_token_accuracy_utility": 0.664224664224664,
       "step": 905
     },
     {
       "epoch": 0.33675640669812196,
       "grad_norm": 0.0,
       "learning_rate": 1.677981838792144e-05,
-      "loss": 2.2793,
-      "mean_token_accuracy": 0.7535164835164835,
-      "mean_token_accuracy_utility": 0.6605982905982903,
+      "loss": 2.2657,
+      "mean_token_accuracy": 0.7575030525030527,
+      "mean_token_accuracy_utility": 0.6593894993894992,
       "step": 910
     },
     {
       "epoch": 0.3386067166250347,
       "grad_norm": 0.0,
       "learning_rate": 1.6732179907039266e-05,
-      "loss": 2.1866,
-      "mean_token_accuracy": 0.7586996336996339,
-      "mean_token_accuracy_utility": 0.6715995115995115,
+      "loss": 2.1714,
+      "mean_token_accuracy": 0.7630219780219782,
+      "mean_token_accuracy_utility": 0.6703540903540902,
       "step": 915
     },
     {
       "epoch": 0.34045702655194743,
       "grad_norm": 0.0,
       "learning_rate": 1.6684260349979637e-05,
-      "loss": 2.2454,
-      "mean_token_accuracy": 0.7550793650793651,
-      "mean_token_accuracy_utility": 0.6655433455433453,
+      "loss": 2.2307,
+      "mean_token_accuracy": 0.7593894993894994,
+      "mean_token_accuracy_utility": 0.6642673992673991,
       "step": 920
     },
     {
       "epoch": 0.3423073364788602,
       "grad_norm": 0.0,
       "learning_rate": 1.6636061717438626e-05,
-      "loss": 2.1484,
-      "mean_token_accuracy": 0.7703968253968254,
-      "mean_token_accuracy_utility": 0.6739194139194138,
+      "loss": 2.1347,
+      "mean_token_accuracy": 0.773840048840049,
+      "mean_token_accuracy_utility": 0.6723687423687421,
       "step": 925
     },
     {
       "epoch": 0.34415764640577295,
       "grad_norm": 0.0,
       "learning_rate": 1.6587586021764022e-05,
-      "loss": 2.1898,
-      "mean_token_accuracy": 0.7653357753357753,
-      "mean_token_accuracy_utility": 0.6685138831493396,
+      "loss": 2.1762,
+      "mean_token_accuracy": 0.7693345543345546,
+      "mean_token_accuracy_utility": 0.6671931683904053,
       "step": 930
     },
     {
       "epoch": 0.3460079563326857,
       "grad_norm": 0.0,
       "learning_rate": 1.653883528687133e-05,
-      "loss": 2.2979,
-      "mean_token_accuracy": 0.7565198586613124,
-      "mean_token_accuracy_utility": 0.651611214762576,
+      "loss": 2.283,
+      "mean_token_accuracy": 0.7609418741049391,
+      "mean_token_accuracy_utility": 0.650332007376855,
       "step": 935
     },
     {
       "epoch": 0.3478582662595985,
       "grad_norm": 0.0,
       "learning_rate": 1.6489811548159245e-05,
-      "loss": 2.2743,
-      "mean_token_accuracy": 0.7516727716727717,
-      "mean_token_accuracy_utility": 0.6588766788766788,
+      "loss": 2.2599,
+      "mean_token_accuracy": 0.7564224664224666,
+      "mean_token_accuracy_utility": 0.6576495726495726,
       "step": 940
     },
     {
       "epoch": 0.34970857618651124,
       "grad_norm": 0.0,
       "learning_rate": 1.6440516852424678e-05,
-      "loss": 2.292,
-      "mean_token_accuracy": 0.7521550671550672,
-      "mean_token_accuracy_utility": 0.6611111111111109,
+      "loss": 2.2782,
+      "mean_token_accuracy": 0.7561721611721614,
+      "mean_token_accuracy_utility": 0.6598229548229547,
       "step": 945
     },
     {
       "epoch": 0.351558886113424,
       "grad_norm": 0.0,
       "learning_rate": 1.6390953257777324e-05,
-      "loss": 2.1965,
-      "mean_token_accuracy": 0.7647863247863248,
-      "mean_token_accuracy_utility": 0.6699694749694747,
+      "loss": 2.1832,
+      "mean_token_accuracy": 0.7690964590964592,
+      "mean_token_accuracy_utility": 0.6684310134310132,
       "step": 950
     },
     {
       "epoch": 0.35340919604033677,
       "grad_norm": 0.0,
       "learning_rate": 1.634112283355369e-05,
-      "loss": 2.3527,
-      "mean_token_accuracy": 0.7516666666666666,
-      "mean_token_accuracy_utility": 0.647203907203907,
+      "loss": 2.3395,
+      "mean_token_accuracy": 0.7561111111111112,
+      "mean_token_accuracy_utility": 0.6459706959706957,
       "step": 955
     },
     {
       "epoch": 0.35525950596724953,
       "grad_norm": 0.0,
       "learning_rate": 1.6291027660230735e-05,
-      "loss": 2.2287,
-      "mean_token_accuracy": 0.7605860805860806,
-      "mean_token_accuracy_utility": 0.6653639303845065,
+      "loss": 2.2144,
+      "mean_token_accuracy": 0.7642918192918196,
+      "mean_token_accuracy_utility": 0.6640987202098313,
       "step": 960
     },
     {
       "epoch": 0.3571098158941623,
       "grad_norm": 0.0,
       "learning_rate": 1.6240669829338992e-05,
-      "loss": 2.3426,
-      "mean_token_accuracy": 0.7447130647130645,
-      "mean_token_accuracy_utility": 0.6506593406593404,
+      "loss": 2.3279,
+      "mean_token_accuracy": 0.7481562881562881,
+      "mean_token_accuracy_utility": 0.6489316239316236,
       "step": 965
     },
     {
       "epoch": 0.358960125821075,
       "grad_norm": 0.0,
       "learning_rate": 1.6190051443375248e-05,
-      "loss": 2.2134,
-      "mean_token_accuracy": 0.7608241758241758,
-      "mean_token_accuracy_utility": 0.6697069597069595,
+      "loss": 2.1988,
+      "mean_token_accuracy": 0.7638705738705739,
+      "mean_token_accuracy_utility": 0.6683455433455431,
       "step": 970
     },
     {
       "epoch": 0.36081043574798777,
       "grad_norm": 0.0,
       "learning_rate": 1.6139174615714753e-05,
-      "loss": 2.2725,
-      "mean_token_accuracy": 0.7598351648351648,
-      "mean_token_accuracy_utility": 0.6599064391970965,
+      "loss": 2.2593,
+      "mean_token_accuracy": 0.7632478632478633,
+      "mean_token_accuracy_utility": 0.6591758664250011,
       "step": 975
     },
     {
       "epoch": 0.36266074567490053,
       "grad_norm": 0.0,
       "learning_rate": 1.6088041470523005e-05,
-      "loss": 2.3252,
-      "mean_token_accuracy": 0.7460256410256412,
-      "mean_token_accuracy_utility": 0.6597985347985345,
+      "loss": 2.3097,
+      "mean_token_accuracy": 0.7501526251526253,
+      "mean_token_accuracy_utility": 0.6588278388278386,
       "step": 980
     },
     {
       "epoch": 0.3645110556018133,
       "grad_norm": 0.0,
       "learning_rate": 1.6036654142667043e-05,
-      "loss": 2.3344,
-      "mean_token_accuracy": 0.7491452991452989,
-      "mean_token_accuracy_utility": 0.6568311849626727,
+      "loss": 2.3195,
+      "mean_token_accuracy": 0.7531684981684983,
+      "mean_token_accuracy_utility": 0.6552784643269071,
       "step": 985
     },
     {
       "epoch": 0.36636136552872606,
       "grad_norm": 0.0,
       "learning_rate": 1.598501477762632e-05,
-      "loss": 2.312,
-      "mean_token_accuracy": 0.7534798534798535,
-      "mean_token_accuracy_utility": 0.6566544566544564,
+      "loss": 2.2978,
+      "mean_token_accuracy": 0.7575885225885226,
+      "mean_token_accuracy_utility": 0.6552442002442,
       "step": 990
     },
     {
       "epoch": 0.3682116754556388,
       "grad_norm": 0.0,
       "learning_rate": 1.5933125531403135e-05,
-      "loss": 2.3536,
-      "mean_token_accuracy": 0.7547399074812324,
-      "mean_token_accuracy_utility": 0.6433821733821732,
+      "loss": 2.3406,
+      "mean_token_accuracy": 0.7588048593537552,
+      "mean_token_accuracy_utility": 0.6422283272283271,
       "step": 995
     },
     {
       "epoch": 0.3700619853825516,
       "grad_norm": 0.0,
       "learning_rate": 1.5880988570432603e-05,
-      "loss": 2.3031,
-      "mean_token_accuracy": 0.7702869352869353,
-      "mean_token_accuracy_utility": 0.6453357753357751,
+      "loss": 2.2903,
+      "mean_token_accuracy": 0.773901098901099,
+      "mean_token_accuracy_utility": 0.6437362637362636,
       "step": 1000
     },
     {
       "epoch": 0.3700619853825516,
-      "eval_mean_token_accuracy": 0.7457067986479751,
-      "eval_reasoning_loss": 0.8698025345802307,
-      "eval_reasoning_runtime": 27.5431,
-      "eval_reasoning_samples_per_second": 4.684,
-      "eval_reasoning_steps_per_second": 1.198,
+      "eval_mean_token_accuracy": 0.7497011702894055,
+      "eval_reasoning_loss": 0.8484104871749878,
+      "eval_reasoning_runtime": 27.6112,
+      "eval_reasoning_samples_per_second": 4.672,
+      "eval_reasoning_steps_per_second": 1.195,
       "step": 1000
     },
     {
       "epoch": 0.3700619853825516,
-      "eval_mean_token_accuracy": 0.6605476942909027,
-      "eval_utility_loss": 1.437069058418274,
-      "eval_utility_runtime": 10.3599,
-      "eval_utility_samples_per_second": 4.73,
-      "eval_utility_steps_per_second": 1.255,
+      "eval_mean_token_accuracy": 0.6593459839813858,
+      "eval_utility_loss": 1.4419615268707275,
+      "eval_utility_runtime": 10.394,
+      "eval_utility_samples_per_second": 4.714,
+      "eval_utility_steps_per_second": 1.251,
       "step": 1000
     },
     {
       "epoch": 0.37191229530946435,
       "grad_norm": 0.0,
       "learning_rate": 1.582860607149222e-05,
-      "loss": 2.2569,
-      "mean_token_accuracy": 0.7570879120879122,
-      "mean_token_accuracy_utility": 0.661196581196581,
+      "loss": 2.2432,
+      "mean_token_accuracy": 0.7608730158730159,
+      "mean_token_accuracy_utility": 0.6599694749694748,
       "step": 1005
     },
     {
       "epoch": 0.3737626052363771,
       "grad_norm": 0.0,
       "learning_rate": 1.5775980221610966e-05,
-      "loss": 2.3095,
-      "mean_token_accuracy": 0.754072039072039,
-      "mean_token_accuracy_utility": 0.6572463651908391,
+      "loss": 2.2957,
+      "mean_token_accuracy": 0.7577228327228328,
+      "mean_token_accuracy_utility": 0.6561070041662672,
       "step": 1010
     },
     {
       "epoch": 0.3756129151632899,
       "grad_norm": 0.0,
       "learning_rate": 1.5723113217978e-05,
-      "loss": 2.2202,
-      "mean_token_accuracy": 0.7577838827838829,
-      "mean_token_accuracy_utility": 0.6748962148962148,
+      "loss": 2.2067,
+      "mean_token_accuracy": 0.7619413919413921,
+      "mean_token_accuracy_utility": 0.6736324786324784,
       "step": 1015
     },
     {
       "epoch": 0.37746322509020264,
       "grad_norm": 0.0,
       "learning_rate": 1.567000726785093e-05,
-      "loss": 2.1303,
-      "mean_token_accuracy": 0.7683699633699634,
-      "mean_token_accuracy_utility": 0.6803479853479851,
+      "loss": 2.116,
+      "mean_token_accuracy": 0.7722710622710626,
+      "mean_token_accuracy_utility": 0.6790659340659337,
       "step": 1020
     },
     {
       "epoch": 0.37931353501711534,
       "grad_norm": 0.0,
       "learning_rate": 1.561666458846365e-05,
-      "loss": 2.3237,
-      "mean_token_accuracy": 0.7428754578754578,
-      "mean_token_accuracy_utility": 0.6585836385836383,
+      "loss": 2.3093,
+      "mean_token_accuracy": 0.7473015873015874,
+      "mean_token_accuracy_utility": 0.6573076923076921,
       "step": 1025
     },
     {
       "epoch": 0.3811638449440281,
       "grad_norm": 0.0,
       "learning_rate": 1.5563087406933762e-05,
-      "loss": 2.2602,
-      "mean_token_accuracy": 0.7501587301587301,
-      "mean_token_accuracy_utility": 0.6636630036630035,
+      "loss": 2.2454,
+      "mean_token_accuracy": 0.7541208791208792,
+      "mean_token_accuracy_utility": 0.6623321123321121,
       "step": 1030
     },
     {
       "epoch": 0.38301415487094087,
       "grad_norm": 0.0,
       "learning_rate": 1.550927796016961e-05,
-      "loss": 2.3249,
-      "mean_token_accuracy": 0.761025641025641,
-      "mean_token_accuracy_utility": 0.6457142857142856,
+      "loss": 2.3109,
+      "mean_token_accuracy": 0.7648351648351648,
+      "mean_token_accuracy_utility": 0.6442124542124541,
       "step": 1035
     },
     {
       "epoch": 0.38486446479785363,
       "grad_norm": 0.0,
       "learning_rate": 1.5455238494776876e-05,
-      "loss": 2.2988,
-      "mean_token_accuracy": 0.7487301587301588,
-      "mean_token_accuracy_utility": 0.6596398046398044,
+      "loss": 2.284,
+      "mean_token_accuracy": 0.7529365079365081,
+      "mean_token_accuracy_utility": 0.6585592185592184,
       "step": 1040
     },
     {
       "epoch": 0.3867147747247664,
       "grad_norm": 0.0,
       "learning_rate": 1.5400971266964772e-05,
-      "loss": 2.2222,
-      "mean_token_accuracy": 0.7680341880341882,
-      "mean_token_accuracy_utility": 0.66012999195926,
+      "loss": 2.2081,
+      "mean_token_accuracy": 0.7720451770451773,
+      "mean_token_accuracy_utility": 0.6589882962565888,
       "step": 1045
     },
     {
       "epoch": 0.38856508465167916,
       "grad_norm": 0.0,
       "learning_rate": 1.5346478542451862e-05,
-      "loss": 2.204,
-      "mean_token_accuracy": 0.752777777777778,
-      "mean_token_accuracy_utility": 0.6768547996474152,
+      "loss": 2.1898,
+      "mean_token_accuracy": 0.7571001221001223,
+      "mean_token_accuracy_utility": 0.6755366722393429,
       "step": 1050
     },
     {
       "epoch": 0.3904153945785919,
       "grad_norm": 0.0,
       "learning_rate": 1.529176259637145e-05,
-      "loss": 2.3652,
-      "mean_token_accuracy": 0.7543467643467645,
-      "mean_token_accuracy_utility": 0.6478876678876677,
+      "loss": 2.3516,
+      "mean_token_accuracy": 0.7586630036630038,
+      "mean_token_accuracy_utility": 0.6462820512820511,
       "step": 1055
     },
     {
       "epoch": 0.3922657045055047,
       "grad_norm": 0.0,
       "learning_rate": 1.5236825713176584e-05,
-      "loss": 2.3101,
-      "mean_token_accuracy": 0.7511721611721611,
-      "mean_token_accuracy_utility": 0.6556410256410253,
+      "loss": 2.2975,
+      "mean_token_accuracy": 0.7548717948717949,
+      "mean_token_accuracy_utility": 0.6540231990231987,
       "step": 1060
     },
     {
       "epoch": 0.39411601443241745,
       "grad_norm": 0.0,
       "learning_rate": 1.5181670186544706e-05,
-      "loss": 2.3488,
-      "mean_token_accuracy": 0.7498046398046398,
-      "mean_token_accuracy_utility": 0.6529174051754695,
+      "loss": 2.3345,
+      "mean_token_accuracy": 0.7542735042735043,
+      "mean_token_accuracy_utility": 0.6514936367326472,
       "step": 1065
     },
     {
       "epoch": 0.3959663243593302,
       "grad_norm": 0.0,
       "learning_rate": 1.5126298319281859e-05,
-      "loss": 2.2634,
-      "mean_token_accuracy": 0.7725518925518928,
-      "mean_token_accuracy_utility": 0.6553968253968252,
+      "loss": 2.2504,
+      "mean_token_accuracy": 0.7765445665445668,
+      "mean_token_accuracy_utility": 0.6540781440781439,
       "step": 1070
     },
     {
       "epoch": 0.3978166342862429,
       "grad_norm": 0.0,
       "learning_rate": 1.5070712423226552e-05,
-      "loss": 2.2045,
-      "mean_token_accuracy": 0.7653052503052502,
-      "mean_token_accuracy_utility": 0.6696703296703295,
+      "loss": 2.1906,
+      "mean_token_accuracy": 0.768931623931624,
+      "mean_token_accuracy_utility": 0.6685286935286934,
       "step": 1075
     },
     {
       "epoch": 0.3996669442131557,
       "grad_norm": 0.0,
       "learning_rate": 1.5014914819153252e-05,
-      "loss": 2.2099,
-      "mean_token_accuracy": 0.7474053724053723,
-      "mean_token_accuracy_utility": 0.6744566544566543,
+      "loss": 2.1955,
+      "mean_token_accuracy": 0.7513492063492062,
+      "mean_token_accuracy_utility": 0.6727655677655676,
       "step": 1080
     },
     {
       "epoch": 0.40151725414006845,
       "grad_norm": 0.0,
       "learning_rate": 1.4958907836675467e-05,
-      "loss": 2.2752,
-      "mean_token_accuracy": 0.7536385836385838,
-      "mean_token_accuracy_utility": 0.6615445665445664,
+      "loss": 2.2609,
+      "mean_token_accuracy": 0.7581257631257632,
+      "mean_token_accuracy_utility": 0.6602258852258851,
       "step": 1085
     },
     {
       "epoch": 0.4033675640669812,
       "grad_norm": 0.0,
       "learning_rate": 1.490269381414849e-05,
-      "loss": 2.2609,
-      "mean_token_accuracy": 0.7574053724053724,
-      "mean_token_accuracy_utility": 0.6588949938949937,
+      "loss": 2.2474,
+      "mean_token_accuracy": 0.7617338217338218,
+      "mean_token_accuracy_utility": 0.6578632478632478,
       "step": 1090
     },
     {
       "epoch": 0.405217873993894,
       "grad_norm": 0.0,
       "learning_rate": 1.484627509857178e-05,
-      "loss": 2.1943,
-      "mean_token_accuracy": 0.7593345543345544,
-      "mean_token_accuracy_utility": 0.6727899877899876,
+      "loss": 2.1807,
+      "mean_token_accuracy": 0.7637423687423688,
+      "mean_token_accuracy_utility": 0.6718192918192916,
       "step": 1095
     },
     {
       "epoch": 0.40706818392080674,
       "grad_norm": 0.0,
       "learning_rate": 1.4789654045490957e-05,
-      "loss": 2.2449,
-      "mean_token_accuracy": 0.7657448107448107,
-      "mean_token_accuracy_utility": 0.6560134310134308,
+      "loss": 2.2322,
+      "mean_token_accuracy": 0.7696520146520147,
+      "mean_token_accuracy_utility": 0.6546581196581195,
       "step": 1100
     },
     {
       "epoch": 0.40706818392080674,
-      "eval_mean_token_accuracy": 0.7460656990068752,
-      "eval_reasoning_loss": 0.8666639924049377,
-      "eval_reasoning_runtime": 27.5089,
-      "eval_reasoning_samples_per_second": 4.689,
-      "eval_reasoning_steps_per_second": 1.2,
+      "eval_mean_token_accuracy": 0.7500785706668062,
+      "eval_reasoning_loss": 0.8455780744552612,
+      "eval_reasoning_runtime": 27.5976,
+      "eval_reasoning_samples_per_second": 4.674,
+      "eval_reasoning_steps_per_second": 1.196,
       "step": 1100
     },
     {
       "epoch": 0.40706818392080674,
-      "eval_mean_token_accuracy": 0.660887214333025,
-      "eval_utility_loss": 1.435528039932251,
-      "eval_utility_runtime": 10.3604,
-      "eval_utility_samples_per_second": 4.73,
-      "eval_utility_steps_per_second": 1.255,
+      "eval_mean_token_accuracy": 0.6601471241877305,
+      "eval_utility_loss": 1.4402730464935303,
+      "eval_utility_runtime": 10.3859,
+      "eval_utility_samples_per_second": 4.718,
+      "eval_utility_steps_per_second": 1.252,
       "step": 1100
     },
     {
       "epoch": 0.4089184938477195,
       "grad_norm": 0.0,
       "learning_rate": 1.4732833018899468e-05,
-      "loss": 2.299,
-      "mean_token_accuracy": 0.7517765567765567,
-      "mean_token_accuracy_utility": 0.662271062271062,
+      "loss": 2.2851,
+      "mean_token_accuracy": 0.7558058608058607,
+      "mean_token_accuracy_utility": 0.661263736263736,
       "step": 1105
     },
     {
       "epoch": 0.41076880377463226,
       "grad_norm": 0.0,
       "learning_rate": 1.4675814391139875e-05,
-      "loss": 2.2554,
-      "mean_token_accuracy": 0.7546703296703298,
-      "mean_token_accuracy_utility": 0.6615445665445664,
+      "loss": 2.2417,
+      "mean_token_accuracy": 0.7588400488400489,
+      "mean_token_accuracy_utility": 0.6601648351648349,
       "step": 1110
     },
     {
       "epoch": 0.412619113701545,
       "grad_norm": 0.0,
       "learning_rate": 1.4618600542804819e-05,
-      "loss": 2.3319,
-      "mean_token_accuracy": 0.7371184371184369,
-      "mean_token_accuracy_utility": 0.6562026862026861,
+      "loss": 2.3182,
+      "mean_token_accuracy": 0.7406593406593407,
+      "mean_token_accuracy_utility": 0.6548473748473747,
       "step": 1115
     },
     {
       "epoch": 0.4144694236284578,
       "grad_norm": 0.0,
       "learning_rate": 1.4561193862637621e-05,
-      "loss": 2.2009,
-      "mean_token_accuracy": 0.7594505494505495,
-      "mean_token_accuracy_utility": 0.6778510378510376,
+      "loss": 2.1874,
+      "mean_token_accuracy": 0.7632905982905985,
+      "mean_token_accuracy_utility": 0.6767643467643466,
       "step": 1120
     },
     {
       "epoch": 0.4163197335553705,
       "grad_norm": 0.0,
       "learning_rate": 1.4503596747432554e-05,
-      "loss": 2.3111,
-      "mean_token_accuracy": 0.7582844932844932,
-      "mean_token_accuracy_utility": 0.6540890232412722,
+      "loss": 2.2979,
+      "mean_token_accuracy": 0.7618803418803418,
+      "mean_token_accuracy_utility": 0.6527167488413161,
       "step": 1125
     },
     {
       "epoch": 0.41817004348228326,
       "grad_norm": 0.0,
       "learning_rate": 1.4445811601934763e-05,
-      "loss": 2.1818,
-      "mean_token_accuracy": 0.7605677655677657,
-      "mean_token_accuracy_utility": 0.6700488400488399,
+      "loss": 2.1677,
+      "mean_token_accuracy": 0.7645299145299147,
+      "mean_token_accuracy_utility": 0.6685347985347984,
       "step": 1130
     },
     {
       "epoch": 0.420020353409196,
       "grad_norm": 0.0,
       "learning_rate": 1.4387840838739875e-05,
-      "loss": 2.3126,
-      "mean_token_accuracy": 0.750958485958486,
-      "mean_token_accuracy_utility": 0.6625702075702073,
+      "loss": 2.2994,
+      "mean_token_accuracy": 0.75493894993895,
+      "mean_token_accuracy_utility": 0.6614041514041512,
       "step": 1135
     },
     {
       "epoch": 0.4218706633361088,
       "grad_norm": 0.0,
       "learning_rate": 1.4329686878193271e-05,
-      "loss": 2.2484,
-      "mean_token_accuracy": 0.752026862026862,
-      "mean_token_accuracy_utility": 0.6647409251186612,
+      "loss": 2.2363,
+      "mean_token_accuracy": 0.7559035409035408,
+      "mean_token_accuracy_utility": 0.6635615213818629,
       "step": 1140
     },
     {
       "epoch": 0.42372097326302155,
       "grad_norm": 0.0,
       "learning_rate": 1.4271352148289025e-05,
-      "loss": 2.2808,
-      "mean_token_accuracy": 0.7707936507936508,
-      "mean_token_accuracy_utility": 0.6515262515262515,
+      "loss": 2.2688,
+      "mean_token_accuracy": 0.7742735042735044,
+      "mean_token_accuracy_utility": 0.6502258852258852,
       "step": 1145
     },
     {
       "epoch": 0.4255712831899343,
       "grad_norm": 0.0,
       "learning_rate": 1.421283908456854e-05,
-      "loss": 2.3178,
-      "mean_token_accuracy": 0.7727411477411479,
-      "mean_token_accuracy_utility": 0.6428205128205126,
+      "loss": 2.3051,
+      "mean_token_accuracy": 0.7766727716727718,
+      "mean_token_accuracy_utility": 0.6413125763125761,
       "step": 1150
     },
     {
       "epoch": 0.4274215931168471,
       "grad_norm": 0.0,
       "learning_rate": 1.4154150130018867e-05,
-      "loss": 2.2253,
-      "mean_token_accuracy": 0.7684004884004885,
-      "mean_token_accuracy_utility": 0.6609035409035406,
+      "loss": 2.212,
+      "mean_token_accuracy": 0.7725091575091576,
+      "mean_token_accuracy_utility": 0.6593284493284491,
       "step": 1155
     },
     {
       "epoch": 0.42927190304375984,
       "grad_norm": 0.0,
       "learning_rate": 1.4095287734970678e-05,
-      "loss": 2.2486,
-      "mean_token_accuracy": 0.7614285714285713,
-      "mean_token_accuracy_utility": 0.6602380952380951,
+      "loss": 2.2347,
+      "mean_token_accuracy": 0.7656043956043957,
+      "mean_token_accuracy_utility": 0.6590720390720389,
       "step": 1160
     },
     {
       "epoch": 0.4311222129706726,
       "grad_norm": 0.0,
       "learning_rate": 1.4036254356996004e-05,
-      "loss": 2.3282,
-      "mean_token_accuracy": 0.7524969474969474,
-      "mean_token_accuracy_utility": 0.6544871794871793,
+      "loss": 2.3147,
+      "mean_token_accuracy": 0.7561477411477412,
+      "mean_token_accuracy_utility": 0.6533394383394382,
       "step": 1165
     },
     {
       "epoch": 0.43297252289758537,
       "grad_norm": 0.0,
       "learning_rate": 1.3977052460805597e-05,
-      "loss": 2.3147,
-      "mean_token_accuracy": 0.7625885225885228,
-      "mean_token_accuracy_utility": 0.6470451770451768,
+      "loss": 2.3006,
+      "mean_token_accuracy": 0.7670512820512824,
+      "mean_token_accuracy_utility": 0.6453418803418802,
       "step": 1170
     },
     {
       "epoch": 0.43482283282449813,
       "grad_norm": 0.0,
       "learning_rate": 1.3917684518146044e-05,
-      "loss": 2.2098,
-      "mean_token_accuracy": 0.7597741147741148,
-      "mean_token_accuracy_utility": 0.6669230769230767,
+      "loss": 2.1963,
+      "mean_token_accuracy": 0.7643223443223445,
+      "mean_token_accuracy_utility": 0.6658669108669106,
       "step": 1175
     },
     {
       "epoch": 0.43667314275141084,
       "grad_norm": 0.0,
       "learning_rate": 1.3858153007696552e-05,
-      "loss": 2.2552,
-      "mean_token_accuracy": 0.7585836385836385,
-      "mean_token_accuracy_utility": 0.6601159951159948,
+      "loss": 2.2425,
+      "mean_token_accuracy": 0.7622100122100121,
+      "mean_token_accuracy_utility": 0.6588156288156285,
       "step": 1180
     },
     {
       "epoch": 0.4385234526783236,
       "grad_norm": 0.0,
       "learning_rate": 1.3798460414965475e-05,
-      "loss": 2.1812,
-      "mean_token_accuracy": 0.771929181929182,
-      "mean_token_accuracy_utility": 0.6662515262515261,
+      "loss": 2.169,
+      "mean_token_accuracy": 0.7754884004884005,
+      "mean_token_accuracy_utility": 0.6644871794871793,
       "step": 1185
     },
     {
       "epoch": 0.44037376260523636,
       "grad_norm": 0.0,
       "learning_rate": 1.3738609232186537e-05,
-      "loss": 2.301,
-      "mean_token_accuracy": 0.7448717948717949,
-      "mean_token_accuracy_utility": 0.6585592185592184,
+      "loss": 2.2869,
+      "mean_token_accuracy": 0.7486691086691086,
+      "mean_token_accuracy_utility": 0.657185592185592,
       "step": 1190
     },
     {
       "epoch": 0.4422240725321491,
       "grad_norm": 0.0,
       "learning_rate": 1.3678601958214779e-05,
-      "loss": 2.1665,
-      "mean_token_accuracy": 0.7693894993894996,
-      "mean_token_accuracy_utility": 0.6684371184371182,
+      "loss": 2.1545,
+      "mean_token_accuracy": 0.7726129426129429,
+      "mean_token_accuracy_utility": 0.667197802197802,
       "step": 1195
     },
     {
       "epoch": 0.4440743824590619,
       "grad_norm": 0.0,
       "learning_rate": 1.3618441098422215e-05,
-      "loss": 2.2438,
-      "mean_token_accuracy": 0.7696520146520147,
-      "mean_token_accuracy_utility": 0.6536935286935285,
+      "loss": 2.2292,
+      "mean_token_accuracy": 0.7743101343101344,
+      "mean_token_accuracy_utility": 0.6524358974358972,
       "step": 1200
     },
     {
       "epoch": 0.4440743824590619,
-      "eval_mean_token_accuracy": 0.7463670316611494,
-      "eval_reasoning_loss": 0.8645576238632202,
-      "eval_reasoning_runtime": 27.5143,
-      "eval_reasoning_samples_per_second": 4.688,
-      "eval_reasoning_steps_per_second": 1.199,
+      "eval_mean_token_accuracy": 0.750652288887583,
+      "eval_reasoning_loss": 0.8436009883880615,
+      "eval_reasoning_runtime": 27.6065,
+      "eval_reasoning_samples_per_second": 4.673,
+      "eval_reasoning_steps_per_second": 1.195,
       "step": 1200
     },
     {
       "epoch": 0.4440743824590619,
-      "eval_mean_token_accuracy": 0.6609557538067689,
-      "eval_utility_loss": 1.4345169067382812,
-      "eval_utility_runtime": 10.3559,
-      "eval_utility_samples_per_second": 4.732,
-      "eval_utility_steps_per_second": 1.255,
+      "eval_mean_token_accuracy": 0.6602935954523029,
+      "eval_utility_loss": 1.43940007686615,
+      "eval_utility_runtime": 10.3885,
+      "eval_utility_samples_per_second": 4.717,
+      "eval_utility_steps_per_second": 1.251,
       "step": 1200
     },
     {
       "epoch": 0.44592469238597465,
       "grad_norm": 0.0,
       "learning_rate": 1.3558129164593256e-05,
-      "loss": 2.3262,
-      "mean_token_accuracy": 0.7565811965811966,
-      "mean_token_accuracy_utility": 0.6491350403901843,
+      "loss": 2.3138,
+      "mean_token_accuracy": 0.7604822954822955,
+      "mean_token_accuracy_utility": 0.6476805750262539,
       "step": 1205
     },
     {
       "epoch": 0.4477750023128874,
       "grad_norm": 0.0,
       "learning_rate": 1.349766867481982e-05,
-      "loss": 2.1816,
-      "mean_token_accuracy": 0.7515262515262516,
-      "mean_token_accuracy_utility": 0.6793467643467641,
+      "loss": 2.1673,
+      "mean_token_accuracy": 0.7556410256410256,
+      "mean_token_accuracy_utility": 0.6778754578754578,
       "step": 1210
     },
     {
       "epoch": 0.4496253122398002,
       "grad_norm": 0.0,
       "learning_rate": 1.3437062153396201e-05,
-      "loss": 2.2082,
-      "mean_token_accuracy": 0.7470634920634922,
-      "mean_token_accuracy_utility": 0.6752686202686201,
+      "loss": 2.1951,
+      "mean_token_accuracy": 0.7513614163614164,
+      "mean_token_accuracy_utility": 0.6736385836385834,
       "step": 1215
     },
     {
       "epoch": 0.45147562216671294,
       "grad_norm": 0.0,
       "learning_rate": 1.337631213071369e-05,
-      "loss": 2.4796,
-      "mean_token_accuracy": 0.7466666666666668,
-      "mean_token_accuracy_utility": 0.632106227106227,
+      "loss": 2.4646,
+      "mean_token_accuracy": 0.7508913308913311,
+      "mean_token_accuracy_utility": 0.6310134310134309,
       "step": 1220
     },
     {
       "epoch": 0.4533259320936257,
       "grad_norm": 0.0,
       "learning_rate": 1.331542114315491e-05,
-      "loss": 2.2723,
-      "mean_token_accuracy": 0.7635286935286933,
-      "mean_token_accuracy_utility": 0.655295633959,
+      "loss": 2.2593,
+      "mean_token_accuracy": 0.7672344322344322,
+      "mean_token_accuracy_utility": 0.6540257926891588,
       "step": 1225
     },
     {
       "epoch": 0.4551762420205384,
       "grad_norm": 0.0,
       "learning_rate": 1.325439173298793e-05,
-      "loss": 2.2377,
-      "mean_token_accuracy": 0.772808302808303,
-      "mean_token_accuracy_utility": 0.6555189255189253,
+      "loss": 2.2267,
+      "mean_token_accuracy": 0.7765567765567767,
+      "mean_token_accuracy_utility": 0.6541575091575089,
       "step": 1230
     },
     {
       "epoch": 0.4570265519474512,
       "grad_norm": 0.0,
       "learning_rate": 1.3193226448260128e-05,
-      "loss": 2.3119,
-      "mean_token_accuracy": 0.7531562881562882,
-      "mean_token_accuracy_utility": 0.654793817293817,
+      "loss": 2.2979,
+      "mean_token_accuracy": 0.7575702075702073,
+      "mean_token_accuracy_utility": 0.6536732711732709,
       "step": 1235
     },
     {
       "epoch": 0.45887686187436394,
       "grad_norm": 0.0,
       "learning_rate": 1.3131927842691793e-05,
-      "loss": 2.3327,
-      "mean_token_accuracy": 0.751837606837607,
-      "mean_token_accuracy_utility": 0.6473443223443222,
+      "loss": 2.3195,
+      "mean_token_accuracy": 0.7553418803418805,
+      "mean_token_accuracy_utility": 0.646105006105006,
       "step": 1240
     },
     {
       "epoch": 0.4607271718012767,
       "grad_norm": 0.0,
       "learning_rate": 1.3070498475569507e-05,
-      "loss": 2.3402,
-      "mean_token_accuracy": 0.736929181929182,
-      "mean_token_accuracy_utility": 0.6586752136752134,
+      "loss": 2.3252,
+      "mean_token_accuracy": 0.7414529914529917,
+      "mean_token_accuracy_utility": 0.657301587301587,
       "step": 1245
     },
     {
       "epoch": 0.46257748172818947,
       "grad_norm": 0.0,
       "learning_rate": 1.3008940911639302e-05,
-      "loss": 2.2833,
-      "mean_token_accuracy": 0.7509768009768008,
-      "mean_token_accuracy_utility": 0.6708692332793953,
+      "loss": 2.2705,
+      "mean_token_accuracy": 0.7550732600732599,
+      "mean_token_accuracy_utility": 0.6698101642076235,
       "step": 1250
     },
     {
       "epoch": 0.46442779165510223,
       "grad_norm": 0.0,
       "learning_rate": 1.2947257720999577e-05,
-      "loss": 2.3993,
-      "mean_token_accuracy": 0.7617521367521369,
-      "mean_token_accuracy_utility": 0.6407753357753357,
+      "loss": 2.3859,
+      "mean_token_accuracy": 0.7657081807081806,
+      "mean_token_accuracy_utility": 0.6395970695970694,
       "step": 1255
     },
     {
       "epoch": 0.466278101582015,
       "grad_norm": 0.0,
       "learning_rate": 1.2885451478993777e-05,
-      "loss": 2.4043,
-      "mean_token_accuracy": 0.7456105006105005,
-      "mean_token_accuracy_utility": 0.6429792429792428,
+      "loss": 2.3902,
+      "mean_token_accuracy": 0.7498778998779,
+      "mean_token_accuracy_utility": 0.6417216117216116,
       "step": 1260
     },
     {
       "epoch": 0.46812841150892776,
       "grad_norm": 0.0,
       "learning_rate": 1.282352476610289e-05,
-      "loss": 2.3037,
-      "mean_token_accuracy": 0.7565811965811967,
-      "mean_token_accuracy_utility": 0.6546092796092793,
+      "loss": 2.2899,
+      "mean_token_accuracy": 0.7604395604395606,
+      "mean_token_accuracy_utility": 0.653351648351648,
       "step": 1265
     },
     {
       "epoch": 0.4699787214358405,
       "grad_norm": 0.0,
       "learning_rate": 1.2761480167837705e-05,
-      "loss": 2.1743,
-      "mean_token_accuracy": 0.7722527472527474,
-      "mean_token_accuracy_utility": 0.6655189255189253,
+      "loss": 2.1618,
+      "mean_token_accuracy": 0.7759645909645909,
+      "mean_token_accuracy_utility": 0.6643894993894992,
       "step": 1270
     },
     {
       "epoch": 0.4718290313627533,
       "grad_norm": 0.0,
       "learning_rate": 1.2699320274630847e-05,
-      "loss": 2.1692,
-      "mean_token_accuracy": 0.763980463980464,
-      "mean_token_accuracy_utility": 0.6744810744810743,
+      "loss": 2.1565,
+      "mean_token_accuracy": 0.7677533577533577,
+      "mean_token_accuracy_utility": 0.673095238095238,
       "step": 1275
     },
     {
       "epoch": 0.47367934128966604,
       "grad_norm": 0.0,
       "learning_rate": 1.263704768172864e-05,
-      "loss": 2.3873,
-      "mean_token_accuracy": 0.7426617826617825,
-      "mean_token_accuracy_utility": 0.6476923076923076,
+      "loss": 2.3739,
+      "mean_token_accuracy": 0.7463553113553113,
+      "mean_token_accuracy_utility": 0.6463186813186812,
       "step": 1280
     },
     {
       "epoch": 0.47552965121657875,
       "grad_norm": 0.0,
       "learning_rate": 1.257466498908276e-05,
-      "loss": 2.1712,
-      "mean_token_accuracy": 0.7627777777777777,
-      "mean_token_accuracy_utility": 0.6751221001220999,
+      "loss": 2.1574,
+      "mean_token_accuracy": 0.7667032967032967,
+      "mean_token_accuracy_utility": 0.6736263736263735,
       "step": 1285
     },
     {
       "epoch": 0.4773799611434915,
       "grad_norm": 0.0,
       "learning_rate": 1.2512174801241657e-05,
-      "loss": 2.2045,
-      "mean_token_accuracy": 0.7720818070818071,
-      "mean_token_accuracy_utility": 0.6593284493284493,
+      "loss": 2.1923,
+      "mean_token_accuracy": 0.775915750915751,
+      "mean_token_accuracy_utility": 0.6581562881562881,
       "step": 1290
     },
     {
       "epoch": 0.4792302710704043,
       "grad_norm": 0.0,
       "learning_rate": 1.2449579727241834e-05,
-      "loss": 2.2339,
-      "mean_token_accuracy": 0.7590964590964592,
-      "mean_token_accuracy_utility": 0.6636630036630035,
+      "loss": 2.2195,
+      "mean_token_accuracy": 0.7627838827838829,
+      "mean_token_accuracy_utility": 0.6625152625152623,
       "step": 1295
     },
     {
       "epoch": 0.48108058099731704,
       "grad_norm": 0.0,
       "learning_rate": 1.2386882380498918e-05,
-      "loss": 2.2791,
-      "mean_token_accuracy": 0.7577106227106226,
-      "mean_token_accuracy_utility": 0.6579426129426128,
+      "loss": 2.2666,
+      "mean_token_accuracy": 0.7613797313797314,
+      "mean_token_accuracy_utility": 0.6567094017094016,
       "step": 1300
     },
     {
       "epoch": 0.48108058099731704,
-      "eval_mean_token_accuracy": 0.7468722998134764,
-      "eval_reasoning_loss": 0.862819254398346,
-      "eval_reasoning_runtime": 27.5303,
-      "eval_reasoning_samples_per_second": 4.686,
-      "eval_reasoning_steps_per_second": 1.199,
+      "eval_mean_token_accuracy": 0.7507464213346562,
+      "eval_reasoning_loss": 0.8419963121414185,
+      "eval_reasoning_runtime": 27.6148,
+      "eval_reasoning_samples_per_second": 4.671,
+      "eval_reasoning_steps_per_second": 1.195,
       "step": 1300
     },
     {
       "epoch": 0.48108058099731704,
-      "eval_mean_token_accuracy": 0.6610985414674292,
-      "eval_utility_loss": 1.4333739280700684,
-      "eval_utility_runtime": 10.3717,
-      "eval_utility_samples_per_second": 4.724,
-      "eval_utility_steps_per_second": 1.253,
+      "eval_mean_token_accuracy": 0.6600860741266804,
+      "eval_utility_loss": 1.4383585453033447,
+      "eval_utility_runtime": 10.3873,
+      "eval_utility_samples_per_second": 4.717,
+      "eval_utility_steps_per_second": 1.252,
       "step": 1300
     },
     {
       "epoch": 0.4829308909242298,
       "grad_norm": 0.0,
       "learning_rate": 1.2324085378698529e-05,
-      "loss": 2.2041,
-      "mean_token_accuracy": 0.7787606837606839,
-      "mean_token_accuracy_utility": 0.6579548229548228,
+      "loss": 2.1925,
+      "mean_token_accuracy": 0.7830219780219781,
+      "mean_token_accuracy_utility": 0.6571001221001219,
       "step": 1305
     },
     {
       "epoch": 0.48478120085114257,
       "grad_norm": 0.0,
       "learning_rate": 1.2261191343687e-05,
-      "loss": 2.2181,
-      "mean_token_accuracy": 0.7590903540903542,
-      "mean_token_accuracy_utility": 0.6663736263736262,
+      "loss": 2.2039,
+      "mean_token_accuracy": 0.7627716727716729,
+      "mean_token_accuracy_utility": 0.6652503052503052,
       "step": 1310
     },
     {
       "epoch": 0.48663151077805533,
       "grad_norm": 0.0,
       "learning_rate": 1.219820290136192e-05,
-      "loss": 2.2732,
-      "mean_token_accuracy": 0.7741452991452993,
-      "mean_token_accuracy_utility": 0.6532783882783881,
+      "loss": 2.2611,
+      "mean_token_accuracy": 0.7781196581196583,
+      "mean_token_accuracy_utility": 0.652173382173382,
       "step": 1315
     },
     {
       "epoch": 0.4884818207049681,
       "grad_norm": 0.0,
       "learning_rate": 1.2135122681562481e-05,
-      "loss": 2.283,
-      "mean_token_accuracy": 0.7545543345543345,
-      "mean_token_accuracy_utility": 0.6598656898656897,
+      "loss": 2.2697,
+      "mean_token_accuracy": 0.7583821733821733,
+      "mean_token_accuracy_utility": 0.6584432234432234,
       "step": 1320
     },
     {
       "epoch": 0.49033213063188086,
       "grad_norm": 0.0,
       "learning_rate": 1.2071953317959692e-05,
-      "loss": 2.2122,
-      "mean_token_accuracy": 0.7629327882940731,
-      "mean_token_accuracy_utility": 0.6667643467643466,
+      "loss": 2.1991,
+      "mean_token_accuracy": 0.766903743496072,
+      "mean_token_accuracy_utility": 0.6652808302808302,
       "step": 1325
     },
     {
       "epoch": 0.4921824405587936,
       "grad_norm": 0.0,
       "learning_rate": 1.2008697447946421e-05,
-      "loss": 2.2079,
-      "mean_token_accuracy": 0.7614407814407815,
-      "mean_token_accuracy_utility": 0.6716056166056164,
+      "loss": 2.1952,
+      "mean_token_accuracy": 0.7652136752136754,
+      "mean_token_accuracy_utility": 0.6703846153846151,
       "step": 1330
     },
     {
       "epoch": 0.49403275048570633,
       "grad_norm": 0.0,
       "learning_rate": 1.1945357712527273e-05,
-      "loss": 2.2989,
-      "mean_token_accuracy": 0.7584371184371184,
-      "mean_token_accuracy_utility": 0.656037851037851,
+      "loss": 2.2854,
+      "mean_token_accuracy": 0.7628327228327227,
+      "mean_token_accuracy_utility": 0.6543772893772892,
       "step": 1335
     },
     {
       "epoch": 0.4958830604126191,
       "grad_norm": 0.0,
       "learning_rate": 1.1881936756208329e-05,
-      "loss": 2.2072,
-      "mean_token_accuracy": 0.7624236874236875,
-      "mean_token_accuracy_utility": 0.6648107448107445,
+      "loss": 2.194,
+      "mean_token_accuracy": 0.7661599511599511,
+      "mean_token_accuracy_utility": 0.6632661782661781,
       "step": 1340
     },
     {
       "epoch": 0.49773337033953186,
       "grad_norm": 0.0,
       "learning_rate": 1.1818437226886738e-05,
-      "loss": 2.2973,
-      "mean_token_accuracy": 0.7640903540903544,
-      "mean_token_accuracy_utility": 0.6511416361416359,
+      "loss": 2.2848,
+      "mean_token_accuracy": 0.7677960927960932,
+      "mean_token_accuracy_utility": 0.6497313797313795,
       "step": 1345
     },
     {
       "epoch": 0.4995836802664446,
       "grad_norm": 0.0,
       "learning_rate": 1.1754861775740163e-05,
-      "loss": 2.1516,
-      "mean_token_accuracy": 0.7615079365079365,
-      "mean_token_accuracy_utility": 0.6780402930402929,
+      "loss": 2.1392,
+      "mean_token_accuracy": 0.7656227106227106,
+      "mean_token_accuracy_utility": 0.6766300366300365,
       "step": 1350
     },
     {
       "epoch": 0.5014339901933574,
       "grad_norm": 0.0,
       "learning_rate": 1.1691213057116082e-05,
-      "loss": 2.2741,
-      "mean_token_accuracy": 0.767655677655678,
-      "mean_token_accuracy_utility": 0.6524603174603172,
+      "loss": 2.2614,
+      "mean_token_accuracy": 0.7713064713064715,
+      "mean_token_accuracy_utility": 0.6509645909645907,
       "step": 1355
     },
     {
       "epoch": 0.5032843001202701,
       "grad_norm": 0.0,
       "learning_rate": 1.1627493728420978e-05,
-      "loss": 2.2136,
-      "mean_token_accuracy": 0.7614590964590966,
-      "mean_token_accuracy_utility": 0.6708029465205212,
+      "loss": 2.2003,
+      "mean_token_accuracy": 0.7651282051282052,
+      "mean_token_accuracy_utility": 0.669395599710258,
       "step": 1360
     },
     {
       "epoch": 0.5051346100471829,
       "grad_norm": 0.0,
       "learning_rate": 1.1563706450009391e-05,
-      "loss": 2.3316,
-      "mean_token_accuracy": 0.7592307692307692,
-      "mean_token_accuracy_utility": 0.6496214896214895,
+      "loss": 2.3188,
+      "mean_token_accuracy": 0.7632112332112333,
+      "mean_token_accuracy_utility": 0.6485103785103783,
       "step": 1365
     },
     {
       "epoch": 0.5069849199740957,
       "grad_norm": 0.0,
       "learning_rate": 1.1499853885072827e-05,
-      "loss": 2.2335,
-      "mean_token_accuracy": 0.775885225885226,
-      "mean_token_accuracy_utility": 0.6519841269841268,
+      "loss": 2.2212,
+      "mean_token_accuracy": 0.7801343101343102,
+      "mean_token_accuracy_utility": 0.6507692307692305,
       "step": 1370
     },
     {
       "epoch": 0.5088352299010084,
       "grad_norm": 0.0,
       "learning_rate": 1.1435938699528586e-05,
-      "loss": 2.3224,
-      "mean_token_accuracy": 0.7431807081807082,
-      "mean_token_accuracy_utility": 0.6587667887667886,
+      "loss": 2.3092,
+      "mean_token_accuracy": 0.7471916971916972,
+      "mean_token_accuracy_utility": 0.6574664224664224,
       "step": 1375
     },
     {
       "epoch": 0.5106855398279212,
       "grad_norm": 0.0,
       "learning_rate": 1.137196356190845e-05,
-      "loss": 2.3588,
-      "mean_token_accuracy": 0.7545787545787547,
-      "mean_token_accuracy_utility": 0.6469939921670025,
+      "loss": 2.3463,
+      "mean_token_accuracy": 0.7582356532356533,
+      "mean_token_accuracy_utility": 0.6455063775132978,
       "step": 1380
     },
     {
       "epoch": 0.5125358497548339,
       "grad_norm": 0.0,
       "learning_rate": 1.1307931143247268e-05,
-      "loss": 2.1914,
-      "mean_token_accuracy": 0.7572832722832722,
-      "mean_token_accuracy_utility": 0.6729426129426128,
+      "loss": 2.178,
+      "mean_token_accuracy": 0.7616117216117215,
+      "mean_token_accuracy_utility": 0.6716971916971914,
       "step": 1385
     },
     {
       "epoch": 0.5143861596817467,
       "grad_norm": 0.0,
       "learning_rate": 1.1243844116971433e-05,
-      "loss": 2.2761,
-      "mean_token_accuracy": 0.7611416361416363,
-      "mean_token_accuracy_utility": 0.6569597069597068,
+      "loss": 2.2632,
+      "mean_token_accuracy": 0.764884004884005,
+      "mean_token_accuracy_utility": 0.6550732600732599,
       "step": 1390
     },
     {
       "epoch": 0.5162364696086594,
       "grad_norm": 0.0,
       "learning_rate": 1.1179705158787276e-05,
-      "loss": 2.2433,
-      "mean_token_accuracy": 0.7525824175824176,
-      "mean_token_accuracy_utility": 0.6625498366989057,
+      "loss": 2.2297,
+      "mean_token_accuracy": 0.7565873015873018,
+      "mean_token_accuracy_utility": 0.6613518289920888,
       "step": 1395
     },
     {
       "epoch": 0.5180867795355723,
       "grad_norm": 0.0,
       "learning_rate": 1.1115516946569333e-05,
-      "loss": 2.3115,
-      "mean_token_accuracy": 0.7641880341880342,
-      "mean_token_accuracy_utility": 0.6518803418803417,
+      "loss": 2.2992,
+      "mean_token_accuracy": 0.7674358974358974,
+      "mean_token_accuracy_utility": 0.6505128205128203,
       "step": 1400
     },
     {
       "epoch": 0.5180867795355723,
-      "eval_mean_token_accuracy": 0.7469516322457503,
-      "eval_reasoning_loss": 0.8616194725036621,
-      "eval_reasoning_runtime": 27.5513,
-      "eval_reasoning_samples_per_second": 4.682,
-      "eval_reasoning_steps_per_second": 1.198,
+      "eval_mean_token_accuracy": 0.7510942216824571,
+      "eval_reasoning_loss": 0.8408248424530029,
+      "eval_reasoning_runtime": 27.6636,
+      "eval_reasoning_samples_per_second": 4.663,
+      "eval_reasoning_steps_per_second": 1.193,
       "step": 1400
     },
     {
       "epoch": 0.5180867795355723,
-      "eval_mean_token_accuracy": 0.6612596074849435,
-      "eval_utility_loss": 1.4327154159545898,
-      "eval_utility_runtime": 10.3678,
-      "eval_utility_samples_per_second": 4.726,
-      "eval_utility_steps_per_second": 1.254,
+      "eval_mean_token_accuracy": 0.6602739204683729,
+      "eval_utility_loss": 1.4376753568649292,
+      "eval_utility_runtime": 10.4264,
+      "eval_utility_samples_per_second": 4.7,
+      "eval_utility_steps_per_second": 1.247,
       "step": 1400
     },
     {
       "epoch": 0.519937089462485,
       "grad_norm": 0.0,
       "learning_rate": 1.105128216024857e-05,
-      "loss": 2.2244,
-      "mean_token_accuracy": 0.7681746031746034,
-      "mean_token_accuracy_utility": 0.666026306845623,
+      "loss": 2.2123,
+      "mean_token_accuracy": 0.7718986568986571,
+      "mean_token_accuracy_utility": 0.6646397930459242,
       "step": 1405
     },
     {
       "epoch": 0.5217873993893978,
       "grad_norm": 0.0,
       "learning_rate": 1.0987003481700456e-05,
-      "loss": 2.227,
-      "mean_token_accuracy": 0.7672100122100123,
-      "mean_token_accuracy_utility": 0.6644993894993891,
+      "loss": 2.2142,
+      "mean_token_accuracy": 0.7707203907203908,
+      "mean_token_accuracy_utility": 0.663272283272283,
       "step": 1410
     },
     {
       "epoch": 0.5236377093163105,
       "grad_norm": 0.0,
       "learning_rate": 1.092268359463302e-05,
-      "loss": 2.2254,
-      "mean_token_accuracy": 0.7700783866281763,
-      "mean_token_accuracy_utility": 0.6590903540903539,
+      "loss": 2.2129,
+      "mean_token_accuracy": 0.7745994398308564,
+      "mean_token_accuracy_utility": 0.6578754578754576,
       "step": 1415
     },
     {
       "epoch": 0.5254880192432232,
       "grad_norm": 0.0,
       "learning_rate": 1.0858325184474796e-05,
-      "loss": 2.2486,
-      "mean_token_accuracy": 0.7635225885225886,
-      "mean_token_accuracy_utility": 0.6603601953601952,
+      "loss": 2.2353,
+      "mean_token_accuracy": 0.7671489621489622,
+      "mean_token_accuracy_utility": 0.6595238095238094,
       "step": 1420
     },
     {
       "epoch": 0.527338329170136,
       "grad_norm": 0.0,
       "learning_rate": 1.0793930938262689e-05,
-      "loss": 2.3198,
-      "mean_token_accuracy": 0.751868131868132,
-      "mean_token_accuracy_utility": 0.6544200244200243,
+      "loss": 2.3065,
+      "mean_token_accuracy": 0.7556715506715507,
+      "mean_token_accuracy_utility": 0.653009768009768,
       "step": 1425
     },
     {
       "epoch": 0.5291886390970487,
       "grad_norm": 0.0,
       "learning_rate": 1.0729503544529814e-05,
-      "loss": 2.279,
-      "mean_token_accuracy": 0.7633516483516484,
-      "mean_token_accuracy_utility": 0.6550488400488399,
+      "loss": 2.2661,
+      "mean_token_accuracy": 0.7668131868131871,
+      "mean_token_accuracy_utility": 0.6538949938949937,
       "step": 1430
     },
     {
       "epoch": 0.5310389490239615,
       "grad_norm": 0.0,
       "learning_rate": 1.0665045693193226e-05,
-      "loss": 2.3183,
-      "mean_token_accuracy": 0.7608058608058608,
-      "mean_token_accuracy_utility": 0.6494810744810743,
+      "loss": 2.306,
+      "mean_token_accuracy": 0.7648412698412697,
+      "mean_token_accuracy_utility": 0.6479426129426128,
       "step": 1435
     },
     {
       "epoch": 0.5328892589508742,
       "grad_norm": 0.0,
       "learning_rate": 1.0600560075441617e-05,
-      "loss": 2.2437,
-      "mean_token_accuracy": 0.7742551892551895,
-      "mean_token_accuracy_utility": 0.6542979242979241,
+      "loss": 2.232,
+      "mean_token_accuracy": 0.7783821733821734,
+      "mean_token_accuracy_utility": 0.6531440781440778,
       "step": 1440
     },
     {
       "epoch": 0.5347395688777871,
       "grad_norm": 0.0,
       "learning_rate": 1.0536049383622966e-05,
-      "loss": 2.2613,
-      "mean_token_accuracy": 0.7498717948717949,
-      "mean_token_accuracy_utility": 0.6667582417582416,
+      "loss": 2.2471,
+      "mean_token_accuracy": 0.7543528693528694,
+      "mean_token_accuracy_utility": 0.6655433455433454,
       "step": 1445
     },
     {
       "epoch": 0.5365898788046998,
       "grad_norm": 0.0,
       "learning_rate": 1.047151631113212e-05,
-      "loss": 2.1868,
-      "mean_token_accuracy": 0.7667094017094019,
-      "mean_token_accuracy_utility": 0.6673687423687421,
+      "loss": 2.1748,
+      "mean_token_accuracy": 0.7704212454212456,
+      "mean_token_accuracy_utility": 0.6658241758241756,
       "step": 1450
     },
     {
       "epoch": 0.5384401887316126,
       "grad_norm": 0.0,
       "learning_rate": 1.0406963552298332e-05,
-      "loss": 2.2013,
-      "mean_token_accuracy": 0.7744749694749696,
-      "mean_token_accuracy_utility": 0.6608485958485956,
+      "loss": 2.1883,
+      "mean_token_accuracy": 0.7783455433455435,
+      "mean_token_accuracy_utility": 0.6594871794871793,
       "step": 1455
     },
     {
       "epoch": 0.5402904986585253,
       "grad_norm": 0.0,
       "learning_rate": 1.034239380227281e-05,
-      "loss": 2.2206,
-      "mean_token_accuracy": 0.7712393162393164,
-      "mean_token_accuracy_utility": 0.6618412227844275,
+      "loss": 2.208,
+      "mean_token_accuracy": 0.7753601953601956,
+      "mean_token_accuracy_utility": 0.6604146946134775,
       "step": 1460
     },
     {
       "epoch": 0.542140808585438,
       "grad_norm": 0.0,
       "learning_rate": 1.0277809756916134e-05,
-      "loss": 2.3461,
-      "mean_token_accuracy": 0.7512881562881562,
-      "mean_token_accuracy_utility": 0.6508058608058606,
+      "loss": 2.3324,
+      "mean_token_accuracy": 0.7551037851037848,
+      "mean_token_accuracy_utility": 0.6497802197802196,
       "step": 1465
     },
     {
       "epoch": 0.5439911185123508,
       "grad_norm": 0.0,
       "learning_rate": 1.0213214112685747e-05,
-      "loss": 2.248,
-      "mean_token_accuracy": 0.763913308913309,
-      "mean_token_accuracy_utility": 0.6586446886446884,
+      "loss": 2.2359,
+      "mean_token_accuracy": 0.7679120879120881,
+      "mean_token_accuracy_utility": 0.6580036630036628,
       "step": 1470
     },
     {
       "epoch": 0.5458414284392635,
       "grad_norm": 0.0,
       "learning_rate": 1.0148609566523358e-05,
-      "loss": 2.2426,
-      "mean_token_accuracy": 0.7550610500610502,
-      "mean_token_accuracy_utility": 0.6654090354090353,
+      "loss": 2.2297,
+      "mean_token_accuracy": 0.7584065934065936,
+      "mean_token_accuracy_utility": 0.6639621489621488,
       "step": 1475
     },
     {
       "epoch": 0.5476917383661764,
       "grad_norm": 0.0,
       "learning_rate": 1.0083998815742335e-05,
-      "loss": 2.1898,
-      "mean_token_accuracy": 0.7538278388278388,
-      "mean_token_accuracy_utility": 0.6799877899877897,
+      "loss": 2.1753,
+      "mean_token_accuracy": 0.7581868131868132,
+      "mean_token_accuracy_utility": 0.6789621489621489,
       "step": 1480
     },
     {
       "epoch": 0.5495420482930891,
       "grad_norm": 0.0,
       "learning_rate": 1.0019384557915099e-05,
-      "loss": 2.2374,
-      "mean_token_accuracy": 0.7595115995115995,
-      "mean_token_accuracy_utility": 0.661190476190476,
+      "loss": 2.2246,
+      "mean_token_accuracy": 0.7636935286935287,
+      "mean_token_accuracy_utility": 0.6599328449328448,
       "step": 1485
     },
     {
       "epoch": 0.5513923582200019,
       "grad_norm": 0.0,
       "learning_rate": 9.9547694907605e-06,
-      "loss": 2.205,
-      "mean_token_accuracy": 0.7605189255189255,
-      "mean_token_accuracy_utility": 0.6662332112332111,
+      "loss": 2.1929,
+      "mean_token_accuracy": 0.7642551892551893,
+      "mean_token_accuracy_utility": 0.6647924297924297,
       "step": 1490
     },
     {
       "epoch": 0.5532426681469146,
       "grad_norm": 0.0,
       "learning_rate": 9.890156312031165e-06,
-      "loss": 2.291,
-      "mean_token_accuracy": 0.7640537240537241,
-      "mean_token_accuracy_utility": 0.659762373733567,
+      "loss": 2.2781,
+      "mean_token_accuracy": 0.7681929181929182,
+      "mean_token_accuracy_utility": 0.6580332668398512,
       "step": 1495
     },
     {
       "epoch": 0.5550929780738274,
       "grad_norm": 0.0,
       "learning_rate": 9.825547719400889e-06,
-      "loss": 2.2714,
-      "mean_token_accuracy": 0.7667521367521368,
-      "mean_token_accuracy_utility": 0.6495665445665444,
+      "loss": 2.2591,
+      "mean_token_accuracy": 0.7707448107448108,
+      "mean_token_accuracy_utility": 0.6483089133089132,
       "step": 1500
     },
     {
       "epoch": 0.5550929780738274,
-      "eval_mean_token_accuracy": 0.7471368500780267,
-      "eval_reasoning_loss": 0.860738217830658,
-      "eval_reasoning_runtime": 27.5487,
-      "eval_reasoning_samples_per_second": 4.683,
-      "eval_reasoning_steps_per_second": 1.198,
+      "eval_mean_token_accuracy": 0.7511552717435074,
+      "eval_reasoning_loss": 0.839961588382721,
+      "eval_reasoning_runtime": 27.6591,
+      "eval_reasoning_samples_per_second": 4.664,
+      "eval_reasoning_steps_per_second": 1.193,
       "step": 1500
     },
     {
       "epoch": 0.5550929780738274,
-      "eval_mean_token_accuracy": 0.6610910520547352,
-      "eval_utility_loss": 1.4322147369384766,
-      "eval_utility_runtime": 10.3673,
-      "eval_utility_samples_per_second": 4.726,
-      "eval_utility_steps_per_second": 1.254,
+      "eval_mean_token_accuracy": 0.6602034780902382,
+      "eval_utility_loss": 1.437183141708374,
+      "eval_utility_runtime": 10.4135,
+      "eval_utility_samples_per_second": 4.705,
+      "eval_utility_steps_per_second": 1.248,
       "step": 1500
     },
     {
       "epoch": 0.5569432880007401,
       "grad_norm": 0.0,
       "learning_rate": 9.760946410351988e-06,
-      "loss": 2.3388,
-      "mean_token_accuracy": 0.7484004884004885,
-      "mean_token_accuracy_utility": 0.6508974358974358,
+      "loss": 2.3259,
+      "mean_token_accuracy": 0.7527960927960928,
+      "mean_token_accuracy_utility": 0.6493833943833942,
       "step": 1505
     },
     {
       "epoch": 0.5587935979276529,
       "grad_norm": 0.0,
       "learning_rate": 9.696355082062679e-06,
-      "loss": 2.3666,
-      "mean_token_accuracy": 0.7553296703296704,
-      "mean_token_accuracy_utility": 0.6386080586080585,
+      "loss": 2.3531,
+      "mean_token_accuracy": 0.7594139194139193,
+      "mean_token_accuracy_utility": 0.637252747252747,
       "step": 1510
     },
     {
       "epoch": 0.5606439078545656,
       "grad_norm": 0.0,
       "learning_rate": 9.631776431294475e-06,
-      "loss": 2.2935,
-      "mean_token_accuracy": 0.7603235653235654,
-      "mean_token_accuracy_utility": 0.6572405372405371,
+      "loss": 2.2821,
+      "mean_token_accuracy": 0.763870573870574,
+      "mean_token_accuracy_utility": 0.6556593406593405,
       "step": 1515
     },
     {
       "epoch": 0.5624942177814783,
       "grad_norm": 0.0,
       "learning_rate": 9.567213154279582e-06,
-      "loss": 2.2054,
-      "mean_token_accuracy": 0.7691269841269841,
-      "mean_token_accuracy_utility": 0.6615384615384613,
+      "loss": 2.193,
+      "mean_token_accuracy": 0.7727106227106229,
+      "mean_token_accuracy_utility": 0.6600671550671549,
       "step": 1520
     },
     {
       "epoch": 0.5643445277083912,
       "grad_norm": 0.0,
       "learning_rate": 9.502667946608332e-06,
-      "loss": 2.2019,
-      "mean_token_accuracy": 0.7610012210012209,
-      "mean_token_accuracy_utility": 0.6723626373626371,
+      "loss": 2.1896,
+      "mean_token_accuracy": 0.7646703296703298,
+      "mean_token_accuracy_utility": 0.6711111111111109,
       "step": 1525
     },
     {
       "epoch": 0.5661948376353039,
       "grad_norm": 0.0,
       "learning_rate": 9.43814350311666e-06,
-      "loss": 2.2283,
-      "mean_token_accuracy": 0.7586507936507936,
-      "mean_token_accuracy_utility": 0.6692633463000822,
+      "loss": 2.2153,
+      "mean_token_accuracy": 0.7624236874236874,
+      "mean_token_accuracy_utility": 0.6682257805642784,
       "step": 1530
     },
     {
       "epoch": 0.5680451475622167,
       "grad_norm": 0.0,
       "learning_rate": 9.37364251777355e-06,
-      "loss": 2.283,
-      "mean_token_accuracy": 0.7475152625152626,
-      "mean_token_accuracy_utility": 0.6619230769230766,
+      "loss": 2.2695,
+      "mean_token_accuracy": 0.7515262515262516,
+      "mean_token_accuracy_utility": 0.6604456654456652,
       "step": 1535
     },
     {
       "epoch": 0.5698954574891294,
       "grad_norm": 0.0,
       "learning_rate": 9.309167683568597e-06,
-      "loss": 2.3721,
-      "mean_token_accuracy": 0.7540231990231993,
-      "mean_token_accuracy_utility": 0.642240537240537,
+      "loss": 2.359,
+      "mean_token_accuracy": 0.7582234432234436,
+      "mean_token_accuracy_utility": 0.641294261294261,
       "step": 1540
     },
     {
       "epoch": 0.5717457674160422,
       "grad_norm": 0.0,
       "learning_rate": 9.244721692399545e-06,
-      "loss": 2.1845,
-      "mean_token_accuracy": 0.7690720390720391,
-      "mean_token_accuracy_utility": 0.666233211233211,
+      "loss": 2.1729,
+      "mean_token_accuracy": 0.773052503052503,
+      "mean_token_accuracy_utility": 0.6651343101343099,
       "step": 1545
     },
     {
       "epoch": 0.5735960773429549,
       "grad_norm": 0.0,
       "learning_rate": 9.180307234959918e-06,
-      "loss": 2.1693,
-      "mean_token_accuracy": 0.7589072039072039,
-      "mean_token_accuracy_utility": 0.6726617826617824,
+      "loss": 2.1574,
+      "mean_token_accuracy": 0.7627228327228326,
+      "mean_token_accuracy_utility": 0.6713003663003662,
       "step": 1550
     },
     {
       "epoch": 0.5754463872698677,
       "grad_norm": 0.0,
       "learning_rate": 9.115927000626665e-06,
-      "loss": 2.2175,
-      "mean_token_accuracy": 0.7567765567765568,
-      "mean_token_accuracy_utility": 0.6639010989010988,
+      "loss": 2.2045,
+      "mean_token_accuracy": 0.760921855921856,
+      "mean_token_accuracy_utility": 0.6626800976800975,
       "step": 1555
     },
     {
       "epoch": 0.5772966971967805,
       "grad_norm": 0.0,
       "learning_rate": 9.051583677347879e-06,
-      "loss": 2.288,
-      "mean_token_accuracy": 0.7574664224664225,
-      "mean_token_accuracy_utility": 0.6515995115995115,
+      "loss": 2.2752,
+      "mean_token_accuracy": 0.7616056166056167,
+      "mean_token_accuracy_utility": 0.6503296703296704,
       "step": 1560
     },
     {
       "epoch": 0.5791470071236933,
       "grad_norm": 0.0,
       "learning_rate": 8.987279951530586e-06,
-      "loss": 2.1983,
-      "mean_token_accuracy": 0.7581257631257633,
-      "mean_token_accuracy_utility": 0.6703724053724052,
+      "loss": 2.1857,
+      "mean_token_accuracy": 0.761825396825397,
+      "mean_token_accuracy_utility": 0.6695299145299144,
       "step": 1565
     },
     {
       "epoch": 0.580997317050606,
       "grad_norm": 0.0,
       "learning_rate": 8.923018507928564e-06,
-      "loss": 2.3404,
-      "mean_token_accuracy": 0.7521367521367524,
-      "mean_token_accuracy_utility": 0.6509340659340658,
+      "loss": 2.3273,
+      "mean_token_accuracy": 0.7559951159951162,
+      "mean_token_accuracy_utility": 0.6499328449328448,
       "step": 1570
     },
     {
       "epoch": 0.5828476269775187,
       "grad_norm": 0.0,
       "learning_rate": 8.85880202953026e-06,
-      "loss": 2.1934,
-      "mean_token_accuracy": 0.7620390720390724,
-      "mean_token_accuracy_utility": 0.6716117216117216,
+      "loss": 2.1809,
+      "mean_token_accuracy": 0.7660012210012213,
+      "mean_token_accuracy_utility": 0.6699572649572648,
       "step": 1575
     },
     {
       "epoch": 0.5846979369044315,
       "grad_norm": 0.0,
       "learning_rate": 8.79463319744677e-06,
-      "loss": 2.2948,
-      "mean_token_accuracy": 0.7448778998778998,
-      "mean_token_accuracy_utility": 0.6644871794871794,
+      "loss": 2.2816,
+      "mean_token_accuracy": 0.7488583638583639,
+      "mean_token_accuracy_utility": 0.6628571428571428,
       "step": 1580
     },
     {
       "epoch": 0.5865482468313442,
       "grad_norm": 0.0,
       "learning_rate": 8.730514690799916e-06,
-      "loss": 2.1902,
-      "mean_token_accuracy": 0.7637301587301586,
-      "mean_token_accuracy_utility": 0.6685653235653236,
+      "loss": 2.1778,
+      "mean_token_accuracy": 0.7672222222222222,
+      "mean_token_accuracy_utility": 0.667142857142857,
       "step": 1585
     },
     {
       "epoch": 0.588398556758257,
       "grad_norm": 0.0,
       "learning_rate": 8.666449186610353e-06,
-      "loss": 2.2385,
-      "mean_token_accuracy": 0.7672344322344326,
-      "mean_token_accuracy_utility": 0.6614285714285713,
+      "loss": 2.2255,
+      "mean_token_accuracy": 0.7711050061050062,
+      "mean_token_accuracy_utility": 0.6602747252747251,
       "step": 1590
     },
     {
       "epoch": 0.5902488666851697,
       "grad_norm": 0.0,
       "learning_rate": 8.60243935968585e-06,
-      "loss": 2.1982,
-      "mean_token_accuracy": 0.7644139194139197,
-      "mean_token_accuracy_utility": 0.6680769230769229,
+      "loss": 2.1858,
+      "mean_token_accuracy": 0.7685103785103787,
+      "mean_token_accuracy_utility": 0.6666544566544564,
       "step": 1595
     },
     {
       "epoch": 0.5920991766120826,
       "grad_norm": 0.0,
       "learning_rate": 8.538487882509568e-06,
-      "loss": 2.2098,
-      "mean_token_accuracy": 0.7627472527472527,
-      "mean_token_accuracy_utility": 0.6693650793650792,
+      "loss": 2.1971,
+      "mean_token_accuracy": 0.7665018315018315,
+      "mean_token_accuracy_utility": 0.668241758241758,
       "step": 1600
     },
     {
       "epoch": 0.5920991766120826,
-      "eval_mean_token_accuracy": 0.7471292324233499,
-      "eval_reasoning_loss": 0.8601270318031311,
-      "eval_reasoning_runtime": 27.5464,
-      "eval_reasoning_samples_per_second": 4.683,
-      "eval_reasoning_steps_per_second": 1.198,
+      "eval_mean_token_accuracy": 0.7512568041979807,
+      "eval_reasoning_loss": 0.8394526243209839,
+      "eval_reasoning_runtime": 27.6608,
+      "eval_reasoning_samples_per_second": 4.664,
+      "eval_reasoning_steps_per_second": 1.193,
       "step": 1600
     },
     {
       "epoch": 0.5920991766120826,
-      "eval_mean_token_accuracy": 0.6618212436056303,
-      "eval_utility_loss": 1.4319357872009277,
-      "eval_utility_runtime": 10.3672,
-      "eval_utility_samples_per_second": 4.726,
-      "eval_utility_steps_per_second": 1.254,
+      "eval_mean_token_accuracy": 0.6605631793936055,
+      "eval_utility_loss": 1.4368093013763428,
+      "eval_utility_runtime": 10.4129,
+      "eval_utility_samples_per_second": 4.706,
+      "eval_utility_steps_per_second": 1.248,
       "step": 1600
     },
     {
       "epoch": 0.5939494865389953,
       "grad_norm": 0.0,
       "learning_rate": 8.474597425128501e-06,
-      "loss": 2.2814,
-      "mean_token_accuracy": 0.7658852258852257,
-      "mean_token_accuracy_utility": 0.6518559218559217,
+      "loss": 2.2686,
+      "mean_token_accuracy": 0.7699145299145298,
+      "mean_token_accuracy_utility": 0.6505494505494503,
       "step": 1605
     },
     {
       "epoch": 0.5957997964659081,
       "grad_norm": 0.0,
       "learning_rate": 8.410770655042003e-06,
-      "loss": 2.1831,
-      "mean_token_accuracy": 0.7618192918192918,
-      "mean_token_accuracy_utility": 0.6732356532356532,
+      "loss": 2.1696,
+      "mean_token_accuracy": 0.7656776556776557,
+      "mean_token_accuracy_utility": 0.6718437118437117,
       "step": 1610
     },
     {
       "epoch": 0.5976501063928208,
       "grad_norm": 0.0,
       "learning_rate": 8.347010237090408e-06,
-      "loss": 2.3003,
-      "mean_token_accuracy": 0.7562820512820514,
-      "mean_token_accuracy_utility": 0.6559437701601879,
+      "loss": 2.287,
+      "mean_token_accuracy": 0.7597313797313798,
+      "mean_token_accuracy_utility": 0.6546492522926143,
       "step": 1615
     },
     {
       "epoch": 0.5995004163197336,
       "grad_norm": 0.0,
       "learning_rate": 8.283318833343773e-06,
-      "loss": 2.2424,
-      "mean_token_accuracy": 0.7503846153846154,
-      "mean_token_accuracy_utility": 0.6694993894993893,
+      "loss": 2.2303,
+      "mean_token_accuracy": 0.7543650793650792,
+      "mean_token_accuracy_utility": 0.6680647130647128,
       "step": 1620
     },
     {
       "epoch": 0.6013507262466463,
       "grad_norm": 0.0,
       "learning_rate": 8.219699102990735e-06,
-      "loss": 2.3024,
-      "mean_token_accuracy": 0.7544066727101442,
-      "mean_token_accuracy_utility": 0.6575152625152623,
+      "loss": 2.2899,
+      "mean_token_accuracy": 0.7587382599980584,
+      "mean_token_accuracy_utility": 0.6564529914529913,
       "step": 1625
     },
     {
       "epoch": 0.603201036173559,
       "grad_norm": 0.0,
       "learning_rate": 8.156153702227484e-06,
-      "loss": 2.2158,
-      "mean_token_accuracy": 0.7627594627594629,
-      "mean_token_accuracy_utility": 0.668980463980464,
+      "loss": 2.204,
+      "mean_token_accuracy": 0.7665445665445665,
+      "mean_token_accuracy_utility": 0.6676678876678876,
       "step": 1630
     },
     {
       "epoch": 0.6050513461004718,
       "grad_norm": 0.0,
       "learning_rate": 8.092685284146865e-06,
-      "loss": 2.2979,
-      "mean_token_accuracy": 0.7422710622710621,
-      "mean_token_accuracy_utility": 0.6632577854669311,
+      "loss": 2.2846,
+      "mean_token_accuracy": 0.7463614163614162,
+      "mean_token_accuracy_utility": 0.6617082645116704,
       "step": 1635
     },
     {
       "epoch": 0.6069016560273846,
       "grad_norm": 0.0,
       "learning_rate": 8.029296498627608e-06,
-      "loss": 2.3257,
-      "mean_token_accuracy": 0.7576617826617827,
-      "mean_token_accuracy_utility": 0.6526617392637586,
+      "loss": 2.3127,
+      "mean_token_accuracy": 0.7615873015873016,
+      "mean_token_accuracy_utility": 0.651193140399481,
       "step": 1640
     },
     {
       "epoch": 0.6087519659542974,
       "grad_norm": 0.0,
       "learning_rate": 7.965989992223693e-06,
-      "loss": 2.1723,
-      "mean_token_accuracy": 0.7647008547008547,
-      "mean_token_accuracy_utility": 0.6758730158730156,
+      "loss": 2.1605,
+      "mean_token_accuracy": 0.7682905982905984,
+      "mean_token_accuracy_utility": 0.6748290598290596,
       "step": 1645
     },
     {
       "epoch": 0.6106022758812101,
       "grad_norm": 0.0,
       "learning_rate": 7.90276840805385e-06,
-      "loss": 2.3312,
-      "mean_token_accuracy": 0.7527106227106228,
-      "mean_token_accuracy_utility": 0.6538958801748251,
+      "loss": 2.3189,
+      "mean_token_accuracy": 0.7558730158730157,
+      "mean_token_accuracy_utility": 0.6527022671030431,
       "step": 1650
     },
     {
       "epoch": 0.6124525858081229,
       "grad_norm": 0.0,
       "learning_rate": 7.839634385691214e-06,
-      "loss": 2.2218,
-      "mean_token_accuracy": 0.7602625152625153,
-      "mean_token_accuracy_utility": 0.6655921855921855,
+      "loss": 2.2085,
+      "mean_token_accuracy": 0.7647863247863247,
+      "mean_token_accuracy_utility": 0.6645482295482295,
       "step": 1655
     },
     {
       "epoch": 0.6143028957350356,
       "grad_norm": 0.0,
       "learning_rate": 7.776590561053117e-06,
-      "loss": 2.1891,
-      "mean_token_accuracy": 0.7545054945054945,
-      "mean_token_accuracy_utility": 0.6778083028083025,
+      "loss": 2.1761,
+      "mean_token_accuracy": 0.7587545787545791,
+      "mean_token_accuracy_utility": 0.6765506715506713,
       "step": 1660
     },
     {
       "epoch": 0.6161532056619484,
       "grad_norm": 0.0,
       "learning_rate": 7.713639566291028e-06,
-      "loss": 2.343,
-      "mean_token_accuracy": 0.7423565323565323,
-      "mean_token_accuracy_utility": 0.6539377289377288,
+      "loss": 2.329,
+      "mean_token_accuracy": 0.7463980463980463,
+      "mean_token_accuracy_utility": 0.6526007326007324,
       "step": 1665
     },
     {
       "epoch": 0.6180035155888611,
       "grad_norm": 0.0,
       "learning_rate": 7.650784029680662e-06,
-      "loss": 2.2156,
-      "mean_token_accuracy": 0.7761538461538461,
-      "mean_token_accuracy_utility": 0.6581562881562879,
+      "loss": 2.2026,
+      "mean_token_accuracy": 0.7802991452991452,
+      "mean_token_accuracy_utility": 0.6568437118437116,
       "step": 1670
     },
     {
       "epoch": 0.6198538255157738,
       "grad_norm": 0.0,
       "learning_rate": 7.58802657551225e-06,
-      "loss": 2.213,
-      "mean_token_accuracy": 0.7705982905982908,
-      "mean_token_accuracy_utility": 0.6595970695970694,
+      "loss": 2.2008,
+      "mean_token_accuracy": 0.7744871794871797,
+      "mean_token_accuracy_utility": 0.6580830280830279,
       "step": 1675
     },
     {
       "epoch": 0.6217041354426867,
       "grad_norm": 0.0,
       "learning_rate": 7.52536982398097e-06,
-      "loss": 2.2551,
-      "mean_token_accuracy": 0.7565323565323567,
-      "mean_token_accuracy_utility": 0.6644322344322343,
+      "loss": 2.2424,
+      "mean_token_accuracy": 0.7601526251526252,
+      "mean_token_accuracy_utility": 0.663058608058608,
       "step": 1680
     },
     {
       "epoch": 0.6235544453695994,
       "grad_norm": 0.0,
       "learning_rate": 7.46281639107755e-06,
-      "loss": 2.4302,
-      "mean_token_accuracy": 0.739005279029158,
-      "mean_token_accuracy_utility": 0.6453663003663002,
+      "loss": 2.4162,
+      "mean_token_accuracy": 0.7427586698893122,
+      "mean_token_accuracy_utility": 0.6444566544566543,
       "step": 1685
     },
     {
       "epoch": 0.6254047552965122,
       "grad_norm": 0.0,
       "learning_rate": 7.400368888479048e-06,
-      "loss": 2.2284,
-      "mean_token_accuracy": 0.7486691086691086,
-      "mean_token_accuracy_utility": 0.6753907203907202,
+      "loss": 2.2157,
+      "mean_token_accuracy": 0.751916971916972,
+      "mean_token_accuracy_utility": 0.674188034188034,
       "step": 1690
     },
     {
       "epoch": 0.6272550652234249,
       "grad_norm": 0.0,
       "learning_rate": 7.3380299234398076e-06,
-      "loss": 2.3693,
-      "mean_token_accuracy": 0.7491880341880341,
-      "mean_token_accuracy_utility": 0.6480097680097677,
+      "loss": 2.3558,
+      "mean_token_accuracy": 0.7534126984126983,
+      "mean_token_accuracy_utility": 0.6470024420024417,
       "step": 1695
     },
     {
       "epoch": 0.6291053751503377,
       "grad_norm": 0.0,
       "learning_rate": 7.275802098682612e-06,
-      "loss": 2.2874,
-      "mean_token_accuracy": 0.7458119658119655,
-      "mean_token_accuracy_utility": 0.6645299145299143,
+      "loss": 2.274,
+      "mean_token_accuracy": 0.7498962148962147,
+      "mean_token_accuracy_utility": 0.6632600732600731,
       "step": 1700
     },
     {
       "epoch": 0.6291053751503377,
-      "eval_mean_token_accuracy": 0.7473179326120505,
-      "eval_reasoning_loss": 0.8597175478935242,
-      "eval_reasoning_runtime": 27.5558,
-      "eval_reasoning_samples_per_second": 4.681,
-      "eval_reasoning_steps_per_second": 1.198,
+      "eval_mean_token_accuracy": 0.7513883719766073,
+      "eval_reasoning_loss": 0.839073121547699,
+      "eval_reasoning_runtime": 27.6637,
+      "eval_reasoning_samples_per_second": 4.663,
+      "eval_reasoning_steps_per_second": 1.193,
       "step": 1700
     },
     {
       "epoch": 0.6291053751503377,
-      "eval_mean_token_accuracy": 0.6612192449624534,
-      "eval_utility_loss": 1.431671142578125,
-      "eval_utility_runtime": 10.3674,
-      "eval_utility_samples_per_second": 4.726,
-      "eval_utility_steps_per_second": 1.254,
+      "eval_mean_token_accuracy": 0.6605556899809117,
+      "eval_utility_loss": 1.4366581439971924,
+      "eval_utility_runtime": 10.4053,
+      "eval_utility_samples_per_second": 4.709,
+      "eval_utility_steps_per_second": 1.249,
       "step": 1700
     },
     {
       "epoch": 0.6309556850772504,
       "grad_norm": 0.0,
       "learning_rate": 7.213688012290004e-06,
-      "loss": 2.1804,
-      "mean_token_accuracy": 0.7547496947496948,
-      "mean_token_accuracy_utility": 0.6772771672771671,
+      "loss": 2.1675,
+      "mean_token_accuracy": 0.7586507936507938,
+      "mean_token_accuracy_utility": 0.6759890109890108,
       "step": 1705
     },
     {
       "epoch": 0.6328059950041632,
       "grad_norm": 0.0,
       "learning_rate": 7.151690257595826e-06,
-      "loss": 2.1979,
-      "mean_token_accuracy": 0.7614590964590965,
-      "mean_token_accuracy_utility": 0.6695909645909643,
+      "loss": 2.1864,
+      "mean_token_accuracy": 0.7657142857142858,
+      "mean_token_accuracy_utility": 0.6683760683760682,
       "step": 1710
     },
     {
       "epoch": 0.6346563049310759,
       "grad_norm": 0.0,
       "learning_rate": 7.089811423076936e-06,
-      "loss": 2.2205,
-      "mean_token_accuracy": 0.7779792429792429,
-      "mean_token_accuracy_utility": 0.6566178266178264,
+      "loss": 2.2091,
+      "mean_token_accuracy": 0.7816178266178267,
+      "mean_token_accuracy_utility": 0.6552319902319901,
       "step": 1715
     },
     {
       "epoch": 0.6365066148579888,
       "grad_norm": 0.0,
       "learning_rate": 7.028054092245134e-06,
-      "loss": 2.2672,
-      "mean_token_accuracy": 0.7517216117216117,
-      "mean_token_accuracy_utility": 0.6617765567765566,
+      "loss": 2.253,
+      "mean_token_accuracy": 0.7555555555555555,
+      "mean_token_accuracy_utility": 0.6603968253968252,
       "step": 1720
     },
     {
       "epoch": 0.6383569247849015,
       "grad_norm": 0.0,
       "learning_rate": 6.966420843539321e-06,
-      "loss": 2.191,
-      "mean_token_accuracy": 0.748919413919414,
-      "mean_token_accuracy_utility": 0.678180708180708,
+      "loss": 2.178,
+      "mean_token_accuracy": 0.753064713064713,
+      "mean_token_accuracy_utility": 0.6768681318681317,
       "step": 1725
     },
     {
       "epoch": 0.6402072347118142,
       "grad_norm": 0.0,
       "learning_rate": 6.90491425021781e-06,
-      "loss": 2.2706,
-      "mean_token_accuracy": 0.7521245421245422,
-      "mean_token_accuracy_utility": 0.6664590964590963,
+      "loss": 2.2573,
+      "mean_token_accuracy": 0.7564835164835164,
+      "mean_token_accuracy_utility": 0.66520757020757,
       "step": 1730
     },
     {
       "epoch": 0.642057544638727,
       "grad_norm": 0.0,
       "learning_rate": 6.843536880250914e-06,
-      "loss": 2.2436,
-      "mean_token_accuracy": 0.7595421245421246,
-      "mean_token_accuracy_utility": 0.6639804639804637,
+      "loss": 2.2303,
+      "mean_token_accuracy": 0.763809523809524,
+      "mean_token_accuracy_utility": 0.6627228327228325,
       "step": 1735
     },
     {
       "epoch": 0.6439078545656397,
       "grad_norm": 0.0,
       "learning_rate": 6.7822912962137225e-06,
-      "loss": 2.1928,
-      "mean_token_accuracy": 0.7533882783882785,
-      "mean_token_accuracy_utility": 0.6705616605616604,
+      "loss": 2.1791,
+      "mean_token_accuracy": 0.7576129426129427,
+      "mean_token_accuracy_utility": 0.6692490842490841,
       "step": 1740
     },
     {
       "epoch": 0.6457581644925525,
       "grad_norm": 0.0,
       "learning_rate": 6.721180055179113e-06,
-      "loss": 2.2356,
-      "mean_token_accuracy": 0.7571075809101483,
-      "mean_token_accuracy_utility": 0.6705006105006103,
+      "loss": 2.2236,
+      "mean_token_accuracy": 0.7608585725483212,
+      "mean_token_accuracy_utility": 0.6687179487179484,
       "step": 1745
     },
     {
       "epoch": 0.6476084744194652,
       "grad_norm": 0.0,
       "learning_rate": 6.660205708610987e-06,
-      "loss": 2.2074,
-      "mean_token_accuracy": 0.7485592185592185,
-      "mean_token_accuracy_utility": 0.6787606837606835,
+      "loss": 2.194,
+      "mean_token_accuracy": 0.7530647130647132,
+      "mean_token_accuracy_utility": 0.6779731379731377,
       "step": 1750
     },
     {
       "epoch": 0.649458784346378,
       "grad_norm": 0.0,
       "learning_rate": 6.599370802257755e-06,
-      "loss": 2.1943,
-      "mean_token_accuracy": 0.7592307692307694,
-      "mean_token_accuracy_utility": 0.6728266178266176,
+      "loss": 2.1802,
+      "mean_token_accuracy": 0.7632417582417585,
+      "mean_token_accuracy_utility": 0.6714163614163612,
       "step": 1755
     },
     {
       "epoch": 0.6513090942732908,
       "grad_norm": 0.0,
       "learning_rate": 6.5386778760460316e-06,
-      "loss": 2.2914,
-      "mean_token_accuracy": 0.7704578754578757,
-      "mean_token_accuracy_utility": 0.6516056166056166,
+      "loss": 2.2793,
+      "mean_token_accuracy": 0.7740781440781442,
+      "mean_token_accuracy_utility": 0.6508913308913308,
       "step": 1760
     },
     {
       "epoch": 0.6531594042002036,
       "grad_norm": 0.0,
       "learning_rate": 6.478129463974598e-06,
-      "loss": 2.3373,
-      "mean_token_accuracy": 0.7481623931623932,
-      "mean_token_accuracy_utility": 0.6550061050061048,
+      "loss": 2.3239,
+      "mean_token_accuracy": 0.7523687423687423,
+      "mean_token_accuracy_utility": 0.6532844932844931,
       "step": 1765
     },
     {
       "epoch": 0.6550097141271163,
       "grad_norm": 0.0,
       "learning_rate": 6.417728094008613e-06,
-      "loss": 2.1645,
-      "mean_token_accuracy": 0.7687362637362638,
-      "mean_token_accuracy_utility": 0.6703724053724052,
+      "loss": 2.153,
+      "mean_token_accuracy": 0.7723626373626373,
+      "mean_token_accuracy_utility": 0.6691391941391939,
       "step": 1770
     },
     {
       "epoch": 0.6568600240540291,
       "grad_norm": 0.0,
       "learning_rate": 6.357476287974051e-06,
-      "loss": 2.173,
-      "mean_token_accuracy": 0.7580219780219781,
-      "mean_token_accuracy_utility": 0.6738827838827837,
+      "loss": 2.1599,
+      "mean_token_accuracy": 0.761758241758242,
+      "mean_token_accuracy_utility": 0.6721001221001219,
       "step": 1775
     },
     {
       "epoch": 0.6587103339809418,
       "grad_norm": 0.0,
       "learning_rate": 6.297376561452428e-06,
-      "loss": 2.2089,
-      "mean_token_accuracy": 0.7663980463980463,
-      "mean_token_accuracy_utility": 0.6652748425176407,
+      "loss": 2.1967,
+      "mean_token_accuracy": 0.7703663003663006,
+      "mean_token_accuracy_utility": 0.6639287096077218,
       "step": 1780
     },
     {
       "epoch": 0.6605606439078545,
       "grad_norm": 0.0,
       "learning_rate": 6.237431423675764e-06,
-      "loss": 2.2773,
-      "mean_token_accuracy": 0.7587301587301589,
-      "mean_token_accuracy_utility": 0.6597741147741145,
+      "loss": 2.2651,
+      "mean_token_accuracy": 0.7624786324786328,
+      "mean_token_accuracy_utility": 0.6583699633699631,
       "step": 1785
     },
     {
       "epoch": 0.6624109538347673,
       "grad_norm": 0.0,
       "learning_rate": 6.177643377421827e-06,
-      "loss": 2.2661,
-      "mean_token_accuracy": 0.7622771672771672,
-      "mean_token_accuracy_utility": 0.6565750915750914,
+      "loss": 2.2542,
+      "mean_token_accuracy": 0.7657936507936507,
+      "mean_token_accuracy_utility": 0.6549023199023198,
       "step": 1790
     },
     {
       "epoch": 0.66426126376168,
       "grad_norm": 0.0,
       "learning_rate": 6.118014918909633e-06,
-      "loss": 2.2548,
-      "mean_token_accuracy": 0.7557814407814408,
-      "mean_token_accuracy_utility": 0.659716852599381,
+      "loss": 2.2416,
+      "mean_token_accuracy": 0.7592185592185593,
+      "mean_token_accuracy_utility": 0.6584945048319979,
       "step": 1795
     },
     {
       "epoch": 0.6661115736885929,
       "grad_norm": 0.0,
       "learning_rate": 6.058548537695225e-06,
-      "loss": 2.2584,
-      "mean_token_accuracy": 0.7732417582417582,
-      "mean_token_accuracy_utility": 0.6547680097680096,
+      "loss": 2.2466,
+      "mean_token_accuracy": 0.7767399267399268,
+      "mean_token_accuracy_utility": 0.6535897435897435,
       "step": 1800
     },
     {
       "epoch": 0.6661115736885929,
-      "eval_mean_token_accuracy": 0.7473884503296266,
-      "eval_reasoning_loss": 0.8594432473182678,
-      "eval_reasoning_runtime": 27.5488,
-      "eval_reasoning_samples_per_second": 4.683,
-      "eval_reasoning_steps_per_second": 1.198,
+      "eval_mean_token_accuracy": 0.7515178721061075,
+      "eval_reasoning_loss": 0.8387583494186401,
+      "eval_reasoning_runtime": 27.6611,
+      "eval_reasoning_samples_per_second": 4.664,
+      "eval_reasoning_steps_per_second": 1.193,
       "step": 1800
     },
     {
       "epoch": 0.6661115736885929,
-      "eval_mean_token_accuracy": 0.661180279067039,
-      "eval_utility_loss": 1.4314672946929932,
-      "eval_utility_runtime": 10.372,
-      "eval_utility_samples_per_second": 4.724,
-      "eval_utility_steps_per_second": 1.253,
+      "eval_mean_token_accuracy": 0.6603913244319306,
+      "eval_utility_loss": 1.4364391565322876,
+      "eval_utility_runtime": 10.4171,
+      "eval_utility_samples_per_second": 4.704,
+      "eval_utility_steps_per_second": 1.248,
       "step": 1800
     },
     {
       "epoch": 0.6679618836155056,
       "grad_norm": 0.0,
       "learning_rate": 5.999246716567737e-06,
-      "loss": 2.206,
-      "mean_token_accuracy": 0.7483272283272283,
-      "mean_token_accuracy_utility": 0.6744871794871792,
+      "loss": 2.1917,
+      "mean_token_accuracy": 0.7526495726495724,
+      "mean_token_accuracy_utility": 0.6731501831501829,
       "step": 1805
     },
     {
       "epoch": 0.6698121935424184,
       "grad_norm": 0.0,
       "learning_rate": 5.940111931445731e-06,
-      "loss": 2.347,
-      "mean_token_accuracy": 0.74014652014652,
-      "mean_token_accuracy_utility": 0.6596642246642244,
+      "loss": 2.3335,
+      "mean_token_accuracy": 0.7438644688644689,
+      "mean_token_accuracy_utility": 0.6585225885225883,
       "step": 1810
     },
     {
       "epoch": 0.6716625034693311,
       "grad_norm": 0.0,
       "learning_rate": 5.881146651273825e-06,
-      "loss": 2.1653,
-      "mean_token_accuracy": 0.7721794871794875,
-      "mean_token_accuracy_utility": 0.6674297924297923,
+      "loss": 2.1534,
+      "mean_token_accuracy": 0.7760195360195363,
+      "mean_token_accuracy_utility": 0.6658608058608056,
       "step": 1815
     },
     {
       "epoch": 0.6735128133962439,
       "grad_norm": 0.0,
       "learning_rate": 5.822353337919616e-06,
-      "loss": 2.1967,
-      "mean_token_accuracy": 0.7620757020757021,
-      "mean_token_accuracy_utility": 0.6633638583638581,
+      "loss": 2.1849,
+      "mean_token_accuracy": 0.7662454212454213,
+      "mean_token_accuracy_utility": 0.6617460317460315,
       "step": 1820
     },
     {
       "epoch": 0.6753631233231566,
       "grad_norm": 0.0,
       "learning_rate": 5.763734446070892e-06,
-      "loss": 2.1175,
-      "mean_token_accuracy": 0.7752930402930402,
-      "mean_token_accuracy_utility": 0.6723809523809523,
+      "loss": 2.1066,
+      "mean_token_accuracy": 0.7788278388278389,
+      "mean_token_accuracy_utility": 0.6708669108669107,
       "step": 1825
     },
     {
       "epoch": 0.6772134332500694,
       "grad_norm": 0.0,
       "learning_rate": 5.705292423133133e-06,
-      "loss": 2.2402,
-      "mean_token_accuracy": 0.7752503052503055,
-      "mean_token_accuracy_utility": 0.6523321123321122,
+      "loss": 2.2294,
+      "mean_token_accuracy": 0.778742368742369,
+      "mean_token_accuracy_utility": 0.6510134310134308,
       "step": 1830
     },
     {
       "epoch": 0.6790637431769821,
       "grad_norm": 0.0,
       "learning_rate": 5.647029709127355e-06,
-      "loss": 2.318,
-      "mean_token_accuracy": 0.7520695970695972,
-      "mean_token_accuracy_utility": 0.6505677655677655,
+      "loss": 2.3046,
+      "mean_token_accuracy": 0.756117216117216,
+      "mean_token_accuracy_utility": 0.6495787545787544,
       "step": 1835
     },
     {
       "epoch": 0.6809140531038949,
       "grad_norm": 0.0,
       "learning_rate": 5.5889487365882065e-06,
-      "loss": 2.2089,
-      "mean_token_accuracy": 0.7644627594627593,
-      "mean_token_accuracy_utility": 0.6647496947496946,
+      "loss": 2.1968,
+      "mean_token_accuracy": 0.768095238095238,
+      "mean_token_accuracy_utility": 0.6633150183150182,
       "step": 1840
     },
     {
       "epoch": 0.6827643630308077,
       "grad_norm": 0.0,
       "learning_rate": 5.531051930462437e-06,
-      "loss": 2.2379,
-      "mean_token_accuracy": 0.7647008547008547,
-      "mean_token_accuracy_utility": 0.662112332112332,
+      "loss": 2.2253,
+      "mean_token_accuracy": 0.7685836385836387,
+      "mean_token_accuracy_utility": 0.6608547008547008,
       "step": 1845
     },
     {
       "epoch": 0.6846146729577204,
       "grad_norm": 0.0,
       "learning_rate": 5.4733417080076325e-06,
-      "loss": 2.2391,
-      "mean_token_accuracy": 0.7662576312576312,
-      "mean_token_accuracy_utility": 0.6570879120879118,
+      "loss": 2.2267,
+      "mean_token_accuracy": 0.7704029304029304,
+      "mean_token_accuracy_utility": 0.6558363858363856,
       "step": 1850
     },
     {
       "epoch": 0.6864649828846332,
       "grad_norm": 0.0,
       "learning_rate": 5.415820478691301e-06,
-      "loss": 2.1641,
-      "mean_token_accuracy": 0.7661111111111112,
-      "mean_token_accuracy_utility": 0.6717948717948716,
+      "loss": 2.152,
+      "mean_token_accuracy": 0.7695543345543348,
+      "mean_token_accuracy_utility": 0.6702380952380951,
       "step": 1855
     },
     {
       "epoch": 0.6883152928115459,
       "grad_norm": 0.0,
       "learning_rate": 5.358490644090263e-06,
-      "loss": 2.2619,
-      "mean_token_accuracy": 0.7537179487179488,
-      "mean_token_accuracy_utility": 0.6611599511599511,
+      "loss": 2.2486,
+      "mean_token_accuracy": 0.7573382173382173,
+      "mean_token_accuracy_utility": 0.66007326007326,
       "step": 1860
     },
     {
       "epoch": 0.6901656027384587,
       "grad_norm": 0.0,
       "learning_rate": 5.3013545977904005e-06,
-      "loss": 2.3518,
-      "mean_token_accuracy": 0.7463797313797313,
-      "mean_token_accuracy_utility": 0.6484737484737483,
+      "loss": 2.3385,
+      "mean_token_accuracy": 0.7501282051282053,
+      "mean_token_accuracy_utility": 0.6473809523809522,
       "step": 1865
     },
     {
       "epoch": 0.6920159126653714,
       "grad_norm": 0.0,
       "learning_rate": 5.244414725286717e-06,
-      "loss": 2.321,
-      "mean_token_accuracy": 0.7484554334554335,
-      "mean_token_accuracy_utility": 0.6595909645909644,
+      "loss": 2.3075,
+      "mean_token_accuracy": 0.7525396825396825,
+      "mean_token_accuracy_utility": 0.658217338217338,
       "step": 1870
     },
     {
       "epoch": 0.6938662225922843,
       "grad_norm": 0.0,
       "learning_rate": 5.187673403883721e-06,
-      "loss": 2.225,
-      "mean_token_accuracy": 0.7700427350427352,
-      "mean_token_accuracy_utility": 0.6584371184371183,
+      "loss": 2.2138,
+      "mean_token_accuracy": 0.7736752136752137,
+      "mean_token_accuracy_utility": 0.657173382173382,
       "step": 1875
     },
     {
       "epoch": 0.695716532519197,
       "grad_norm": 0.0,
       "learning_rate": 5.131133002596199e-06,
-      "loss": 2.2388,
-      "mean_token_accuracy": 0.7737728937728938,
-      "mean_token_accuracy_utility": 0.6514738787263726,
+      "loss": 2.2264,
+      "mean_token_accuracy": 0.7773931623931624,
+      "mean_token_accuracy_utility": 0.6503252661042375,
       "step": 1880
     },
     {
       "epoch": 0.6975668424461097,
       "grad_norm": 0.0,
       "learning_rate": 5.074795882050293e-06,
-      "loss": 2.2271,
-      "mean_token_accuracy": 0.75513431013431,
-      "mean_token_accuracy_utility": 0.6701953601953601,
+      "loss": 2.2139,
+      "mean_token_accuracy": 0.7589499389499389,
+      "mean_token_accuracy_utility": 0.6692246642246642,
       "step": 1885
     },
     {
       "epoch": 0.6994171523730225,
       "grad_norm": 0.0,
       "learning_rate": 5.018664394384942e-06,
-      "loss": 2.2963,
-      "mean_token_accuracy": 0.7683882783882786,
-      "mean_token_accuracy_utility": 0.648119658119658,
+      "loss": 2.2841,
+      "mean_token_accuracy": 0.7725518925518926,
+      "mean_token_accuracy_utility": 0.6468620268620267,
       "step": 1890
     },
     {
       "epoch": 0.7012674622999352,
       "grad_norm": 0.0,
       "learning_rate": 4.9627408831536705e-06,
-      "loss": 2.2631,
-      "mean_token_accuracy": 0.7619108669108673,
-      "mean_token_accuracy_utility": 0.6594200244200243,
+      "loss": 2.2509,
+      "mean_token_accuracy": 0.7655494505494509,
+      "mean_token_accuracy_utility": 0.6579242979242979,
       "step": 1895
     },
     {
       "epoch": 0.703117772226848,
       "grad_norm": 0.0,
       "learning_rate": 4.907027683226761e-06,
-      "loss": 2.1867,
-      "mean_token_accuracy": 0.7667778316759639,
-      "mean_token_accuracy_utility": 0.6677228327228326,
+      "loss": 2.1753,
+      "mean_token_accuracy": 0.7699871473555684,
+      "mean_token_accuracy_utility": 0.6663736263736263,
       "step": 1900
     },
     {
       "epoch": 0.703117772226848,
-      "eval_mean_token_accuracy": 0.747469850411027,
-      "eval_reasoning_loss": 0.8592001795768738,
-      "eval_reasoning_runtime": 27.5455,
-      "eval_reasoning_samples_per_second": 4.683,
-      "eval_reasoning_steps_per_second": 1.198,
+      "eval_mean_token_accuracy": 0.7514956720839072,
+      "eval_reasoning_loss": 0.838676393032074,
+      "eval_reasoning_runtime": 27.6564,
+      "eval_reasoning_samples_per_second": 4.664,
+      "eval_reasoning_steps_per_second": 1.193,
       "step": 1900
     },
     {
       "epoch": 0.703117772226848,
-      "eval_mean_token_accuracy": 0.6612427257551652,
-      "eval_utility_loss": 1.4313732385635376,
-      "eval_utility_runtime": 10.3748,
-      "eval_utility_samples_per_second": 4.723,
-      "eval_utility_steps_per_second": 1.253,
+      "eval_mean_token_accuracy": 0.6602067776217047,
+      "eval_utility_loss": 1.4363669157028198,
+      "eval_utility_runtime": 10.4118,
+      "eval_utility_samples_per_second": 4.706,
+      "eval_utility_steps_per_second": 1.249,
       "step": 1900
     },
     {
       "epoch": 0.7049680821537607,
       "grad_norm": 0.0,
       "learning_rate": 4.85152712069375e-06,
-      "loss": 2.2479,
-      "mean_token_accuracy": 0.7604945054945058,
-      "mean_token_accuracy_utility": 0.6614713064713063,
+      "loss": 2.2355,
+      "mean_token_accuracy": 0.7643833943833946,
+      "mean_token_accuracy_utility": 0.6605860805860804,
       "step": 1905
     },
     {
       "epoch": 0.7068183920806735,
       "grad_norm": 0.0,
       "learning_rate": 4.7962415127663265e-06,
-      "loss": 2.2847,
-      "mean_token_accuracy": 0.7630036630036632,
-      "mean_token_accuracy_utility": 0.6554456654456653,
+      "loss": 2.2721,
+      "mean_token_accuracy": 0.7672893772893774,
+      "mean_token_accuracy_utility": 0.6539010989010987,
       "step": 1910
     },
     {
       "epoch": 0.7086687020075862,
       "grad_norm": 0.0,
       "learning_rate": 4.74117316768158e-06,
-      "loss": 2.293,
-      "mean_token_accuracy": 0.7551526251526252,
-      "mean_token_accuracy_utility": 0.6594566544566544,
+      "loss": 2.2803,
+      "mean_token_accuracy": 0.7589133089133088,
+      "mean_token_accuracy_utility": 0.658052503052503,
       "step": 1915
     },
     {
       "epoch": 0.7105190119344991,
       "grad_norm": 0.0,
       "learning_rate": 4.686324384605629e-06,
-      "loss": 2.2567,
-      "mean_token_accuracy": 0.7551343101343101,
-      "mean_token_accuracy_utility": 0.6641025641025639,
+      "loss": 2.2444,
+      "mean_token_accuracy": 0.758913308913309,
+      "mean_token_accuracy_utility": 0.6631562881562879,
       "step": 1920
     },
     {
       "epoch": 0.7123693218614118,
       "grad_norm": 0.0,
       "learning_rate": 4.631697453537623e-06,
-      "loss": 2.2813,
-      "mean_token_accuracy": 0.7577655677655679,
-      "mean_token_accuracy_utility": 0.6588339438339437,
+      "loss": 2.2702,
+      "mean_token_accuracy": 0.7617155067155068,
+      "mean_token_accuracy_utility": 0.6575213675213674,
       "step": 1925
     },
     {
       "epoch": 0.7142196317883246,
       "grad_norm": 0.0,
       "learning_rate": 4.577294655214144e-06,
-      "loss": 2.236,
-      "mean_token_accuracy": 0.754053724053724,
-      "mean_token_accuracy_utility": 0.667062717715291,
+      "loss": 2.2224,
+      "mean_token_accuracy": 0.7577167277167278,
+      "mean_token_accuracy_utility": 0.665872017076061,
       "step": 1930
     },
     {
       "epoch": 0.7160699417152373,
       "grad_norm": 0.0,
       "learning_rate": 4.523118261013969e-06,
-      "loss": 2.2863,
-      "mean_token_accuracy": 0.7531440781440781,
-      "mean_token_accuracy_utility": 0.6620753261240393,
+      "loss": 2.2733,
+      "mean_token_accuracy": 0.7572893772893772,
+      "mean_token_accuracy_utility": 0.6605134377468935,
       "step": 1935
     },
     {
       "epoch": 0.71792025164215,
       "grad_norm": 0.0,
       "learning_rate": 4.469170532863254e-06,
-      "loss": 2.2936,
-      "mean_token_accuracy": 0.7436568986568987,
-      "mean_token_accuracy_utility": 0.6659951159951158,
+      "loss": 2.2804,
+      "mean_token_accuracy": 0.7474297924297926,
+      "mean_token_accuracy_utility": 0.6642612942612941,
       "step": 1940
     },
     {
       "epoch": 0.7197705615690628,
       "grad_norm": 0.0,
       "learning_rate": 4.415453723141081e-06,
-      "loss": 2.3858,
-      "mean_token_accuracy": 0.7492124542124543,
-      "mean_token_accuracy_utility": 0.6512576312576311,
+      "loss": 2.373,
+      "mean_token_accuracy": 0.7530647130647131,
+      "mean_token_accuracy_utility": 0.6501404151404151,
       "step": 1945
     },
     {
       "epoch": 0.7216208714959755,
       "grad_norm": 0.0,
       "learning_rate": 4.361970074585426e-06,
-      "loss": 2.1146,
-      "mean_token_accuracy": 0.7679548229548232,
-      "mean_token_accuracy_utility": 0.6836813186813184,
+      "loss": 2.1029,
+      "mean_token_accuracy": 0.7720573870573871,
+      "mean_token_accuracy_utility": 0.682222222222222,
       "step": 1950
     },
     {
       "epoch": 0.7234711814228884,
       "grad_norm": 0.0,
       "learning_rate": 4.308721820199529e-06,
-      "loss": 2.3013,
-      "mean_token_accuracy": 0.7527838827838828,
-      "mean_token_accuracy_utility": 0.6562820512820513,
+      "loss": 2.2889,
+      "mean_token_accuracy": 0.7564713064713063,
+      "mean_token_accuracy_utility": 0.6548901098901099,
       "step": 1955
     },
     {
       "epoch": 0.7253214913498011,
       "grad_norm": 0.0,
       "learning_rate": 4.255711183158635e-06,
-      "loss": 2.2064,
-      "mean_token_accuracy": 0.7626678876678875,
-      "mean_token_accuracy_utility": 0.6684126984126982,
+      "loss": 2.1939,
+      "mean_token_accuracy": 0.7663980463980465,
+      "mean_token_accuracy_utility": 0.6669719169719168,
       "step": 1960
     },
     {
       "epoch": 0.7271718012767139,
       "grad_norm": 0.0,
       "learning_rate": 4.2029403767172175e-06,
-      "loss": 2.2841,
-      "mean_token_accuracy": 0.7586874236874237,
-      "mean_token_accuracy_utility": 0.6576617826617824,
+      "loss": 2.271,
+      "mean_token_accuracy": 0.7627716727716728,
+      "mean_token_accuracy_utility": 0.6563125763125762,
       "step": 1965
     },
     {
       "epoch": 0.7290221112036266,
       "grad_norm": 0.0,
       "learning_rate": 4.150411604116531e-06,
-      "loss": 2.3032,
-      "mean_token_accuracy": 0.7559829059829057,
-      "mean_token_accuracy_utility": 0.654139194139194,
+      "loss": 2.2901,
+      "mean_token_accuracy": 0.7603785103785103,
+      "mean_token_accuracy_utility": 0.6528327228327226,
       "step": 1970
     },
     {
       "epoch": 0.7308724211305394,
       "grad_norm": 0.0,
       "learning_rate": 4.098127058492652e-06,
-      "loss": 2.1941,
-      "mean_token_accuracy": 0.7543711843711844,
-      "mean_token_accuracy_utility": 0.6729181929181928,
+      "loss": 2.1817,
+      "mean_token_accuracy": 0.7584493284493286,
+      "mean_token_accuracy_utility": 0.6719413919413917,
       "step": 1975
     },
     {
       "epoch": 0.7327227310574521,
       "grad_norm": 0.0,
       "learning_rate": 4.0460889227849e-06,
-      "loss": 2.27,
-      "mean_token_accuracy": 0.75519536019536,
-      "mean_token_accuracy_utility": 0.6611660561660561,
+      "loss": 2.2575,
+      "mean_token_accuracy": 0.7592002442002445,
+      "mean_token_accuracy_utility": 0.6598412698412698,
       "step": 1980
     },
     {
       "epoch": 0.7345730409843649,
       "grad_norm": 0.0,
       "learning_rate": 3.9942993696447045e-06,
-      "loss": 2.2834,
-      "mean_token_accuracy": 0.7605006105006105,
-      "mean_token_accuracy_utility": 0.6590964590964589,
+      "loss": 2.2715,
+      "mean_token_accuracy": 0.7642124542124544,
+      "mean_token_accuracy_utility": 0.658211233211233,
       "step": 1985
     },
     {
       "epoch": 0.7364233509112776,
       "grad_norm": 0.0,
       "learning_rate": 3.942760561344877e-06,
-      "loss": 2.1879,
-      "mean_token_accuracy": 0.7538034188034187,
-      "mean_token_accuracy_utility": 0.6765628815628814,
+      "loss": 2.1758,
+      "mean_token_accuracy": 0.7572405372405374,
+      "mean_token_accuracy_utility": 0.6750549450549449,
       "step": 1990
     },
     {
       "epoch": 0.7382736608381903,
       "grad_norm": 0.0,
       "learning_rate": 3.891474649689362e-06,
-      "loss": 2.2562,
-      "mean_token_accuracy": 0.7495482295482296,
-      "mean_token_accuracy_utility": 0.6667887667887668,
+      "loss": 2.243,
+      "mean_token_accuracy": 0.752899877899878,
+      "mean_token_accuracy_utility": 0.6652136752136751,
       "step": 1995
     },
     {
       "epoch": 0.7401239707651032,
       "grad_norm": 0.0,
       "learning_rate": 3.840443775923365e-06,
-      "loss": 2.1726,
-      "mean_token_accuracy": 0.7568070818070819,
-      "mean_token_accuracy_utility": 0.6756349206349205,
+      "loss": 2.1604,
+      "mean_token_accuracy": 0.7606532356532358,
+      "mean_token_accuracy_utility": 0.6746642246642245,
       "step": 2000
     },
     {
       "epoch": 0.7401239707651032,
-      "eval_mean_token_accuracy": 0.7475551681434033,
-      "eval_reasoning_loss": 0.8590500950813293,
-      "eval_reasoning_runtime": 27.5409,
-      "eval_reasoning_samples_per_second": 4.684,
-      "eval_reasoning_steps_per_second": 1.198,
+      "eval_mean_token_accuracy": 0.7514531220413573,
+      "eval_reasoning_loss": 0.8385465145111084,
+      "eval_reasoning_runtime": 27.6596,
+      "eval_reasoning_samples_per_second": 4.664,
+      "eval_reasoning_steps_per_second": 1.193,
       "step": 2000
     },
     {
       "epoch": 0.7401239707651032,
-      "eval_mean_token_accuracy": 0.661378914352892,
-      "eval_utility_loss": 1.4313037395477295,
-      "eval_utility_runtime": 10.368,
-      "eval_utility_samples_per_second": 4.726,
-      "eval_utility_steps_per_second": 1.254,
+      "eval_mean_token_accuracy": 0.6604758552856922,
+      "eval_utility_loss": 1.4363021850585938,
+      "eval_utility_runtime": 10.4113,
+      "eval_utility_samples_per_second": 4.706,
+      "eval_utility_steps_per_second": 1.249,
       "step": 2000
     },
     {
       "epoch": 0.7419742806920159,
       "grad_norm": 0.0,
       "learning_rate": 3.7896700706439826e-06,
-      "loss": 2.3341,
-      "mean_token_accuracy": 0.7579975579975582,
-      "mean_token_accuracy_utility": 0.6476421377846008,
+      "loss": 2.3223,
+      "mean_token_accuracy": 0.7618742368742369,
+      "mean_token_accuracy_utility": 0.6465399036217049,
       "step": 2005
     },
     {
       "epoch": 0.7438245906189287,
       "grad_norm": 0.0,
       "learning_rate": 3.7391556537112282e-06,
-      "loss": 2.1854,
-      "mean_token_accuracy": 0.7669780219780222,
-      "mean_token_accuracy_utility": 0.6649206349206348,
+      "loss": 2.1735,
+      "mean_token_accuracy": 0.7708974358974361,
+      "mean_token_accuracy_utility": 0.6634065934065932,
       "step": 2010
     },
     {
       "epoch": 0.7456749005458414,
       "grad_norm": 0.0,
       "learning_rate": 3.6889026341595378e-06,
-      "loss": 2.2504,
-      "mean_token_accuracy": 0.7700244200244202,
-      "mean_token_accuracy_utility": 0.6544139194139192,
+      "loss": 2.2387,
+      "mean_token_accuracy": 0.773870573870574,
+      "mean_token_accuracy_utility": 0.6529853479853478,
       "step": 2015
     },
     {
       "epoch": 0.7475252104727542,
       "grad_norm": 0.0,
       "learning_rate": 3.6389131101096953e-06,
-      "loss": 2.2469,
-      "mean_token_accuracy": 0.7617032967032966,
-      "mean_token_accuracy_utility": 0.658113553113553,
+      "loss": 2.2348,
+      "mean_token_accuracy": 0.7657326007326007,
+      "mean_token_accuracy_utility": 0.6564590964590963,
       "step": 2020
     },
     {
       "epoch": 0.7493755203996669,
       "grad_norm": 0.0,
       "learning_rate": 3.5891891686812597e-06,
-      "loss": 2.1994,
-      "mean_token_accuracy": 0.7508073642115167,
-      "mean_token_accuracy_utility": 0.6717467219259498,
+      "loss": 2.1868,
+      "mean_token_accuracy": 0.7548155792916779,
+      "mean_token_accuracy_utility": 0.6704507940996909,
       "step": 2025
     },
     {
       "epoch": 0.7512258303265797,
       "grad_norm": 0.0,
       "learning_rate": 3.5397328859054138e-06,
-      "loss": 2.1669,
-      "mean_token_accuracy": 0.7551648351648353,
-      "mean_token_accuracy_utility": 0.6808424908424907,
+      "loss": 2.1543,
+      "mean_token_accuracy": 0.7587484737484737,
+      "mean_token_accuracy_utility": 0.6796398046398044,
       "step": 2030
     },
     {
       "epoch": 0.7530761402534925,
       "grad_norm": 0.0,
       "learning_rate": 3.490546326638273e-06,
-      "loss": 2.2786,
-      "mean_token_accuracy": 0.7483993097957086,
-      "mean_token_accuracy_utility": 0.6652564102564101,
+      "loss": 2.2653,
+      "mean_token_accuracy": 0.7526783638310223,
+      "mean_token_accuracy_utility": 0.6643101343101342,
       "step": 2035
     },
     {
       "epoch": 0.7549264501804053,
       "grad_norm": 0.0,
       "learning_rate": 3.441631544474705e-06,
-      "loss": 2.2194,
-      "mean_token_accuracy": 0.7736385836385837,
-      "mean_token_accuracy_utility": 0.6592673992673992,
+      "loss": 2.2081,
+      "mean_token_accuracy": 0.7772832722832723,
+      "mean_token_accuracy_utility": 0.6575213675213674,
       "step": 2040
     },
     {
       "epoch": 0.756776760107318,
       "grad_norm": 0.0,
       "learning_rate": 3.3929905816625653e-06,
-      "loss": 2.3881,
-      "mean_token_accuracy": 0.7590964590964592,
-      "mean_token_accuracy_utility": 0.6314468864468863,
+      "loss": 2.3763,
+      "mean_token_accuracy": 0.7633211233211236,
+      "mean_token_accuracy_utility": 0.6302319902319901,
       "step": 2045
     },
     {
       "epoch": 0.7586270700342307,
       "grad_norm": 0.0,
       "learning_rate": 3.344625469017445e-06,
-      "loss": 2.202,
-      "mean_token_accuracy": 0.7543772893772893,
-      "mean_token_accuracy_utility": 0.6775030525030522,
+      "loss": 2.1898,
+      "mean_token_accuracy": 0.7579731379731379,
+      "mean_token_accuracy_utility": 0.6759645909645907,
       "step": 2050
     },
     {
       "epoch": 0.7604773799611435,
       "grad_norm": 0.0,
       "learning_rate": 3.2965382258378674e-06,
-      "loss": 2.1845,
-      "mean_token_accuracy": 0.7744871794871797,
-      "mean_token_accuracy_utility": 0.6637912087912086,
+      "loss": 2.1721,
+      "mean_token_accuracy": 0.7785103785103786,
+      "mean_token_accuracy_utility": 0.6625641025641024,
       "step": 2055
     },
     {
       "epoch": 0.7623276898880562,
       "grad_norm": 0.0,
       "learning_rate": 3.248730859821002e-06,
-      "loss": 2.2819,
-      "mean_token_accuracy": 0.7563431013431013,
-      "mean_token_accuracy_utility": 0.663217338217338,
+      "loss": 2.2699,
+      "mean_token_accuracy": 0.7607692307692309,
+      "mean_token_accuracy_utility": 0.6620085470085468,
       "step": 2060
     },
     {
       "epoch": 0.764177999814969,
       "grad_norm": 0.0,
       "learning_rate": 3.2012053669788136e-06,
-      "loss": 2.2619,
-      "mean_token_accuracy": 0.7541391941391942,
-      "mean_token_accuracy_utility": 0.6585470085470082,
+      "loss": 2.2495,
+      "mean_token_accuracy": 0.7579426129426129,
+      "mean_token_accuracy_utility": 0.6572039072039069,
       "step": 2065
     },
     {
       "epoch": 0.7660283097418817,
       "grad_norm": 0.0,
       "learning_rate": 3.1539637315547524e-06,
-      "loss": 2.1194,
-      "mean_token_accuracy": 0.7624603174603177,
-      "mean_token_accuracy_utility": 0.6819474969474968,
+      "loss": 2.1069,
+      "mean_token_accuracy": 0.7660012210012213,
+      "mean_token_accuracy_utility": 0.6805616605616603,
       "step": 2070
     },
     {
       "epoch": 0.7678786196687946,
       "grad_norm": 0.0,
       "learning_rate": 3.1070079259408934e-06,
-      "loss": 2.1984,
-      "mean_token_accuracy": 0.7521733821733821,
-      "mean_token_accuracy_utility": 0.6751526251526249,
+      "loss": 2.1867,
+      "mean_token_accuracy": 0.7556471306471308,
+      "mean_token_accuracy_utility": 0.6739010989010986,
       "step": 2075
     },
     {
       "epoch": 0.7697289295957073,
       "grad_norm": 0.0,
       "learning_rate": 3.0603399105955966e-06,
-      "loss": 2.338,
-      "mean_token_accuracy": 0.7580402930402931,
-      "mean_token_accuracy_utility": 0.6529059829059827,
+      "loss": 2.3257,
+      "mean_token_accuracy": 0.7617582417582418,
+      "mean_token_accuracy_utility": 0.6513736263736262,
       "step": 2080
     },
     {
       "epoch": 0.7715792395226201,
       "grad_norm": 0.0,
       "learning_rate": 3.0139616339616394e-06,
-      "loss": 2.233,
-      "mean_token_accuracy": 0.7611721611721614,
-      "mean_token_accuracy_utility": 0.6666213917786152,
+      "loss": 2.22,
+      "mean_token_accuracy": 0.7649572649572649,
+      "mean_token_accuracy_utility": 0.6650482473298005,
       "step": 2085
     },
     {
       "epoch": 0.7734295494495328,
       "grad_norm": 0.0,
       "learning_rate": 2.9678750323848893e-06,
-      "loss": 2.1814,
-      "mean_token_accuracy": 0.7634615384615384,
-      "mean_token_accuracy_utility": 0.6743650793650792,
+      "loss": 2.169,
+      "mean_token_accuracy": 0.7677167277167277,
+      "mean_token_accuracy_utility": 0.6730830280830279,
       "step": 2090
     },
     {
       "epoch": 0.7752798593764455,
       "grad_norm": 0.0,
       "learning_rate": 2.922082030033446e-06,
-      "loss": 2.2369,
-      "mean_token_accuracy": 0.7471672771672772,
-      "mean_token_accuracy_utility": 0.6692493103604213,
+      "loss": 2.2234,
+      "mean_token_accuracy": 0.7504761904761905,
+      "mean_token_accuracy_utility": 0.6680177405280283,
       "step": 2095
     },
     {
       "epoch": 0.7771301693033583,
       "grad_norm": 0.0,
       "learning_rate": 2.8765845388172955e-06,
-      "loss": 2.2547,
-      "mean_token_accuracy": 0.7675457875457876,
-      "mean_token_accuracy_utility": 0.6548230791702782,
+      "loss": 2.2427,
+      "mean_token_accuracy": 0.770970695970696,
+      "mean_token_accuracy_utility": 0.6537896965381821,
       "step": 2100
     },
     {
       "epoch": 0.7771301693033583,
-      "eval_mean_token_accuracy": 0.7474920504332272,
-      "eval_reasoning_loss": 0.8589840531349182,
-      "eval_reasoning_runtime": 27.5519,
-      "eval_reasoning_samples_per_second": 4.682,
-      "eval_reasoning_steps_per_second": 1.198,
+      "eval_mean_token_accuracy": 0.7513458219340573,
+      "eval_reasoning_loss": 0.838448703289032,
+      "eval_reasoning_runtime": 27.6142,
+      "eval_reasoning_samples_per_second": 4.672,
+      "eval_reasoning_steps_per_second": 1.195,
       "step": 2100
     },
     {
       "epoch": 0.7771301693033583,
-      "eval_mean_token_accuracy": 0.6613474378701718,
-      "eval_utility_loss": 1.4313603639602661,
-      "eval_utility_runtime": 10.3664,
-      "eval_utility_samples_per_second": 4.727,
-      "eval_utility_steps_per_second": 1.254,
+      "eval_mean_token_accuracy": 0.6604744586586165,
+      "eval_utility_loss": 1.436232328414917,
+      "eval_utility_runtime": 10.3927,
+      "eval_utility_samples_per_second": 4.715,
+      "eval_utility_steps_per_second": 1.251,
       "step": 2100
     },
     {
       "epoch": 0.778980479230271,
       "grad_norm": 0.0,
       "learning_rate": 2.831384458308518e-06,
-      "loss": 2.2054,
-      "mean_token_accuracy": 0.7729242979242981,
-      "mean_token_accuracy_utility": 0.6578815628815626,
+      "loss": 2.1938,
+      "mean_token_accuracy": 0.7767032967032967,
+      "mean_token_accuracy_utility": 0.6564407814407812,
       "step": 2105
     },
     {
       "epoch": 0.7808307891571838,
       "grad_norm": 0.0,
       "learning_rate": 2.7864836756619407e-06,
-      "loss": 2.2615,
-      "mean_token_accuracy": 0.7601709401709403,
-      "mean_token_accuracy_utility": 0.6551465201465201,
+      "loss": 2.2477,
+      "mean_token_accuracy": 0.7645909645909648,
+      "mean_token_accuracy_utility": 0.6538034188034187,
       "step": 2110
     },
     {
       "epoch": 0.7826810990840966,
       "grad_norm": 0.0,
       "learning_rate": 2.741884065536373e-06,
-      "loss": 2.2393,
-      "mean_token_accuracy": 0.7741941391941393,
-      "mean_token_accuracy_utility": 0.6535347985347983,
+      "loss": 2.2273,
+      "mean_token_accuracy": 0.7777960927960929,
+      "mean_token_accuracy_utility": 0.6516727716727716,
       "step": 2115
     },
     {
       "epoch": 0.7845314090110094,
       "grad_norm": 0.0,
       "learning_rate": 2.6975874900163223e-06,
-      "loss": 2.2799,
-      "mean_token_accuracy": 0.7590293040293041,
-      "mean_token_accuracy_utility": 0.6555189255189253,
+      "loss": 2.2669,
+      "mean_token_accuracy": 0.7637484737484737,
+      "mean_token_accuracy_utility": 0.654218559218559,
       "step": 2120
     },
     {
       "epoch": 0.7863817189379221,
       "grad_norm": 0.0,
       "learning_rate": 2.6535957985342653e-06,
-      "loss": 2.1835,
-      "mean_token_accuracy": 0.7661172161172163,
-      "mean_token_accuracy_utility": 0.6702930402930402,
+      "loss": 2.1713,
+      "mean_token_accuracy": 0.7696153846153848,
+      "mean_token_accuracy_utility": 0.6687728937728937,
       "step": 2125
     },
     {
       "epoch": 0.7882320288648349,
       "grad_norm": 0.0,
       "learning_rate": 2.6099108277934105e-06,
-      "loss": 2.1626,
-      "mean_token_accuracy": 0.7742429792429792,
-      "mean_token_accuracy_utility": 0.6661035364456687,
+      "loss": 2.15,
+      "mean_token_accuracy": 0.778070818070818,
+      "mean_token_accuracy_utility": 0.6651599308244653,
       "step": 2130
     },
     {
       "epoch": 0.7900823387917476,
       "grad_norm": 0.0,
       "learning_rate": 2.5665344016910367e-06,
-      "loss": 2.1077,
-      "mean_token_accuracy": 0.7662087912087913,
-      "mean_token_accuracy_utility": 0.6768925518925517,
+      "loss": 2.0957,
+      "mean_token_accuracy": 0.7703113553113552,
+      "mean_token_accuracy_utility": 0.6754700854700852,
       "step": 2135
     },
     {
       "epoch": 0.7919326487186604,
       "grad_norm": 0.0,
       "learning_rate": 2.523468331242329e-06,
-      "loss": 2.1723,
-      "mean_token_accuracy": 0.7587973137973141,
-      "mean_token_accuracy_utility": 0.6742857142857142,
+      "loss": 2.1606,
+      "mean_token_accuracy": 0.7622954822954824,
+      "mean_token_accuracy_utility": 0.6725274725274725,
       "step": 2140
     },
     {
       "epoch": 0.7937829586455731,
       "grad_norm": 0.0,
       "learning_rate": 2.4807144145047734e-06,
-      "loss": 2.2362,
-      "mean_token_accuracy": 0.7457509157509156,
-      "mean_token_accuracy_utility": 0.6713247863247862,
+      "loss": 2.2236,
+      "mean_token_accuracy": 0.7499633699633698,
+      "mean_token_accuracy_utility": 0.6697435897435895,
       "step": 2145
     },
     {
       "epoch": 0.7956332685724858,
       "grad_norm": 0.0,
       "learning_rate": 2.438274436503074e-06,
-      "loss": 2.2051,
-      "mean_token_accuracy": 0.7735653235653236,
-      "mean_token_accuracy_utility": 0.6603908661423581,
+      "loss": 2.1936,
+      "mean_token_accuracy": 0.777118437118437,
+      "mean_token_accuracy_utility": 0.6588198556131273,
       "step": 2150
     },
     {
       "epoch": 0.7974835784993987,
       "grad_norm": 0.0,
       "learning_rate": 2.396150169154644e-06,
-      "loss": 2.3296,
-      "mean_token_accuracy": 0.7524297924297924,
-      "mean_token_accuracy_utility": 0.6530463980463977,
+      "loss": 2.3166,
+      "mean_token_accuracy": 0.7567948717948718,
+      "mean_token_accuracy_utility": 0.6517399267399264,
       "step": 2155
     },
     {
       "epoch": 0.7993338884263114,
       "grad_norm": 0.0,
       "learning_rate": 2.3543433711956197e-06,
-      "loss": 2.2669,
-      "mean_token_accuracy": 0.7549755799755801,
-      "mean_token_accuracy_utility": 0.6606898656898654,
+      "loss": 2.2546,
+      "mean_token_accuracy": 0.7586080586080588,
+      "mean_token_accuracy_utility": 0.6592673992673991,
       "step": 2160
     },
     {
       "epoch": 0.8011841983532242,
       "grad_norm": 0.0,
       "learning_rate": 2.3128557881074153e-06,
-      "loss": 2.2481,
-      "mean_token_accuracy": 0.7650366300366301,
-      "mean_token_accuracy_utility": 0.6647985347985347,
+      "loss": 2.2364,
+      "mean_token_accuracy": 0.7686752136752137,
+      "mean_token_accuracy_utility": 0.6634126984126982,
       "step": 2165
     },
     {
       "epoch": 0.8030345082801369,
       "grad_norm": 0.0,
       "learning_rate": 2.271689152043873e-06,
-      "loss": 2.2634,
-      "mean_token_accuracy": 0.7574175824175825,
-      "mean_token_accuracy_utility": 0.6574420024420023,
+      "loss": 2.2506,
+      "mean_token_accuracy": 0.7613675213675215,
+      "mean_token_accuracy_utility": 0.6563308913308912,
       "step": 2170
     },
     {
       "epoch": 0.8048848182070497,
       "grad_norm": 0.0,
       "learning_rate": 2.230845181758928e-06,
-      "loss": 2.2602,
-      "mean_token_accuracy": 0.7417399267399267,
-      "mean_token_accuracy_utility": 0.6733638583638581,
+      "loss": 2.2466,
+      "mean_token_accuracy": 0.7455128205128206,
+      "mean_token_accuracy_utility": 0.6718437118437117,
       "step": 2175
     },
     {
       "epoch": 0.8067351281339624,
       "grad_norm": 0.0,
       "learning_rate": 2.1903255825348533e-06,
-      "loss": 2.3444,
-      "mean_token_accuracy": 0.7670207570207571,
-      "mean_token_accuracy_utility": 0.6419352869352868,
+      "loss": 2.3317,
+      "mean_token_accuracy": 0.7713858363858366,
+      "mean_token_accuracy_utility": 0.6410561660561659,
       "step": 2180
     },
     {
       "epoch": 0.8085854380608752,
       "grad_norm": 0.0,
       "learning_rate": 2.150132046111054e-06,
-      "loss": 2.1894,
-      "mean_token_accuracy": 0.7574664224664224,
-      "mean_token_accuracy_utility": 0.6730586080586078,
+      "loss": 2.1759,
+      "mean_token_accuracy": 0.7614896214896215,
+      "mean_token_accuracy_utility": 0.6716849816849815,
       "step": 2185
     },
     {
       "epoch": 0.810435747987788,
       "grad_norm": 0.0,
       "learning_rate": 2.1102662506134506e-06,
-      "loss": 2.3301,
-      "mean_token_accuracy": 0.7425579975579977,
-      "mean_token_accuracy_utility": 0.6548778998778998,
+      "loss": 2.316,
+      "mean_token_accuracy": 0.7466361416361418,
+      "mean_token_accuracy_utility": 0.6536752136752135,
       "step": 2190
     },
     {
       "epoch": 0.8122860579147008,
       "grad_norm": 0.0,
       "learning_rate": 2.0707298604843964e-06,
-      "loss": 2.2746,
-      "mean_token_accuracy": 0.7612148962148965,
-      "mean_token_accuracy_utility": 0.6563675213675211,
+      "loss": 2.2629,
+      "mean_token_accuracy": 0.76503663003663,
+      "mean_token_accuracy_utility": 0.6551221001220998,
       "step": 2195
     },
     {
       "epoch": 0.8141363678416135,
       "grad_norm": 0.0,
       "learning_rate": 2.03152452641321e-06,
-      "loss": 2.2226,
-      "mean_token_accuracy": 0.7720207570207571,
-      "mean_token_accuracy_utility": 0.6547680097680096,
+      "loss": 2.2103,
+      "mean_token_accuracy": 0.7753846153846154,
+      "mean_token_accuracy_utility": 0.6534737484737484,
       "step": 2200
     },
     {
       "epoch": 0.8141363678416135,
-      "eval_mean_token_accuracy": 0.7474273003684768,
-      "eval_reasoning_loss": 0.8590350151062012,
-      "eval_reasoning_runtime": 27.49,
-      "eval_reasoning_samples_per_second": 4.693,
-      "eval_reasoning_steps_per_second": 1.2,
+      "eval_mean_token_accuracy": 0.7514512720395075,
+      "eval_reasoning_loss": 0.8384374380111694,
+      "eval_reasoning_runtime": 27.5507,
+      "eval_reasoning_samples_per_second": 4.682,
+      "eval_reasoning_steps_per_second": 1.198,
       "step": 2200
     },
     {
       "epoch": 0.8141363678416135,
-      "eval_mean_token_accuracy": 0.6613615263457985,
-      "eval_utility_loss": 1.4312834739685059,
-      "eval_utility_runtime": 10.3472,
-      "eval_utility_samples_per_second": 4.736,
+      "eval_mean_token_accuracy": 0.6603194854267204,
+      "eval_utility_loss": 1.436269998550415,
+      "eval_utility_runtime": 10.352,
+      "eval_utility_samples_per_second": 4.733,
       "eval_utility_steps_per_second": 1.256,
       "step": 2200
     },
@@ -4368,1001 +4368,1001 @@
       "epoch": 0.8159866777685262,
       "grad_norm": 0.0,
       "learning_rate": 1.9926518852672294e-06,
-      "loss": 2.1782,
-      "mean_token_accuracy": 0.757936507936508,
-      "mean_token_accuracy_utility": 0.6738217338217336,
+      "loss": 2.1658,
+      "mean_token_accuracy": 0.7619780219780221,
+      "mean_token_accuracy_utility": 0.6722893772893771,
       "step": 2205
     },
     {
       "epoch": 0.817836987695439,
       "grad_norm": 0.0,
       "learning_rate": 1.9541135600234917e-06,
-      "loss": 2.192,
-      "mean_token_accuracy": 0.7684737484737487,
-      "mean_token_accuracy_utility": 0.6697680097680095,
+      "loss": 2.1799,
+      "mean_token_accuracy": 0.7719719169719171,
+      "mean_token_accuracy_utility": 0.6685592185592183,
       "step": 2210
     },
     {
       "epoch": 0.8196872976223517,
       "grad_norm": 0.0,
       "learning_rate": 1.9159111597009584e-06,
-      "loss": 2.2589,
-      "mean_token_accuracy": 0.766978021978022,
-      "mean_token_accuracy_utility": 0.6561294261294259,
+      "loss": 2.2464,
+      "mean_token_accuracy": 0.7711477411477411,
+      "mean_token_accuracy_utility": 0.6549816849816847,
       "step": 2215
     },
     {
       "epoch": 0.8215376075492645,
       "grad_norm": 0.0,
       "learning_rate": 1.8780462792933473e-06,
-      "loss": 2.2889,
-      "mean_token_accuracy": 0.7686263736263735,
-      "mean_token_accuracy_utility": 0.646025641025641,
+      "loss": 2.2768,
+      "mean_token_accuracy": 0.7726190476190478,
+      "mean_token_accuracy_utility": 0.6449267399267399,
       "step": 2220
     },
     {
       "epoch": 0.8233879174761772,
       "grad_norm": 0.0,
       "learning_rate": 1.8405204997025394e-06,
-      "loss": 2.3178,
-      "mean_token_accuracy": 0.738199023199023,
-      "mean_token_accuracy_utility": 0.6600976800976799,
+      "loss": 2.3045,
+      "mean_token_accuracy": 0.7419658119658119,
+      "mean_token_accuracy_utility": 0.6590354090354088,
       "step": 2225
     },
     {
       "epoch": 0.82523822740309,
       "grad_norm": 0.0,
       "learning_rate": 1.8033353876725578e-06,
-      "loss": 2.2631,
-      "mean_token_accuracy": 0.7426007326007324,
-      "mean_token_accuracy_utility": 0.6720146520146518,
+      "loss": 2.2501,
+      "mean_token_accuracy": 0.7465201465201465,
+      "mean_token_accuracy_utility": 0.6706043956043954,
       "step": 2230
     },
     {
       "epoch": 0.8270885373300028,
       "grad_norm": 0.0,
       "learning_rate": 1.766492495724178e-06,
-      "loss": 2.2543,
-      "mean_token_accuracy": 0.7656898656898659,
-      "mean_token_accuracy_utility": 0.6563003663003661,
+      "loss": 2.2426,
+      "mean_token_accuracy": 0.769847374847375,
+      "mean_token_accuracy_utility": 0.65525641025641,
       "step": 2235
     },
     {
       "epoch": 0.8289388472569156,
       "grad_norm": 0.0,
       "learning_rate": 1.7299933620900945e-06,
-      "loss": 2.2315,
-      "mean_token_accuracy": 0.7537545787545789,
-      "mean_token_accuracy_utility": 0.6672954822954822,
+      "loss": 2.2187,
+      "mean_token_accuracy": 0.7577777777777778,
+      "mean_token_accuracy_utility": 0.6660500610500609,
       "step": 2240
     },
     {
       "epoch": 0.8307891571838283,
       "grad_norm": 0.0,
       "learning_rate": 1.6938395106507034e-06,
-      "loss": 2.131,
-      "mean_token_accuracy": 0.7830891330891332,
-      "mean_token_accuracy_utility": 0.6687973137973136,
+      "loss": 2.1197,
+      "mean_token_accuracy": 0.7865934065934067,
+      "mean_token_accuracy_utility": 0.6671916971916969,
       "step": 2245
     },
     {
       "epoch": 0.832639467110741,
       "grad_norm": 0.0,
       "learning_rate": 1.658032450870467e-06,
-      "loss": 2.2559,
-      "mean_token_accuracy": 0.7608547008547009,
-      "mean_token_accuracy_utility": 0.6587484737484737,
+      "loss": 2.244,
+      "mean_token_accuracy": 0.7648046398046396,
+      "mean_token_accuracy_utility": 0.657112332112332,
       "step": 2250
     },
     {
       "epoch": 0.8344897770376538,
       "grad_norm": 0.0,
       "learning_rate": 1.622573677734911e-06,
-      "loss": 2.2029,
-      "mean_token_accuracy": 0.7712087912087913,
-      "mean_token_accuracy_utility": 0.6619536019536019,
+      "loss": 2.19,
+      "mean_token_accuracy": 0.7755372405372407,
+      "mean_token_accuracy_utility": 0.6608424908424907,
       "step": 2255
     },
     {
       "epoch": 0.8363400869645665,
       "grad_norm": 0.0,
       "learning_rate": 1.587464671688187e-06,
-      "loss": 2.1771,
-      "mean_token_accuracy": 0.7529670329670328,
-      "mean_token_accuracy_utility": 0.6767887667887666,
+      "loss": 2.1635,
+      "mean_token_accuracy": 0.756965811965812,
+      "mean_token_accuracy_utility": 0.6749572649572648,
       "step": 2260
     },
     {
       "epoch": 0.8381903968914793,
       "grad_norm": 0.0,
       "learning_rate": 1.552706898571288e-06,
-      "loss": 2.2341,
-      "mean_token_accuracy": 0.7503174603174603,
-      "mean_token_accuracy_utility": 0.668443520714151,
+      "loss": 2.2209,
+      "mean_token_accuracy": 0.7546520146520146,
+      "mean_token_accuracy_utility": 0.6667255982812106,
       "step": 2265
     },
     {
       "epoch": 0.840040706818392,
       "grad_norm": 0.0,
       "learning_rate": 1.5183018095608138e-06,
-      "loss": 2.2445,
-      "mean_token_accuracy": 0.7538209646325424,
-      "mean_token_accuracy_utility": 0.6655947050239731,
+      "loss": 2.2311,
+      "mean_token_accuracy": 0.7574781889821619,
+      "mean_token_accuracy_utility": 0.6640489413025996,
       "step": 2270
     },
     {
       "epoch": 0.8418910167453049,
       "grad_norm": 0.0,
       "learning_rate": 1.4842508411084145e-06,
-      "loss": 2.1278,
-      "mean_token_accuracy": 0.7685958485958487,
-      "mean_token_accuracy_utility": 0.684224664224664,
+      "loss": 2.1168,
+      "mean_token_accuracy": 0.7721306471306473,
+      "mean_token_accuracy_utility": 0.6826800976800975,
       "step": 2275
     },
     {
       "epoch": 0.8437413266722176,
       "grad_norm": 0.0,
       "learning_rate": 1.4505554148807954e-06,
-      "loss": 2.3875,
-      "mean_token_accuracy": 0.7509218559218558,
-      "mean_token_accuracy_utility": 0.6572527472527471,
+      "loss": 2.375,
+      "mean_token_accuracy": 0.7545421245421245,
+      "mean_token_accuracy_utility": 0.6558058608058607,
       "step": 2280
     },
     {
       "epoch": 0.8455916365991304,
       "grad_norm": 0.0,
       "learning_rate": 1.4172169377003775e-06,
-      "loss": 2.212,
-      "mean_token_accuracy": 0.7584676434676435,
-      "mean_token_accuracy_utility": 0.6760927960927958,
+      "loss": 2.199,
+      "mean_token_accuracy": 0.762881562881563,
+      "mean_token_accuracy_utility": 0.6749328449328448,
       "step": 2285
     },
     {
       "epoch": 0.8474419465260431,
       "grad_norm": 0.0,
       "learning_rate": 1.3842368014865414e-06,
-      "loss": 2.2215,
-      "mean_token_accuracy": 0.7637606837606838,
-      "mean_token_accuracy_utility": 0.6696030233271211,
+      "loss": 2.2081,
+      "mean_token_accuracy": 0.7678937728937729,
+      "mean_token_accuracy_utility": 0.6681550664473427,
       "step": 2290
     },
     {
       "epoch": 0.8492922564529559,
       "grad_norm": 0.0,
       "learning_rate": 1.3516163831975337e-06,
-      "loss": 2.2541,
-      "mean_token_accuracy": 0.7649328449328451,
-      "mean_token_accuracy_utility": 0.6585958485958484,
+      "loss": 2.2412,
+      "mean_token_accuracy": 0.768827838827839,
+      "mean_token_accuracy_utility": 0.6573260073260071,
       "step": 2295
     },
     {
       "epoch": 0.8511425663798686,
       "grad_norm": 0.0,
       "learning_rate": 1.3193570447729642e-06,
-      "loss": 2.2581,
-      "mean_token_accuracy": 0.752014652014652,
-      "mean_token_accuracy_utility": 0.6657995847462592,
+      "loss": 2.2458,
+      "mean_token_accuracy": 0.7554151404151404,
+      "mean_token_accuracy_utility": 0.6646015776897227,
       "step": 2300
     },
     {
       "epoch": 0.8511425663798686,
-      "eval_mean_token_accuracy": 0.7476345005756769,
-      "eval_reasoning_loss": 0.8590121269226074,
-      "eval_reasoning_runtime": 27.4637,
-      "eval_reasoning_samples_per_second": 4.697,
-      "eval_reasoning_steps_per_second": 1.202,
+      "eval_mean_token_accuracy": 0.7515049220931571,
+      "eval_reasoning_loss": 0.8383774757385254,
+      "eval_reasoning_runtime": 27.5457,
+      "eval_reasoning_samples_per_second": 4.683,
+      "eval_reasoning_steps_per_second": 1.198,
       "step": 2300
     },
     {
       "epoch": 0.8511425663798686,
-      "eval_mean_token_accuracy": 0.6616010304314567,
-      "eval_utility_loss": 1.431130051612854,
-      "eval_utility_runtime": 10.3417,
-      "eval_utility_samples_per_second": 4.738,
-      "eval_utility_steps_per_second": 1.257,
+      "eval_mean_token_accuracy": 0.6606261323590464,
+      "eval_utility_loss": 1.4362255334854126,
+      "eval_utility_runtime": 10.3545,
+      "eval_utility_samples_per_second": 4.732,
+      "eval_utility_steps_per_second": 1.255,
       "step": 2300
     },
     {
       "epoch": 0.8529928763067813,
       "grad_norm": 0.0,
       "learning_rate": 1.2874601330769488e-06,
-      "loss": 2.2803,
-      "mean_token_accuracy": 0.7492673992673994,
-      "mean_token_accuracy_utility": 0.66407218226556,
+      "loss": 2.2673,
+      "mean_token_accuracy": 0.7531440781440784,
+      "mean_token_accuracy_utility": 0.6629159038726491,
       "step": 2305
     },
     {
       "epoch": 0.8548431862336942,
       "grad_norm": 0.0,
       "learning_rate": 1.255926979841876e-06,
-      "loss": 2.2394,
-      "mean_token_accuracy": 0.7675946275946277,
-      "mean_token_accuracy_utility": 0.661037851037851,
+      "loss": 2.2268,
+      "mean_token_accuracy": 0.7713919413919413,
+      "mean_token_accuracy_utility": 0.6594505494505494,
       "step": 2310
     },
     {
       "epoch": 0.8566934961606069,
       "grad_norm": 0.0,
       "learning_rate": 1.224758901612796e-06,
-      "loss": 2.2841,
-      "mean_token_accuracy": 0.7635225885225887,
-      "mean_token_accuracy_utility": 0.6536324786324784,
+      "loss": 2.2717,
+      "mean_token_accuracy": 0.7672100122100123,
+      "mean_token_accuracy_utility": 0.652130647130647,
       "step": 2315
     },
     {
       "epoch": 0.8585438060875197,
       "grad_norm": 0.0,
       "learning_rate": 1.1939571996924738e-06,
-      "loss": 2.2386,
-      "mean_token_accuracy": 0.771166056166056,
-      "mean_token_accuracy_utility": 0.6588888888888887,
+      "loss": 2.2277,
+      "mean_token_accuracy": 0.77489010989011,
+      "mean_token_accuracy_utility": 0.6576739926739925,
       "step": 2320
     },
     {
       "epoch": 0.8603941160144324,
       "grad_norm": 0.0,
       "learning_rate": 1.1635231600870334e-06,
-      "loss": 2.3071,
-      "mean_token_accuracy": 0.7491575091575092,
-      "mean_token_accuracy_utility": 0.6523076923076921,
+      "loss": 2.2951,
+      "mean_token_accuracy": 0.7529487179487181,
+      "mean_token_accuracy_utility": 0.6508608058608056,
       "step": 2325
     },
     {
       "epoch": 0.8622444259413452,
       "grad_norm": 0.0,
       "learning_rate": 1.1334580534522932e-06,
-      "loss": 2.2239,
-      "mean_token_accuracy": 0.7583577533577532,
-      "mean_token_accuracy_utility": 0.6679059829059827,
+      "loss": 2.2132,
+      "mean_token_accuracy": 0.7619536019536017,
+      "mean_token_accuracy_utility": 0.6664896214896212,
       "step": 2330
     },
     {
       "epoch": 0.8640947358682579,
       "grad_norm": 0.0,
       "learning_rate": 1.1037631350406874e-06,
-      "loss": 2.2024,
-      "mean_token_accuracy": 0.7857692307692308,
-      "mean_token_accuracy_utility": 0.6520248640533108,
+      "loss": 2.1907,
+      "mean_token_accuracy": 0.7898595848595851,
+      "mean_token_accuracy_utility": 0.6502568937128135,
       "step": 2335
     },
     {
       "epoch": 0.8659450457951707,
       "grad_norm": 0.0,
       "learning_rate": 1.0744396446488781e-06,
-      "loss": 2.2747,
-      "mean_token_accuracy": 0.7555311355311355,
-      "mean_token_accuracy_utility": 0.6603357753357751,
+      "loss": 2.2629,
+      "mean_token_accuracy": 0.7592368742368744,
+      "mean_token_accuracy_utility": 0.6588461538461536,
       "step": 2340
     },
     {
       "epoch": 0.8677953557220834,
       "grad_norm": 0.0,
       "learning_rate": 1.0454888065659775e-06,
-      "loss": 2.2986,
-      "mean_token_accuracy": 0.7431257631257632,
-      "mean_token_accuracy_utility": 0.6590537240537239,
+      "loss": 2.2853,
+      "mean_token_accuracy": 0.7472832722832724,
+      "mean_token_accuracy_utility": 0.6577777777777777,
       "step": 2345
     },
     {
       "epoch": 0.8696456656489963,
       "grad_norm": 0.0,
       "learning_rate": 1.0169118295224488e-06,
-      "loss": 2.3764,
-      "mean_token_accuracy": 0.7488827838827838,
-      "mean_token_accuracy_utility": 0.6493772893772893,
+      "loss": 2.3642,
+      "mean_token_accuracy": 0.7524664224664223,
+      "mean_token_accuracy_utility": 0.6475946275946275,
       "step": 2350
     },
     {
       "epoch": 0.871495975575909,
       "grad_norm": 0.0,
       "learning_rate": 9.887099066396178e-07,
-      "loss": 2.2788,
-      "mean_token_accuracy": 0.7486691086691086,
-      "mean_token_accuracy_utility": 0.6612515262515262,
+      "loss": 2.2656,
+      "mean_token_accuracy": 0.7520634920634921,
+      "mean_token_accuracy_utility": 0.6595299145299144,
       "step": 2355
     },
     {
       "epoch": 0.8733462855028217,
       "grad_norm": 0.0,
       "learning_rate": 9.608842153798903e-07,
-      "loss": 2.3259,
-      "mean_token_accuracy": 0.7423260073260073,
-      "mean_token_accuracy_utility": 0.6634976299827783,
+      "loss": 2.3122,
+      "mean_token_accuracy": 0.7465934065934065,
+      "mean_token_accuracy_utility": 0.6625484361622974,
       "step": 2360
     },
     {
       "epoch": 0.8751965954297345,
       "grad_norm": 0.0,
       "learning_rate": 9.33435917497556e-07,
-      "loss": 2.2006,
-      "mean_token_accuracy": 0.7624603174603174,
-      "mean_token_accuracy_utility": 0.667155067155067,
+      "loss": 2.188,
+      "mean_token_accuracy": 0.766013431013431,
+      "mean_token_accuracy_utility": 0.6658241758241756,
       "step": 2365
     },
     {
       "epoch": 0.8770469053566472,
       "grad_norm": 0.0,
       "learning_rate": 9.063661589903116e-07,
-      "loss": 2.2085,
-      "mean_token_accuracy": 0.7600732600732601,
-      "mean_token_accuracy_utility": 0.6676881986436858,
+      "loss": 2.1966,
+      "mean_token_accuracy": 0.7640781440781442,
+      "mean_token_accuracy_utility": 0.6664105141042975,
       "step": 2370
     },
     {
       "epoch": 0.87889721528356,
       "grad_norm": 0.0,
       "learning_rate": 8.796760700513984e-07,
-      "loss": 2.2167,
-      "mean_token_accuracy": 0.7456471306471306,
-      "mean_token_accuracy_utility": 0.681039516039516,
+      "loss": 2.2025,
+      "mean_token_accuracy": 0.7495299145299145,
+      "mean_token_accuracy_utility": 0.6796600621600621,
       "step": 2375
     },
     {
       "epoch": 0.8807475252104727,
       "grad_norm": 0.0,
       "learning_rate": 8.533667650224253e-07,
-      "loss": 2.2588,
-      "mean_token_accuracy": 0.7599267399267401,
-      "mean_token_accuracy_utility": 0.6596703296703296,
+      "loss": 2.2465,
+      "mean_token_accuracy": 0.7635897435897436,
+      "mean_token_accuracy_utility": 0.6580280830280829,
       "step": 2380
     },
     {
       "epoch": 0.8825978351373855,
       "grad_norm": 0.0,
       "learning_rate": 8.274393423468385e-07,
-      "loss": 2.3484,
-      "mean_token_accuracy": 0.7440354090354091,
-      "mean_token_accuracy_utility": 0.657191697191697,
+      "loss": 2.3356,
+      "mean_token_accuracy": 0.748095238095238,
+      "mean_token_accuracy_utility": 0.6558730158730157,
       "step": 2385
     },
     {
       "epoch": 0.8844481450642983,
       "grad_norm": 0.0,
       "learning_rate": 8.018948845240538e-07,
-      "loss": 2.3152,
-      "mean_token_accuracy": 0.7528693528693526,
-      "mean_token_accuracy_utility": 0.6553968253968252,
+      "loss": 2.3017,
+      "mean_token_accuracy": 0.7569597069597067,
+      "mean_token_accuracy_utility": 0.654133089133089,
       "step": 2390
     },
     {
       "epoch": 0.8862984549912111,
       "grad_norm": 0.0,
       "learning_rate": 7.767344580642821e-07,
-      "loss": 2.3042,
-      "mean_token_accuracy": 0.7616605616605617,
-      "mean_token_accuracy_utility": 0.651092796092796,
+      "loss": 2.291,
+      "mean_token_accuracy": 0.7655433455433456,
+      "mean_token_accuracy_utility": 0.6499450549450548,
       "step": 2395
     },
     {
       "epoch": 0.8881487649181238,
       "grad_norm": 0.0,
       "learning_rate": 7.519591134439753e-07,
-      "loss": 2.2846,
-      "mean_token_accuracy": 0.7610134310134311,
-      "mean_token_accuracy_utility": 0.6564590964590963,
+      "loss": 2.2715,
+      "mean_token_accuracy": 0.7652564102564103,
+      "mean_token_accuracy_utility": 0.655286935286935,
       "step": 2400
     },
     {
       "epoch": 0.8881487649181238,
-      "eval_mean_token_accuracy": 0.74734383263795,
-      "eval_reasoning_loss": 0.8589988350868225,
-      "eval_reasoning_runtime": 27.5023,
-      "eval_reasoning_samples_per_second": 4.691,
-      "eval_reasoning_steps_per_second": 1.2,
+      "eval_mean_token_accuracy": 0.7515197221079571,
+      "eval_reasoning_loss": 0.8383514881134033,
+      "eval_reasoning_runtime": 27.5537,
+      "eval_reasoning_samples_per_second": 4.682,
+      "eval_reasoning_steps_per_second": 1.198,
       "step": 2400
     },
     {
       "epoch": 0.8881487649181238,
-      "eval_mean_token_accuracy": 0.6613333493945446,
-      "eval_utility_loss": 1.4312708377838135,
-      "eval_utility_runtime": 10.3586,
-      "eval_utility_samples_per_second": 4.73,
-      "eval_utility_steps_per_second": 1.255,
+      "eval_mean_token_accuracy": 0.6606261323590464,
+      "eval_utility_loss": 1.436227798461914,
+      "eval_utility_runtime": 10.3715,
+      "eval_utility_samples_per_second": 4.725,
+      "eval_utility_steps_per_second": 1.253,
       "step": 2400
     },
     {
       "epoch": 0.8899990748450366,
       "grad_norm": 0.0,
       "learning_rate": 7.275698850619861e-07,
-      "loss": 2.2777,
-      "mean_token_accuracy": 0.7624175824175824,
-      "mean_token_accuracy_utility": 0.6561660561660559,
+      "loss": 2.2662,
+      "mean_token_accuracy": 0.7663736263736263,
+      "mean_token_accuracy_utility": 0.6550793650793649,
       "step": 2405
     },
     {
       "epoch": 0.8918493847719493,
       "grad_norm": 0.0,
       "learning_rate": 7.035677911963712e-07,
-      "loss": 2.1718,
-      "mean_token_accuracy": 0.7709768009768009,
-      "mean_token_accuracy_utility": 0.6705555555555553,
+      "loss": 2.1593,
+      "mean_token_accuracy": 0.7745421245421246,
+      "mean_token_accuracy_utility": 0.6688400488400486,
       "step": 2410
     },
     {
       "epoch": 0.893699694698862,
       "grad_norm": 0.0,
       "learning_rate": 6.799538339618838e-07,
-      "loss": 2.1764,
-      "mean_token_accuracy": 0.7537973137973137,
-      "mean_token_accuracy_utility": 0.678058608058608,
+      "loss": 2.1627,
+      "mean_token_accuracy": 0.7576923076923077,
+      "mean_token_accuracy_utility": 0.677118437118437,
       "step": 2415
     },
     {
       "epoch": 0.8955500046257748,
       "grad_norm": 0.0,
       "learning_rate": 6.567289992681258e-07,
-      "loss": 2.233,
-      "mean_token_accuracy": 0.7644200244200244,
-      "mean_token_accuracy_utility": 0.6586324786324785,
+      "loss": 2.2204,
+      "mean_token_accuracy": 0.767881562881563,
+      "mean_token_accuracy_utility": 0.6571672771672771,
       "step": 2420
     },
     {
       "epoch": 0.8974003145526875,
       "grad_norm": 0.0,
       "learning_rate": 6.33894256778399e-07,
-      "loss": 2.3099,
-      "mean_token_accuracy": 0.7542002442002442,
-      "mean_token_accuracy_utility": 0.6547924297924297,
+      "loss": 2.2972,
+      "mean_token_accuracy": 0.7575763125763126,
+      "mean_token_accuracy_utility": 0.6533943833943833,
       "step": 2425
     },
     {
       "epoch": 0.8992506244796004,
       "grad_norm": 0.0,
       "learning_rate": 6.114505598692011e-07,
-      "loss": 2.2867,
-      "mean_token_accuracy": 0.755726495726496,
-      "mean_token_accuracy_utility": 0.65521978021978,
+      "loss": 2.2734,
+      "mean_token_accuracy": 0.7596336996336998,
+      "mean_token_accuracy_utility": 0.654194139194139,
       "step": 2430
     },
     {
       "epoch": 0.9011009344065131,
       "grad_norm": 0.0,
       "learning_rate": 5.893988455904387e-07,
-      "loss": 2.1916,
-      "mean_token_accuracy": 0.7701343101343103,
-      "mean_token_accuracy_utility": 0.6636887519809577,
+      "loss": 2.1798,
+      "mean_token_accuracy": 0.7739316239316241,
+      "mean_token_accuracy_utility": 0.6626061927506404,
       "step": 2435
     },
     {
       "epoch": 0.9029512443334259,
       "grad_norm": 0.0,
       "learning_rate": 5.677400346262918e-07,
-      "loss": 2.2646,
-      "mean_token_accuracy": 0.7659035409035412,
-      "mean_token_accuracy_utility": 0.6555494505494502,
+      "loss": 2.2525,
+      "mean_token_accuracy": 0.7700061050061053,
+      "mean_token_accuracy_utility": 0.6543467643467641,
       "step": 2440
     },
     {
       "epoch": 0.9048015542603386,
       "grad_norm": 0.0,
       "learning_rate": 5.464750312567835e-07,
-      "loss": 2.1066,
-      "mean_token_accuracy": 0.7831623931623934,
-      "mean_token_accuracy_utility": 0.6738149564776139,
+      "loss": 2.0948,
+      "mean_token_accuracy": 0.7862148962148965,
+      "mean_token_accuracy_utility": 0.6724909769584453,
       "step": 2445
     },
     {
       "epoch": 0.9066518641872514,
       "grad_norm": 0.0,
       "learning_rate": 5.256047233200201e-07,
-      "loss": 2.2703,
-      "mean_token_accuracy": 0.7588156288156291,
-      "mean_token_accuracy_utility": 0.6577167277167275,
+      "loss": 2.2578,
+      "mean_token_accuracy": 0.7626923076923078,
+      "mean_token_accuracy_utility": 0.6564102564102562,
       "step": 2450
     },
     {
       "epoch": 0.9085021741141641,
       "grad_norm": 0.0,
       "learning_rate": 5.051299821751254e-07,
-      "loss": 2.1413,
-      "mean_token_accuracy": 0.7556959706959709,
-      "mean_token_accuracy_utility": 0.6848412698412696,
+      "loss": 2.1293,
+      "mean_token_accuracy": 0.758901098901099,
+      "mean_token_accuracy_utility": 0.6834920634920633,
       "step": 2455
     },
     {
       "epoch": 0.9103524840410768,
       "grad_norm": 0.0,
       "learning_rate": 4.850516626658585e-07,
-      "loss": 2.3256,
-      "mean_token_accuracy": 0.7471306471306473,
-      "mean_token_accuracy_utility": 0.6612454212454211,
+      "loss": 2.3126,
+      "mean_token_accuracy": 0.7509279609279611,
+      "mean_token_accuracy_utility": 0.6598046398046397,
       "step": 2460
     },
     {
       "epoch": 0.9122027939679896,
       "grad_norm": 0.0,
       "learning_rate": 4.653706030849214e-07,
-      "loss": 2.2368,
-      "mean_token_accuracy": 0.7588949938949942,
-      "mean_token_accuracy_utility": 0.6650854700854698,
+      "loss": 2.225,
+      "mean_token_accuracy": 0.7622283272283275,
+      "mean_token_accuracy_utility": 0.6633333333333331,
       "step": 2465
     },
     {
       "epoch": 0.9140531038949024,
       "grad_norm": 0.0,
       "learning_rate": 4.4608762513896455e-07,
-      "loss": 2.2771,
-      "mean_token_accuracy": 0.742020757020757,
-      "mean_token_accuracy_utility": 0.6686324786324784,
+      "loss": 2.2635,
+      "mean_token_accuracy": 0.7457142857142858,
+      "mean_token_accuracy_utility": 0.6671184371184369,
       "step": 2470
     },
     {
       "epoch": 0.9159034138218152,
       "grad_norm": 0.0,
       "learning_rate": 4.2720353391427547e-07,
-      "loss": 2.1939,
-      "mean_token_accuracy": 0.7641391941391943,
-      "mean_token_accuracy_utility": 0.6632513429775092,
+      "loss": 2.182,
+      "mean_token_accuracy": 0.7681990231990234,
+      "mean_token_accuracy_utility": 0.6618918831900574,
       "step": 2475
     },
     {
       "epoch": 0.9177537237487279,
       "grad_norm": 0.0,
       "learning_rate": 4.087191178431682e-07,
-      "loss": 2.1912,
-      "mean_token_accuracy": 0.7540659340659343,
-      "mean_token_accuracy_utility": 0.6736630036630036,
+      "loss": 2.1785,
+      "mean_token_accuracy": 0.7576800976800977,
+      "mean_token_accuracy_utility": 0.6727899877899878,
       "step": 2480
     },
     {
       "epoch": 0.9196040336756407,
       "grad_norm": 0.0,
       "learning_rate": 3.9063514867105914e-07,
-      "loss": 2.3445,
-      "mean_token_accuracy": 0.7368376068376066,
-      "mean_token_accuracy_utility": 0.6629609279609279,
+      "loss": 2.3308,
+      "mean_token_accuracy": 0.7410500610500609,
+      "mean_token_accuracy_utility": 0.661269841269841,
       "step": 2485
     },
     {
       "epoch": 0.9214543436025534,
       "grad_norm": 0.0,
       "learning_rate": 3.729523814242608e-07,
-      "loss": 2.262,
-      "mean_token_accuracy": 0.7564041514041511,
-      "mean_token_accuracy_utility": 0.6614713064713063,
+      "loss": 2.2497,
+      "mean_token_accuracy": 0.7605738705738705,
+      "mean_token_accuracy_utility": 0.66019536019536,
       "step": 2490
     },
     {
       "epoch": 0.9233046535294662,
       "grad_norm": 0.0,
       "learning_rate": 3.5567155437843725e-07,
-      "loss": 2.1515,
-      "mean_token_accuracy": 0.7656898656898659,
-      "mean_token_accuracy_utility": 0.6756776556776556,
+      "loss": 2.14,
+      "mean_token_accuracy": 0.7691391941391943,
+      "mean_token_accuracy_utility": 0.6744383394383393,
       "step": 2495
     },
     {
       "epoch": 0.9251549634563789,
       "grad_norm": 0.0,
       "learning_rate": 3.3879338902779945e-07,
-      "loss": 2.215,
-      "mean_token_accuracy": 0.7751648351648354,
-      "mean_token_accuracy_utility": 0.6508974358974358,
+      "loss": 2.2027,
+      "mean_token_accuracy": 0.7796275946275946,
+      "mean_token_accuracy_utility": 0.6494993894993893,
       "step": 2500
     },
     {
       "epoch": 0.9251549634563789,
-      "eval_mean_token_accuracy": 0.7475142504554273,
-      "eval_reasoning_loss": 0.8589881062507629,
-      "eval_reasoning_runtime": 27.4961,
-      "eval_reasoning_samples_per_second": 4.692,
-      "eval_reasoning_steps_per_second": 1.2,
+      "eval_mean_token_accuracy": 0.7515363721246077,
+      "eval_reasoning_loss": 0.8384270668029785,
+      "eval_reasoning_runtime": 27.5441,
+      "eval_reasoning_samples_per_second": 4.683,
+      "eval_reasoning_steps_per_second": 1.198,
       "step": 2500
     },
     {
       "epoch": 0.9251549634563789,
-      "eval_mean_token_accuracy": 0.661474234150814,
-      "eval_utility_loss": 1.4311845302581787,
-      "eval_utility_runtime": 10.3603,
-      "eval_utility_samples_per_second": 4.73,
-      "eval_utility_steps_per_second": 1.255,
+      "eval_mean_token_accuracy": 0.6602570387385944,
+      "eval_utility_loss": 1.4361841678619385,
+      "eval_utility_runtime": 10.3693,
+      "eval_utility_samples_per_second": 4.725,
+      "eval_utility_steps_per_second": 1.254,
       "step": 2500
     },
     {
       "epoch": 0.9270052733832918,
       "grad_norm": 0.0,
       "learning_rate": 3.223185900549686e-07,
-      "loss": 2.1835,
-      "mean_token_accuracy": 0.7541391941391942,
-      "mean_token_accuracy_utility": 0.679249084249084,
+      "loss": 2.1708,
+      "mean_token_accuracy": 0.7577655677655678,
+      "mean_token_accuracy_utility": 0.6778388278388277,
       "step": 2505
     },
     {
       "epoch": 0.9288555833102045,
       "grad_norm": 0.0,
       "learning_rate": 3.0624784530156384e-07,
-      "loss": 2.309,
-      "mean_token_accuracy": 0.7561660561660561,
-      "mean_token_accuracy_utility": 0.6559951159951158,
+      "loss": 2.2965,
+      "mean_token_accuracy": 0.7603846153846154,
+      "mean_token_accuracy_utility": 0.6545909645909644,
       "step": 2510
     },
     {
       "epoch": 0.9307058932371172,
       "grad_norm": 0.0,
       "learning_rate": 2.905818257394799e-07,
-      "loss": 2.0799,
-      "mean_token_accuracy": 0.7727228327228328,
-      "mean_token_accuracy_utility": 0.687148962148962,
+      "loss": 2.0689,
+      "mean_token_accuracy": 0.7767277167277167,
+      "mean_token_accuracy_utility": 0.6859340659340659,
       "step": 2515
     },
     {
       "epoch": 0.93255620316403,
       "grad_norm": 0.0,
       "learning_rate": 2.753211854428728e-07,
-      "loss": 2.1901,
-      "mean_token_accuracy": 0.7708730158730159,
-      "mean_token_accuracy_utility": 0.6615079365079363,
+      "loss": 2.1792,
+      "mean_token_accuracy": 0.7752686202686204,
+      "mean_token_accuracy_utility": 0.6598473748473748,
       "step": 2520
     },
     {
       "epoch": 0.9344065130909427,
       "grad_norm": 0.0,
       "learning_rate": 2.604665615608526e-07,
-      "loss": 2.3398,
-      "mean_token_accuracy": 0.7572297299451345,
-      "mean_token_accuracy_utility": 0.6472329759734482,
+      "loss": 2.3272,
+      "mean_token_accuracy": 0.7612085999292265,
+      "mean_token_accuracy_utility": 0.6458536725467875,
       "step": 2525
     },
     {
       "epoch": 0.9362568230178555,
       "grad_norm": 0.0,
       "learning_rate": 2.460185742908816e-07,
-      "loss": 2.2313,
-      "mean_token_accuracy": 0.746098901098901,
-      "mean_token_accuracy_utility": 0.6714041514041511,
+      "loss": 2.218,
+      "mean_token_accuracy": 0.7500488400488401,
+      "mean_token_accuracy_utility": 0.6699877899877897,
       "step": 2530
     },
     {
       "epoch": 0.9381071329447682,
       "grad_norm": 0.0,
       "learning_rate": 2.3197782685288385e-07,
-      "loss": 2.3077,
-      "mean_token_accuracy": 0.7532112332112332,
-      "mean_token_accuracy_utility": 0.6538705738705737,
+      "loss": 2.2951,
+      "mean_token_accuracy": 0.7575213675213676,
+      "mean_token_accuracy_utility": 0.6526007326007324,
       "step": 2535
     },
     {
       "epoch": 0.939957442871681,
       "grad_norm": 0.0,
       "learning_rate": 2.1834490546405186e-07,
-      "loss": 2.2332,
-      "mean_token_accuracy": 0.7576434676434676,
-      "mean_token_accuracy_utility": 0.6668347173085559,
+      "loss": 2.2197,
+      "mean_token_accuracy": 0.7617216117216117,
+      "mean_token_accuracy_utility": 0.6658633009927722,
       "step": 2540
     },
     {
       "epoch": 0.9418077527985937,
       "grad_norm": 0.0,
       "learning_rate": 2.0512037931437855e-07,
-      "loss": 2.3183,
-      "mean_token_accuracy": 0.7500427350427351,
-      "mean_token_accuracy_utility": 0.6551159951159949,
+      "loss": 2.3048,
+      "mean_token_accuracy": 0.7536752136752136,
+      "mean_token_accuracy_utility": 0.6537240537240535,
       "step": 2545
     },
     {
       "epoch": 0.9436580627255066,
       "grad_norm": 0.0,
       "learning_rate": 1.9230480054288958e-07,
-      "loss": 2.1965,
-      "mean_token_accuracy": 0.7747741147741148,
-      "mean_token_accuracy_utility": 0.661214896214896,
+      "loss": 2.184,
+      "mean_token_accuracy": 0.7790964590964593,
+      "mean_token_accuracy_utility": 0.66014652014652,
       "step": 2550
     },
     {
       "epoch": 0.9455083726524193,
       "grad_norm": 0.0,
       "learning_rate": 1.7989870421459498e-07,
-      "loss": 2.1662,
-      "mean_token_accuracy": 0.7663675213675216,
-      "mean_token_accuracy_utility": 0.6716889995616722,
+      "loss": 2.1537,
+      "mean_token_accuracy": 0.7702442002442004,
+      "mean_token_accuracy_utility": 0.6701373294538355,
       "step": 2555
     },
     {
       "epoch": 0.9473586825793321,
       "grad_norm": 0.0,
       "learning_rate": 1.6790260829814053e-07,
-      "loss": 2.1742,
-      "mean_token_accuracy": 0.7508730158730159,
-      "mean_token_accuracy_utility": 0.6846031746031744,
+      "loss": 2.1605,
+      "mean_token_accuracy": 0.7548351648351649,
+      "mean_token_accuracy_utility": 0.6829731379731379,
       "step": 2560
     },
     {
       "epoch": 0.9492089925062448,
       "grad_norm": 0.0,
       "learning_rate": 1.5631701364419492e-07,
-      "loss": 2.0628,
-      "mean_token_accuracy": 0.782875457875458,
-      "mean_token_accuracy_utility": 0.6822100122100119,
+      "loss": 2.0529,
+      "mean_token_accuracy": 0.7862332112332115,
+      "mean_token_accuracy_utility": 0.6807203907203905,
       "step": 2565
     },
     {
       "epoch": 0.9510593024331575,
       "grad_norm": 0.0,
       "learning_rate": 1.4514240396452438e-07,
-      "loss": 2.2927,
-      "mean_token_accuracy": 0.7660744810744811,
-      "mean_token_accuracy_utility": 0.651184371184371,
+      "loss": 2.2798,
+      "mean_token_accuracy": 0.77,
+      "mean_token_accuracy_utility": 0.6502747252747251,
       "step": 2570
     },
     {
       "epoch": 0.9529096123600703,
       "grad_norm": 0.0,
       "learning_rate": 1.3437924581181205e-07,
-      "loss": 2.1566,
-      "mean_token_accuracy": 0.7754761904761909,
-      "mean_token_accuracy_utility": 0.6654273504273502,
+      "loss": 2.1434,
+      "mean_token_accuracy": 0.7790354090354094,
+      "mean_token_accuracy_utility": 0.6643711843711843,
       "step": 2575
     },
     {
       "epoch": 0.954759922286983,
       "grad_norm": 0.0,
       "learning_rate": 1.2402798856016474e-07,
-      "loss": 2.2581,
-      "mean_token_accuracy": 0.7443833943833944,
-      "mean_token_accuracy_utility": 0.6729426129426128,
+      "loss": 2.2444,
+      "mean_token_accuracy": 0.7481623931623931,
+      "mean_token_accuracy_utility": 0.6715140415140415,
       "step": 2580
     },
     {
       "epoch": 0.9566102322138959,
       "grad_norm": 0.0,
       "learning_rate": 1.1408906438636236e-07,
-      "loss": 2.337,
-      "mean_token_accuracy": 0.7673870573870574,
-      "mean_token_accuracy_utility": 0.6448962148962148,
+      "loss": 2.3248,
+      "mean_token_accuracy": 0.7713675213675215,
+      "mean_token_accuracy_utility": 0.6438949938949938,
       "step": 2585
     },
     {
       "epoch": 0.9584605421408086,
       "grad_norm": 0.0,
       "learning_rate": 1.045628882518046e-07,
-      "loss": 2.1694,
-      "mean_token_accuracy": 0.7703113553113552,
-      "mean_token_accuracy_utility": 0.6684249084249082,
+      "loss": 2.1569,
+      "mean_token_accuracy": 0.7745482295482298,
+      "mean_token_accuracy_utility": 0.6671611721611719,
       "step": 2590
     },
     {
       "epoch": 0.9603108520677214,
       "grad_norm": 0.0,
       "learning_rate": 9.544985788519589e-08,
-      "loss": 2.2447,
-      "mean_token_accuracy": 0.7483211233211232,
-      "mean_token_accuracy_utility": 0.6771065037490775,
+      "loss": 2.2323,
+      "mean_token_accuracy": 0.7526190476190476,
+      "mean_token_accuracy_utility": 0.6755667815598754,
       "step": 2595
     },
     {
       "epoch": 0.9621611619946341,
       "grad_norm": 0.0,
       "learning_rate": 8.675035376593088e-08,
-      "loss": 2.2921,
-      "mean_token_accuracy": 0.7588949938949939,
-      "mean_token_accuracy_utility": 0.6524053724053722,
+      "loss": 2.2789,
+      "mean_token_accuracy": 0.7626007326007327,
+      "mean_token_accuracy_utility": 0.6508852258852257,
       "step": 2600
     },
     {
       "epoch": 0.9621611619946341,
-      "eval_mean_token_accuracy": 0.7476049005460773,
-      "eval_reasoning_loss": 0.8589947819709778,
-      "eval_reasoning_runtime": 27.5926,
-      "eval_reasoning_samples_per_second": 4.675,
-      "eval_reasoning_steps_per_second": 1.196,
+      "eval_mean_token_accuracy": 0.7515160221042569,
+      "eval_reasoning_loss": 0.8384010791778564,
+      "eval_reasoning_runtime": 27.6854,
+      "eval_reasoning_samples_per_second": 4.659,
+      "eval_reasoning_steps_per_second": 1.192,
       "step": 2600
     },
     {
       "epoch": 0.9621611619946341,
-      "eval_mean_token_accuracy": 0.6615211957362374,
-      "eval_utility_loss": 1.4311784505844116,
-      "eval_utility_runtime": 10.3825,
-      "eval_utility_samples_per_second": 4.719,
-      "eval_utility_steps_per_second": 1.252,
+      "eval_mean_token_accuracy": 0.6606322251446642,
+      "eval_utility_loss": 1.436171293258667,
+      "eval_utility_runtime": 10.4181,
+      "eval_utility_samples_per_second": 4.703,
+      "eval_utility_steps_per_second": 1.248,
       "step": 2600
     },
     {
       "epoch": 0.9640114719215469,
       "grad_norm": 0.0,
       "learning_rate": 7.846473910821162e-08,
-      "loss": 2.2126,
-      "mean_token_accuracy": 0.7542857142857144,
-      "mean_token_accuracy_utility": 0.6709719116523318,
+      "loss": 2.2001,
+      "mean_token_accuracy": 0.7579548229548231,
+      "mean_token_accuracy_utility": 0.6692661992316015,
       "step": 2605
     },
     {
       "epoch": 0.9658617818484596,
       "grad_norm": 0.0,
       "learning_rate": 7.059335984588634e-08,
-      "loss": 2.2293,
-      "mean_token_accuracy": 0.7772344322344323,
-      "mean_token_accuracy_utility": 0.6517973248175487,
+      "loss": 2.2181,
+      "mean_token_accuracy": 0.7810683760683763,
+      "mean_token_accuracy_utility": 0.6507873947695108,
       "step": 2610
     },
     {
       "epoch": 0.9677120917753724,
       "grad_norm": 0.0,
       "learning_rate": 6.313654461800322e-08,
-      "loss": 2.2632,
-      "mean_token_accuracy": 0.7418009768009768,
-      "mean_token_accuracy_utility": 0.6748412698412697,
+      "loss": 2.251,
+      "mean_token_accuracy": 0.7456898656898657,
+      "mean_token_accuracy_utility": 0.6733882783882782,
       "step": 2615
     },
     {
       "epoch": 0.9695624017022851,
       "grad_norm": 0.0,
       "learning_rate": 5.609460475509032e-08,
-      "loss": 2.1662,
-      "mean_token_accuracy": 0.7694566544566546,
-      "mean_token_accuracy_utility": 0.6647924297924296,
+      "loss": 2.1537,
+      "mean_token_accuracy": 0.7736080586080587,
+      "mean_token_accuracy_utility": 0.6632905982905981,
       "step": 2620
     },
     {
       "epoch": 0.9714127116291978,
       "grad_norm": 0.0,
       "learning_rate": 4.9467834266154756e-08,
-      "loss": 2.2222,
-      "mean_token_accuracy": 0.7502442002442002,
-      "mean_token_accuracy_utility": 0.6716544566544564,
+      "loss": 2.2108,
+      "mean_token_accuracy": 0.754017094017094,
+      "mean_token_accuracy_utility": 0.6703418803418801,
       "step": 2625
     },
     {
       "epoch": 0.9732630215561107,
       "grad_norm": 0.0,
       "learning_rate": 4.325650982641039e-08,
-      "loss": 2.3199,
-      "mean_token_accuracy": 0.7567765567765568,
-      "mean_token_accuracy_utility": 0.6587374794621453,
+      "loss": 2.3078,
+      "mean_token_accuracy": 0.7609218559218558,
+      "mean_token_accuracy_utility": 0.6574671250456794,
       "step": 2630
     },
     {
       "epoch": 0.9751133314830234,
       "grad_norm": 0.0,
       "learning_rate": 3.746089076572701e-08,
-      "loss": 2.3563,
-      "mean_token_accuracy": 0.7626739926739929,
-      "mean_token_accuracy_utility": 0.6384188034188033,
+      "loss": 2.3442,
+      "mean_token_accuracy": 0.766739926739927,
+      "mean_token_accuracy_utility": 0.6372039072039071,
       "step": 2635
     },
     {
       "epoch": 0.9769636414099362,
       "grad_norm": 0.0,
       "learning_rate": 3.208121905779904e-08,
-      "loss": 2.1862,
-      "mean_token_accuracy": 0.7735225885225886,
-      "mean_token_accuracy_utility": 0.6607509157509156,
+      "loss": 2.1745,
+      "mean_token_accuracy": 0.7771123321123322,
+      "mean_token_accuracy_utility": 0.6595909645909644,
       "step": 2640
     },
     {
       "epoch": 0.9788139513368489,
       "grad_norm": 0.0,
       "learning_rate": 2.711771931004692e-08,
-      "loss": 2.2765,
-      "mean_token_accuracy": 0.7633089133089133,
-      "mean_token_accuracy_utility": 0.6542612942612942,
+      "loss": 2.2634,
+      "mean_token_accuracy": 0.7669902319902319,
+      "mean_token_accuracy_utility": 0.6529670329670328,
       "step": 2645
     },
     {
       "epoch": 0.9806642612637617,
       "grad_norm": 0.0,
       "learning_rate": 2.257059875423795e-08,
-      "loss": 2.3083,
-      "mean_token_accuracy": 0.7415201465201466,
-      "mean_token_accuracy_utility": 0.6657142857142857,
+      "loss": 2.2937,
+      "mean_token_accuracy": 0.7456105006105007,
+      "mean_token_accuracy_utility": 0.6643833943833943,
       "step": 2650
     },
     {
       "epoch": 0.9825145711906744,
       "grad_norm": 0.0,
       "learning_rate": 1.8440047237832105e-08,
-      "loss": 2.0995,
-      "mean_token_accuracy": 0.7841514041514042,
-      "mean_token_accuracy_utility": 0.6675946275946275,
+      "loss": 2.0882,
+      "mean_token_accuracy": 0.7876068376068377,
+      "mean_token_accuracy_utility": 0.6661111111111111,
       "step": 2655
     },
     {
       "epoch": 0.9843648811175872,
       "grad_norm": 0.0,
       "learning_rate": 1.472623721606059e-08,
-      "loss": 2.1898,
-      "mean_token_accuracy": 0.76490231990232,
-      "mean_token_accuracy_utility": 0.6740293040293037,
+      "loss": 2.1786,
+      "mean_token_accuracy": 0.7686324786324787,
+      "mean_token_accuracy_utility": 0.6725763125763123,
       "step": 2660
     },
     {
       "epoch": 0.9862151910445,
       "grad_norm": 0.0,
       "learning_rate": 1.1429323744720499e-08,
-      "loss": 2.2446,
-      "mean_token_accuracy": 0.7598290598290598,
-      "mean_token_accuracy_utility": 0.6598717948717947,
+      "loss": 2.233,
+      "mean_token_accuracy": 0.7638095238095239,
+      "mean_token_accuracy_utility": 0.6582295482295482,
       "step": 2665
     },
     {
       "epoch": 0.9880655009714127,
       "grad_norm": 0.0,
       "learning_rate": 8.549444473702207e-09,
-      "loss": 2.3235,
-      "mean_token_accuracy": 0.7444749694749696,
-      "mean_token_accuracy_utility": 0.6563492063492062,
+      "loss": 2.3101,
+      "mean_token_accuracy": 0.7484188034188035,
+      "mean_token_accuracy_utility": 0.65506105006105,
       "step": 2670
     },
     {
       "epoch": 0.9899158108983255,
       "grad_norm": 0.0,
       "learning_rate": 6.086719641246186e-09,
-      "loss": 2.1646,
-      "mean_token_accuracy": 0.7752747252747253,
-      "mean_token_accuracy_utility": 0.6644749694749692,
+      "loss": 2.1535,
+      "mean_token_accuracy": 0.7785775335775337,
+      "mean_token_accuracy_utility": 0.6629914529914527,
       "step": 2675
     },
     {
       "epoch": 0.9917661208252382,
       "grad_norm": 0.0,
       "learning_rate": 4.041252068918145e-09,
-      "loss": 2.1928,
-      "mean_token_accuracy": 0.7727350427350428,
-      "mean_token_accuracy_utility": 0.6591758241758239,
+      "loss": 2.1805,
+      "mean_token_accuracy": 0.7762637362637363,
+      "mean_token_accuracy_utility": 0.6581379731379728,
       "step": 2680
     },
     {
       "epoch": 0.993616430752151,
       "grad_norm": 0.0,
       "learning_rate": 2.4131271573191172e-09,
-      "loss": 2.3666,
-      "mean_token_accuracy": 0.7641086691086694,
-      "mean_token_accuracy_utility": 0.6429426129426128,
+      "loss": 2.3548,
+      "mean_token_accuracy": 0.767692307692308,
+      "mean_token_accuracy_utility": 0.6416727716727715,
       "step": 2685
     },
     {
       "epoch": 0.9954667406790637,
       "grad_norm": 0.0,
       "learning_rate": 1.2024128825172121e-09,
-      "loss": 2.0871,
-      "mean_token_accuracy": 0.7637606837606837,
-      "mean_token_accuracy_utility": 0.6863186813186811,
+      "loss": 2.075,
+      "mean_token_accuracy": 0.767918192918193,
+      "mean_token_accuracy_utility": 0.685152625152625,
       "step": 2690
     },
     {
       "epoch": 0.9973170506059765,
       "grad_norm": 0.0,
       "learning_rate": 4.0915979321320967e-10,
-      "loss": 2.1549,
-      "mean_token_accuracy": 0.7558974358974357,
-      "mean_token_accuracy_utility": 0.6745909645909645,
+      "loss": 2.1423,
+      "mean_token_accuracy": 0.75997557997558,
+      "mean_token_accuracy_utility": 0.6734615384615383,
       "step": 2695
     },
     {
       "epoch": 0.9991673605328892,
       "grad_norm": 0.0,
       "learning_rate": 3.3401008625588706e-11,
-      "loss": 2.2829,
-      "mean_token_accuracy": 0.7637423687423688,
-      "mean_token_accuracy_utility": 0.6530219780219778,
+      "loss": 2.2701,
+      "mean_token_accuracy": 0.7675702075702077,
+      "mean_token_accuracy_utility": 0.6519780219780218,
       "step": 2700
     },
     {
       "epoch": 0.9991673605328892,
-      "eval_mean_token_accuracy": 0.7476273182155535,
-      "eval_reasoning_loss": 0.858944296836853,
-      "eval_reasoning_runtime": 27.6653,
-      "eval_reasoning_samples_per_second": 4.663,
-      "eval_reasoning_steps_per_second": 1.193,
+      "eval_mean_token_accuracy": 0.7514882720765073,
+      "eval_reasoning_loss": 0.838329017162323,
+      "eval_reasoning_runtime": 27.6149,
+      "eval_reasoning_samples_per_second": 4.671,
+      "eval_reasoning_steps_per_second": 1.195,
       "step": 2700
     },
     {
       "epoch": 0.9991673605328892,
-      "eval_mean_token_accuracy": 0.6615057106335344,
-      "eval_utility_loss": 1.4312485456466675,
-      "eval_utility_runtime": 10.4257,
-      "eval_utility_samples_per_second": 4.7,
-      "eval_utility_steps_per_second": 1.247,
+      "eval_mean_token_accuracy": 0.6606899756742479,
+      "eval_utility_loss": 1.4362167119979858,
+      "eval_utility_runtime": 10.386,
+      "eval_utility_samples_per_second": 4.718,
+      "eval_utility_steps_per_second": 1.252,
       "step": 2700
     },
     {
       "epoch": 0.9999074845036544,
-      "mean_token_accuracy": 0.7905525030525029,
-      "mean_token_accuracy_utility": 0.6711233211233212,
+      "mean_token_accuracy": 0.7934981684981683,
+      "mean_token_accuracy_utility": 0.6699175824175825,
       "step": 2702,
       "total_flos": 0.0,
-      "train_loss": 2.3145499420024835,
-      "train_runtime": 31780.9844,
-      "train_samples_per_second": 0.68,
+      "train_loss": 2.300789678476194,
+      "train_runtime": 31930.6254,
+      "train_samples_per_second": 0.677,
       "train_steps_per_second": 0.085
     }
   ],