diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,5390 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "eval_steps": 100,
+  "global_step": 2703,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.001849796522382538,
+      "grad_norm": 0.0,
+      "learning_rate": 3.690036900369004e-07,
+      "loss": 2.4349,
+      "mean_token_accuracy": 0.7332835853513895,
+      "mean_token_accuracy_utility": 0.6865681333092679,
+      "step": 5
+    },
+    {
+      "epoch": 0.003699593044765076,
+      "grad_norm": 0.0,
+      "learning_rate": 7.380073800738008e-07,
+      "loss": 2.4331,
+      "mean_token_accuracy": 0.7258153835744728,
+      "mean_token_accuracy_utility": 0.6907303636507772,
+      "step": 10
+    },
+    {
+      "epoch": 0.005549389567147614,
+      "grad_norm": 0.0,
+      "learning_rate": 1.1070110701107011e-06,
+      "loss": 2.4646,
+      "mean_token_accuracy": 0.7401440905332599,
+      "mean_token_accuracy_utility": 0.670546543714475,
+      "step": 15
+    },
+    {
+      "epoch": 0.007399186089530152,
+      "grad_norm": 0.0,
+      "learning_rate": 1.4760147601476015e-06,
+      "loss": 2.396,
+      "mean_token_accuracy": 0.7347886541958214,
+      "mean_token_accuracy_utility": 0.6847830005059126,
+      "step": 20
+    },
+    {
+      "epoch": 0.00924898261191269,
+      "grad_norm": 0.0,
+      "learning_rate": 1.845018450184502e-06,
+      "loss": 2.3695,
+      "mean_token_accuracy": 0.7200360891939331,
+      "mean_token_accuracy_utility": 0.6926472885627634,
+      "step": 25
+    },
+    {
+      "epoch": 0.011098779134295227,
+      "grad_norm": 0.0,
+      "learning_rate": 2.2140221402214023e-06,
+      "loss": 2.2555,
+      "mean_token_accuracy": 0.7203309933923006,
+      "mean_token_accuracy_utility": 0.7038649798960377,
+      "step": 30
+    },
+    {
+      "epoch": 0.012948575656677765,
+      "grad_norm": 0.0,
+      "learning_rate": 2.5830258302583027e-06,
+      "loss": 2.0561,
+      "mean_token_accuracy": 0.7507204680620536,
+      "mean_token_accuracy_utility": 0.70900154729327,
+      "step": 35
+    },
+    {
+      "epoch": 0.014798372179060304,
+      "grad_norm": 0.0,
+      "learning_rate": 2.952029520295203e-06,
+      "loss": 2.1267,
+      "mean_token_accuracy": 0.7433367348318429,
+      "mean_token_accuracy_utility": 0.69693243692621,
+      "step": 40
+    },
+    {
+      "epoch": 0.016648168701442843,
+      "grad_norm": 0.0,
+      "learning_rate": 3.3210332103321034e-06,
+      "loss": 2.0606,
+      "mean_token_accuracy": 0.7285045929614512,
+      "mean_token_accuracy_utility": 0.7184004904550563,
+      "step": 45
+    },
+    {
+      "epoch": 0.01849796522382538,
+      "grad_norm": 0.0,
+      "learning_rate": 3.690036900369004e-06,
+      "loss": 2.0363,
+      "mean_token_accuracy": 0.7367152973563061,
+      "mean_token_accuracy_utility": 0.7137676358288755,
+      "step": 50
+    },
+    {
+      "epoch": 0.020347761746207917,
+      "grad_norm": 0.0,
+      "learning_rate": 4.059040590405905e-06,
+      "loss": 1.899,
+      "mean_token_accuracy": 0.7547901296035205,
+      "mean_token_accuracy_utility": 0.7256265365913043,
+      "step": 55
+    },
+    {
+      "epoch": 0.022197558268590455,
+      "grad_norm": 0.0,
+      "learning_rate": 4.428044280442805e-06,
+      "loss": 2.0102,
+      "mean_token_accuracy": 0.7416289638212571,
+      "mean_token_accuracy_utility": 0.7101105334497977,
+      "step": 60
+    },
+    {
+      "epoch": 0.024047354790972992,
+      "grad_norm": 0.0,
+      "learning_rate": 4.797047970479705e-06,
+      "loss": 1.9676,
+      "mean_token_accuracy": 0.7400402713424542,
+      "mean_token_accuracy_utility": 0.716452120001519,
+      "step": 65
+    },
+    {
+      "epoch": 0.02589715131335553,
+      "grad_norm": 0.0,
+      "learning_rate": 5.166051660516605e-06,
+      "loss": 1.9416,
+      "mean_token_accuracy": 0.7373601892572861,
+      "mean_token_accuracy_utility": 0.7296506662645195,
+      "step": 70
+    },
+    {
+      "epoch": 0.02774694783573807,
+      "grad_norm": 0.0,
+      "learning_rate": 5.535055350553506e-06,
+      "loss": 1.9128,
+      "mean_token_accuracy": 0.7458992689810409,
+      "mean_token_accuracy_utility": 0.7256271496637314,
+      "step": 75
+    },
+    {
+      "epoch": 0.029596744358120607,
+      "grad_norm": 0.0,
+      "learning_rate": 5.904059040590406e-06,
+      "loss": 1.8322,
+      "mean_token_accuracy": 0.7577962317100831,
+      "mean_token_accuracy_utility": 0.7357267405466137,
+      "step": 80
+    },
+    {
+      "epoch": 0.031446540880503145,
+      "grad_norm": 0.0,
+      "learning_rate": 6.273062730627307e-06,
+      "loss": 1.8815,
+      "mean_token_accuracy": 0.7611881953048192,
+      "mean_token_accuracy_utility": 0.7178880317642897,
+      "step": 85
+    },
+    {
+      "epoch": 0.033296337402885685,
+      "grad_norm": 0.0,
+      "learning_rate": 6.642066420664207e-06,
+      "loss": 1.8767,
+      "mean_token_accuracy": 0.7495895722443693,
+      "mean_token_accuracy_utility": 0.7277335310418096,
+      "step": 90
+    },
+    {
+      "epoch": 0.03514613392526822,
+      "grad_norm": 0.0,
+      "learning_rate": 7.011070110701108e-06,
+      "loss": 1.8694,
+      "mean_token_accuracy": 0.7617876841347911,
+      "mean_token_accuracy_utility": 0.722386133951006,
+      "step": 95
+    },
+    {
+      "epoch": 0.03699593044765076,
+      "grad_norm": 0.0,
+      "learning_rate": 7.380073800738008e-06,
+      "loss": 1.8551,
+      "mean_token_accuracy": 0.7528114650667405,
+      "mean_token_accuracy_utility": 0.7286875402728609,
+      "step": 100
+    },
+    {
+      "epoch": 0.03699593044765076,
+      "eval_mean_token_accuracy": 0.7464199963919508,
+      "eval_reasoning_loss": 0.87906414270401,
+      "eval_reasoning_runtime": 92.0013,
+      "eval_reasoning_samples_per_second": 1.402,
+      "eval_reasoning_steps_per_second": 0.359,
+      "step": 100
+    },
+    {
+      "epoch": 0.03699593044765076,
+      "eval_mean_token_accuracy": 0.7275092723743046,
+      "eval_utility_loss": 1.0093536376953125,
+      "eval_utility_runtime": 34.3103,
+      "eval_utility_samples_per_second": 1.428,
+      "eval_utility_steps_per_second": 0.379,
+      "step": 100
+    },
+    {
+      "epoch": 0.038845726970033294,
+      "grad_norm": 0.0,
+      "learning_rate": 7.749077490774908e-06,
+      "loss": 1.75,
+      "mean_token_accuracy": 0.7621413399335756,
+      "mean_token_accuracy_utility": 0.7418923388416755,
+      "step": 105
+    },
+    {
+      "epoch": 0.040695523492415835,
+      "grad_norm": 0.0,
+      "learning_rate": 8.11808118081181e-06,
+      "loss": 1.7699,
+      "mean_token_accuracy": 0.7699213264317597,
+      "mean_token_accuracy_utility": 0.7282805047755609,
+      "step": 110
+    },
+    {
+      "epoch": 0.042545320014798375,
+      "grad_norm": 0.0,
+      "learning_rate": 8.48708487084871e-06,
+      "loss": 1.8513,
+      "mean_token_accuracy": 0.7632512691178958,
+      "mean_token_accuracy_utility": 0.718816848325339,
+      "step": 115
+    },
+    {
+      "epoch": 0.04439511653718091,
+      "grad_norm": 0.0,
+      "learning_rate": 8.85608856088561e-06,
+      "loss": 1.8541,
+      "mean_token_accuracy": 0.7551577122618713,
+      "mean_token_accuracy_utility": 0.7241721382622885,
+      "step": 120
+    },
+    {
+      "epoch": 0.04624491305956345,
+      "grad_norm": 0.0,
+      "learning_rate": 9.22509225092251e-06,
+      "loss": 1.87,
+      "mean_token_accuracy": 0.753375742122824,
+      "mean_token_accuracy_utility": 0.7229886629763008,
+      "step": 125
+    },
+    {
+      "epoch": 0.048094709581945984,
+      "grad_norm": 0.0,
+      "learning_rate": 9.59409594095941e-06,
+      "loss": 1.8433,
+      "mean_token_accuracy": 0.7582636728299,
+      "mean_token_accuracy_utility": 0.7202868610814527,
+      "step": 130
+    },
+    {
+      "epoch": 0.049944506104328525,
+      "grad_norm": 0.0,
+      "learning_rate": 9.963099630996312e-06,
+      "loss": 1.772,
+      "mean_token_accuracy": 0.7643527012585991,
+      "mean_token_accuracy_utility": 0.7303040124921532,
+      "step": 135
+    },
+    {
+      "epoch": 0.05179430262671106,
+      "grad_norm": 0.0,
+      "learning_rate": 1.033210332103321e-05,
+      "loss": 1.8774,
+      "mean_token_accuracy": 0.7444156676609961,
+      "mean_token_accuracy_utility": 0.7267372383507629,
+      "step": 140
+    },
+    {
+      "epoch": 0.0536440991490936,
+      "grad_norm": 0.0,
+      "learning_rate": 1.0701107011070112e-05,
+      "loss": 1.8279,
+      "mean_token_accuracy": 0.7618354682660459,
+      "mean_token_accuracy_utility": 0.7262385049980667,
+      "step": 145
+    },
+    {
+      "epoch": 0.05549389567147614,
+      "grad_norm": 0.0,
+      "learning_rate": 1.1070110701107012e-05,
+      "loss": 1.8193,
+      "mean_token_accuracy": 0.7520773415804773,
+      "mean_token_accuracy_utility": 0.7304241108699985,
+      "step": 150
+    },
+    {
+      "epoch": 0.057343692193858674,
+      "grad_norm": 0.0,
+      "learning_rate": 1.1439114391143913e-05,
+      "loss": 1.7042,
+      "mean_token_accuracy": 0.7766770050039395,
+      "mean_token_accuracy_utility": 0.7378825855449885,
+      "step": 155
+    },
+    {
+      "epoch": 0.059193488716241215,
+      "grad_norm": 0.0,
+      "learning_rate": 1.1808118081180812e-05,
+      "loss": 1.8161,
+      "mean_token_accuracy": 0.7638029995138023,
+      "mean_token_accuracy_utility": 0.7219715167595495,
+      "step": 160
+    },
+    {
+      "epoch": 0.06104328523862375,
+      "grad_norm": 0.0,
+      "learning_rate": 1.2177121771217713e-05,
+      "loss": 1.8094,
+      "mean_token_accuracy": 0.7625361709877503,
+      "mean_token_accuracy_utility": 0.7267580136969152,
+      "step": 165
+    },
+    {
+      "epoch": 0.06289308176100629,
+      "grad_norm": 0.0,
+      "learning_rate": 1.2546125461254614e-05,
+      "loss": 1.7574,
+      "mean_token_accuracy": 0.7736754920758429,
+      "mean_token_accuracy_utility": 0.7288909492270748,
+      "step": 170
+    },
+    {
+      "epoch": 0.06474287828338883,
+      "grad_norm": 0.0,
+      "learning_rate": 1.2915129151291515e-05,
+      "loss": 1.7386,
+      "mean_token_accuracy": 0.7652630034435517,
+      "mean_token_accuracy_utility": 0.7337469098486531,
+      "step": 175
+    },
+    {
+      "epoch": 0.06659267480577137,
+      "grad_norm": 0.0,
+      "learning_rate": 1.3284132841328414e-05,
+      "loss": 1.779,
+      "mean_token_accuracy": 0.7630636797209405,
+      "mean_token_accuracy_utility": 0.726525004886766,
+      "step": 180
+    },
+    {
+      "epoch": 0.0684424713281539,
+      "grad_norm": 0.0,
+      "learning_rate": 1.3653136531365315e-05,
+      "loss": 1.6532,
+      "mean_token_accuracy": 0.7817794394003582,
+      "mean_token_accuracy_utility": 0.7410552657612379,
+      "step": 185
+    },
+    {
+      "epoch": 0.07029226785053644,
+      "grad_norm": 0.0,
+      "learning_rate": 1.4022140221402215e-05,
+      "loss": 1.7009,
+      "mean_token_accuracy": 0.781692523101617,
+      "mean_token_accuracy_utility": 0.7312295006322731,
+      "step": 190
+    },
+    {
+      "epoch": 0.07214206437291898,
+      "grad_norm": 0.0,
+      "learning_rate": 1.4391143911439116e-05,
+      "loss": 1.7765,
+      "mean_token_accuracy": 0.7668051067253161,
+      "mean_token_accuracy_utility": 0.7256368550289768,
+      "step": 195
+    },
+    {
+      "epoch": 0.07399186089530152,
+      "grad_norm": 0.0,
+      "learning_rate": 1.4760147601476015e-05,
+      "loss": 1.7252,
+      "mean_token_accuracy": 0.7707869655408566,
+      "mean_token_accuracy_utility": 0.7297637144899167,
+      "step": 200
+    },
+    {
+      "epoch": 0.07399186089530152,
+      "eval_mean_token_accuracy": 0.7644279025583934,
+      "eval_reasoning_loss": 0.7819119095802307,
+      "eval_reasoning_runtime": 90.1921,
+      "eval_reasoning_samples_per_second": 1.43,
+      "eval_reasoning_steps_per_second": 0.366,
+      "step": 200
+    },
+    {
+      "epoch": 0.07399186089530152,
+      "eval_mean_token_accuracy": 0.7320154745923281,
+      "eval_utility_loss": 0.9823516011238098,
+      "eval_utility_runtime": 33.9575,
+      "eval_utility_samples_per_second": 1.443,
+      "eval_utility_steps_per_second": 0.383,
+      "step": 200
+    },
+    {
+      "epoch": 0.07584165741768406,
+      "grad_norm": 0.0,
+      "learning_rate": 1.5129151291512916e-05,
+      "loss": 1.7535,
+      "mean_token_accuracy": 0.7592969246206909,
+      "mean_token_accuracy_utility": 0.7331951698230642,
+      "step": 205
+    },
+    {
+      "epoch": 0.07769145394006659,
+      "grad_norm": 0.0,
+      "learning_rate": 1.5498154981549817e-05,
+      "loss": 1.8305,
+      "mean_token_accuracy": 0.7666876408794361,
+      "mean_token_accuracy_utility": 0.7167117781964942,
+      "step": 210
+    },
+    {
+      "epoch": 0.07954125046244913,
+      "grad_norm": 0.0,
+      "learning_rate": 1.5867158671586716e-05,
+      "loss": 1.6346,
+      "mean_token_accuracy": 0.777809904775085,
+      "mean_token_accuracy_utility": 0.7445735235834552,
+      "step": 215
+    },
+    {
+      "epoch": 0.08139104698483167,
+      "grad_norm": 0.0,
+      "learning_rate": 1.623616236162362e-05,
+      "loss": 1.7467,
+      "mean_token_accuracy": 0.7746032934053209,
+      "mean_token_accuracy_utility": 0.7241276979062945,
+      "step": 220
+    },
+    {
+      "epoch": 0.08324084350721421,
+      "grad_norm": 0.0,
+      "learning_rate": 1.6605166051660518e-05,
+      "loss": 1.731,
+      "mean_token_accuracy": 0.7747848419258958,
+      "mean_token_accuracy_utility": 0.7260973901975303,
+      "step": 225
+    },
+    {
+      "epoch": 0.08509064002959675,
+      "grad_norm": 0.0,
+      "learning_rate": 1.697416974169742e-05,
+      "loss": 1.7542,
+      "mean_token_accuracy": 0.7660526373603214,
+      "mean_token_accuracy_utility": 0.7297221536413332,
+      "step": 230
+    },
+    {
+      "epoch": 0.08694043655197928,
+      "grad_norm": 0.0,
+      "learning_rate": 1.734317343173432e-05,
+      "loss": 1.7122,
+      "mean_token_accuracy": 0.7843387257202252,
+      "mean_token_accuracy_utility": 0.7278456785917015,
+      "step": 235
+    },
+    {
+      "epoch": 0.08879023307436182,
+      "grad_norm": 0.0,
+      "learning_rate": 1.771217712177122e-05,
+      "loss": 1.7134,
+      "mean_token_accuracy": 0.7704482082687314,
+      "mean_token_accuracy_utility": 0.7336378823223254,
+      "step": 240
+    },
+    {
+      "epoch": 0.09064002959674436,
+      "grad_norm": 0.0,
+      "learning_rate": 1.8081180811808117e-05,
+      "loss": 1.6643,
+      "mean_token_accuracy": 0.7772448929355129,
+      "mean_token_accuracy_utility": 0.7424889789990667,
+      "step": 245
+    },
+    {
+      "epoch": 0.0924898261191269,
+      "grad_norm": 0.0,
+      "learning_rate": 1.845018450184502e-05,
+      "loss": 1.7758,
+      "mean_token_accuracy": 0.7626401985644387,
+      "mean_token_accuracy_utility": 0.7325436707644473,
+      "step": 250
+    },
+    {
+      "epoch": 0.09433962264150944,
+      "grad_norm": 0.0,
+      "learning_rate": 1.8819188191881922e-05,
+      "loss": 1.6608,
+      "mean_token_accuracy": 0.7845206596731371,
+      "mean_token_accuracy_utility": 0.7361681094760504,
+      "step": 255
+    },
+    {
+      "epoch": 0.09618941916389197,
+      "grad_norm": 0.0,
+      "learning_rate": 1.918819188191882e-05,
+      "loss": 1.6714,
+      "mean_token_accuracy": 0.7787202543166603,
+      "mean_token_accuracy_utility": 0.7354817040464466,
+      "step": 260
+    },
+    {
+      "epoch": 0.09803921568627451,
+      "grad_norm": 0.0,
+      "learning_rate": 1.955719557195572e-05,
+      "loss": 1.7587,
+      "mean_token_accuracy": 0.7786917812906708,
+      "mean_token_accuracy_utility": 0.7216370857095453,
+      "step": 265
+    },
+    {
+      "epoch": 0.09988901220865705,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9926199261992623e-05,
+      "loss": 1.6561,
+      "mean_token_accuracy": 0.7982589536104924,
+      "mean_token_accuracy_utility": 0.7232248695454092,
+      "step": 270
+    },
+    {
+      "epoch": 0.10173880873103959,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9999866506037346e-05,
+      "loss": 1.7448,
+      "mean_token_accuracy": 0.775553137263134,
+      "mean_token_accuracy_utility": 0.7283458472516793,
+      "step": 275
+    },
+    {
+      "epoch": 0.10358860525342212,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9999324192922497e-05,
+      "loss": 1.6801,
+      "mean_token_accuracy": 0.7916855237248414,
+      "mean_token_accuracy_utility": 0.7291901912455241,
+      "step": 280
+    },
+    {
+      "epoch": 0.10543840177580466,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9998364739888954e-05,
+      "loss": 1.6237,
+      "mean_token_accuracy": 0.7820422742150644,
+      "mean_token_accuracy_utility": 0.7431596434181016,
+      "step": 285
+    },
+    {
+      "epoch": 0.1072881982981872,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9996988186962044e-05,
+      "loss": 1.6253,
+      "mean_token_accuracy": 0.792486481005592,
+      "mean_token_accuracy_utility": 0.7349954695988097,
+      "step": 290
+    },
+    {
+      "epoch": 0.10913799482056974,
+      "grad_norm": 0.0,
+      "learning_rate": 1.999519459156716e-05,
+      "loss": 1.5918,
+      "mean_token_accuracy": 0.7874805555052626,
+      "mean_token_accuracy_utility": 0.7431603697545299,
+      "step": 295
+    },
+    {
+      "epoch": 0.11098779134295228,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9992984028527388e-05,
+      "loss": 1.6132,
+      "mean_token_accuracy": 0.785312437319819,
+      "mean_token_accuracy_utility": 0.742114719224453,
+      "step": 300
+    },
+    {
+      "epoch": 0.11098779134295228,
+      "eval_mean_token_accuracy": 0.7736581779170286,
+      "eval_reasoning_loss": 0.734140157699585,
+      "eval_reasoning_runtime": 91.032,
+      "eval_reasoning_samples_per_second": 1.417,
+      "eval_reasoning_steps_per_second": 0.363,
+      "step": 300
+    },
+    {
+      "epoch": 0.11098779134295228,
+      "eval_mean_token_accuracy": 0.7353442822796448,
+      "eval_utility_loss": 0.9670671820640564,
+      "eval_utility_runtime": 34.2734,
+      "eval_utility_samples_per_second": 1.43,
+      "eval_utility_steps_per_second": 0.379,
+      "step": 300
+    },
+    {
+      "epoch": 0.1128375878653348,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9990356590060363e-05,
+      "loss": 1.6869,
+      "mean_token_accuracy": 0.7788980021729541,
+      "mean_token_accuracy_utility": 0.7316531037214384,
+      "step": 305
+    },
+    {
+      "epoch": 0.11468738438771735,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9987312385774444e-05,
+      "loss": 1.574,
+      "mean_token_accuracy": 0.7989620378153496,
+      "mean_token_accuracy_utility": 0.7444030271760771,
+      "step": 310
+    },
+    {
+      "epoch": 0.11653718091009989,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9983851542664125e-05,
+      "loss": 1.6378,
+      "mean_token_accuracy": 0.787880240533928,
+      "mean_token_accuracy_utility": 0.7384136327699556,
+      "step": 315
+    },
+    {
+      "epoch": 0.11838697743248243,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9979974205104743e-05,
+      "loss": 1.6306,
+      "mean_token_accuracy": 0.78010953871451,
+      "mean_token_accuracy_utility": 0.7390810986786986,
+      "step": 320
+    },
+    {
+      "epoch": 0.12023677395486497,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9975680534846457e-05,
+      "loss": 1.6331,
+      "mean_token_accuracy": 0.7865092481734638,
+      "mean_token_accuracy_utility": 0.739441503765209,
+      "step": 325
+    },
+    {
+      "epoch": 0.1220865704772475,
+      "grad_norm": 0.0,
+      "learning_rate": 1.99709707110075e-05,
+      "loss": 1.6458,
+      "mean_token_accuracy": 0.7833926639358049,
+      "mean_token_accuracy_utility": 0.7342138906305375,
+      "step": 330
+    },
+    {
+      "epoch": 0.12393636699963004,
+      "grad_norm": 0.0,
+      "learning_rate": 1.99658449300667e-05,
+      "loss": 1.6236,
+      "mean_token_accuracy": 0.7767692554801876,
+      "mean_token_accuracy_utility": 0.7452420818349104,
+      "step": 335
+    },
+    {
+      "epoch": 0.12578616352201258,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9960303405855295e-05,
+      "loss": 1.6786,
+      "mean_token_accuracy": 0.7877326976833463,
+      "mean_token_accuracy_utility": 0.7285483115265973,
+      "step": 340
+    },
+    {
+      "epoch": 0.12763596004439512,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9954346369548002e-05,
+      "loss": 1.5627,
+      "mean_token_accuracy": 0.8040369432241796,
+      "mean_token_accuracy_utility": 0.7399751077244556,
+      "step": 345
+    },
+    {
+      "epoch": 0.12948575656677766,
+      "grad_norm": 0.0,
+      "learning_rate": 1.994797406965338e-05,
+      "loss": 1.6634,
+      "mean_token_accuracy": 0.7742994167439111,
+      "mean_token_accuracy_utility": 0.7395679914245183,
+      "step": 350
+    },
+    {
+      "epoch": 0.1313355530891602,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9941186772003463e-05,
+      "loss": 1.6516,
+      "mean_token_accuracy": 0.7699775683368658,
+      "mean_token_accuracy_utility": 0.7439692594518084,
+      "step": 355
+    },
+    {
+      "epoch": 0.13318534961154274,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9933984759742664e-05,
+      "loss": 1.5774,
+      "mean_token_accuracy": 0.796407951579247,
+      "mean_token_accuracy_utility": 0.7448314793193832,
+      "step": 360
+    },
+    {
+      "epoch": 0.13503514613392525,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9926368333315964e-05,
+      "loss": 1.6387,
+      "mean_token_accuracy": 0.795452631650292,
+      "mean_token_accuracy_utility": 0.7328011085110017,
+      "step": 365
+    },
+    {
+      "epoch": 0.1368849426563078,
+      "grad_norm": 0.0,
+      "learning_rate": 1.991833781045639e-05,
+      "loss": 1.5461,
+      "mean_token_accuracy": 0.7993687633650743,
+      "mean_token_accuracy_utility": 0.7467482120703459,
+      "step": 370
+    },
+    {
+      "epoch": 0.13873473917869034,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9909893526171745e-05,
+      "loss": 1.6278,
+      "mean_token_accuracy": 0.7942682076499757,
+      "mean_token_accuracy_utility": 0.7377687272814144,
+      "step": 375
+    },
+    {
+      "epoch": 0.14058453570107288,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9901035832730647e-05,
+      "loss": 1.6469,
+      "mean_token_accuracy": 0.7773208018185016,
+      "mean_token_accuracy_utility": 0.7459934316865239,
+      "step": 380
+    },
+    {
+      "epoch": 0.14243433222345542,
+      "grad_norm": 0.0,
+      "learning_rate": 1.989176509964781e-05,
+      "loss": 1.5439,
+      "mean_token_accuracy": 0.7960874070364053,
+      "mean_token_accuracy_utility": 0.7504608893621103,
+      "step": 385
+    },
+    {
+      "epoch": 0.14428412874583796,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9882081713668665e-05,
+      "loss": 1.7327,
+      "mean_token_accuracy": 0.7741965534045541,
+      "mean_token_accuracy_utility": 0.7330720668303171,
+      "step": 390
+    },
+    {
+      "epoch": 0.1461339252682205,
+      "grad_norm": 0.0,
+      "learning_rate": 1.987198607875319e-05,
+      "loss": 1.6383,
+      "mean_token_accuracy": 0.8041237159813519,
+      "mean_token_accuracy_utility": 0.7349281644158017,
+      "step": 395
+    },
+    {
+      "epoch": 0.14798372179060304,
+      "grad_norm": 0.0,
+      "learning_rate": 1.986147861605908e-05,
+      "loss": 1.563,
+      "mean_token_accuracy": 0.7778670040335585,
+      "mean_token_accuracy_utility": 0.7571117432956742,
+      "step": 400
+    },
+    {
+      "epoch": 0.14798372179060304,
+      "eval_mean_token_accuracy": 0.7780364821605021,
+      "eval_reasoning_loss": 0.712975800037384,
+      "eval_reasoning_runtime": 90.6803,
+      "eval_reasoning_samples_per_second": 1.423,
+      "eval_reasoning_steps_per_second": 0.364,
+      "step": 400
+    },
+    {
+      "epoch": 0.14798372179060304,
+      "eval_mean_token_accuracy": 0.7369054990364875,
+      "eval_utility_loss": 0.9591025710105896,
+      "eval_utility_runtime": 34.0116,
+      "eval_utility_samples_per_second": 1.441,
+      "eval_utility_steps_per_second": 0.382,
+      "step": 400
+    },
+    {
+      "epoch": 0.14983351831298558,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9850559763924176e-05,
+      "loss": 1.6973,
+      "mean_token_accuracy": 0.7847627880636276,
+      "mean_token_accuracy_utility": 0.7250651694079209,
+      "step": 405
+    },
+    {
+      "epoch": 0.15168331483536812,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9839229977848164e-05,
+      "loss": 1.6847,
+      "mean_token_accuracy": 0.7745285592119512,
+      "mean_token_accuracy_utility": 0.7352641508080233,
+      "step": 410
+    },
+    {
+      "epoch": 0.15353311135775063,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9827489730473597e-05,
+      "loss": 1.6192,
+      "mean_token_accuracy": 0.8014127423771562,
+      "mean_token_accuracy_utility": 0.7350436544815817,
+      "step": 415
+    },
+    {
+      "epoch": 0.15538290788013318,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9815339511566148e-05,
+      "loss": 1.646,
+      "mean_token_accuracy": 0.7785625297210315,
+      "mean_token_accuracy_utility": 0.7378192890570616,
+      "step": 420
+    },
+    {
+      "epoch": 0.15723270440251572,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9802779827994214e-05,
+      "loss": 1.5718,
+      "mean_token_accuracy": 0.7826337980664549,
+      "mean_token_accuracy_utility": 0.7494993793068561,
+      "step": 425
+    },
+    {
+      "epoch": 0.15908250092489826,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9789811203707736e-05,
+      "loss": 1.645,
+      "mean_token_accuracy": 0.7845349985453558,
+      "mean_token_accuracy_utility": 0.7380305000311128,
+      "step": 430
+    },
+    {
+      "epoch": 0.1609322974472808,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9776434179716365e-05,
+      "loss": 1.5874,
+      "mean_token_accuracy": 0.7820796745999372,
+      "mean_token_accuracy_utility": 0.7492028148002882,
+      "step": 435
+    },
+    {
+      "epoch": 0.16278209396966334,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9762649314066888e-05,
+      "loss": 1.5365,
+      "mean_token_accuracy": 0.8036080041216456,
+      "mean_token_accuracy_utility": 0.7498171675041352,
+      "step": 440
+    },
+    {
+      "epoch": 0.16463189049204588,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9748457181819937e-05,
+      "loss": 1.595,
+      "mean_token_accuracy": 0.7836013663891406,
+      "mean_token_accuracy_utility": 0.7463781617521162,
+      "step": 445
+    },
+    {
+      "epoch": 0.16648168701442842,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9733858375026015e-05,
+      "loss": 1.6219,
+      "mean_token_accuracy": 0.799349090407528,
+      "mean_token_accuracy_utility": 0.7311774861677404,
+      "step": 450
+    },
+    {
+      "epoch": 0.16833148353681096,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9718853502700783e-05,
+      "loss": 1.6143,
+      "mean_token_accuracy": 0.786329083574207,
+      "mean_token_accuracy_utility": 0.7448906972276971,
+      "step": 455
+    },
+    {
+      "epoch": 0.1701812800591935,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9703443190799673e-05,
+      "loss": 1.6263,
+      "mean_token_accuracy": 0.7975003606152313,
+      "mean_token_accuracy_utility": 0.7333450756515173,
+      "step": 460
+    },
+    {
+      "epoch": 0.17203107658157601,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9687628082191748e-05,
+      "loss": 1.6553,
+      "mean_token_accuracy": 0.7832494306330073,
+      "mean_token_accuracy_utility": 0.734740171903834,
+      "step": 465
+    },
+    {
+      "epoch": 0.17388087310395856,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9671408836632915e-05,
+      "loss": 1.6025,
+      "mean_token_accuracy": 0.8038620491045789,
+      "mean_token_accuracy_utility": 0.7336802134678775,
+      "step": 470
+    },
+    {
+      "epoch": 0.1757306696263411,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9654786130738372e-05,
+      "loss": 1.6906,
+      "mean_token_accuracy": 0.7872269881028978,
+      "mean_token_accuracy_utility": 0.7291075363414395,
+      "step": 475
+    },
+    {
+      "epoch": 0.17758046614872364,
+      "grad_norm": 0.0,
+      "learning_rate": 1.96377606579544e-05,
+      "loss": 1.6348,
+      "mean_token_accuracy": 0.7782332609810034,
+      "mean_token_accuracy_utility": 0.7435790505698467,
+      "step": 480
+    },
+    {
+      "epoch": 0.17943026267110618,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9620333128529436e-05,
+      "loss": 1.5312,
+      "mean_token_accuracy": 0.8007622047538858,
+      "mean_token_accuracy_utility": 0.7489016004193436,
+      "step": 485
+    },
+    {
+      "epoch": 0.18128005919348872,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9602504269484432e-05,
+      "loss": 1.5954,
+      "mean_token_accuracy": 0.7797774370638224,
+      "mean_token_accuracy_utility": 0.7511323977603571,
+      "step": 490
+    },
+    {
+      "epoch": 0.18312985571587126,
+      "grad_norm": 0.0,
+      "learning_rate": 1.958427482458253e-05,
+      "loss": 1.646,
+      "mean_token_accuracy": 0.783821938174815,
+      "mean_token_accuracy_utility": 0.7367630267777819,
+      "step": 495
+    },
+    {
+      "epoch": 0.1849796522382538,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9565645554298043e-05,
+      "loss": 1.6066,
+      "mean_token_accuracy": 0.8077234346647186,
+      "mean_token_accuracy_utility": 0.7325853878797186,
+      "step": 500
+    },
+    {
+      "epoch": 0.1849796522382538,
+      "eval_mean_token_accuracy": 0.7804912077918045,
+      "eval_reasoning_loss": 0.7005118131637573,
+      "eval_reasoning_runtime": 90.0281,
+      "eval_reasoning_samples_per_second": 1.433,
+      "eval_reasoning_steps_per_second": 0.367,
+      "step": 500
+    },
+    {
+      "epoch": 0.1849796522382538,
+      "eval_mean_token_accuracy": 0.7372408455159208,
+      "eval_utility_loss": 0.9542421698570251,
+      "eval_utility_runtime": 33.921,
+      "eval_utility_samples_per_second": 1.445,
+      "eval_utility_steps_per_second": 0.383,
+      "step": 500
+    },
+    {
+      "epoch": 0.18682944876063634,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9546617235784716e-05,
+      "loss": 1.6094,
+      "mean_token_accuracy": 0.7780880131174539,
+      "mean_token_accuracy_utility": 0.7460220007050765,
+      "step": 505
+    },
+    {
+      "epoch": 0.18867924528301888,
+      "grad_norm": 0.0,
+      "learning_rate": 1.952719066284333e-05,
+      "loss": 1.5872,
+      "mean_token_accuracy": 0.7896033143457031,
+      "mean_token_accuracy_utility": 0.7425022004288537,
+      "step": 510
+    },
+    {
+      "epoch": 0.1905290418054014,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9507366645888544e-05,
+      "loss": 1.6041,
+      "mean_token_accuracy": 0.7939051421962724,
+      "mean_token_accuracy_utility": 0.73778907451771,
+      "step": 515
+    },
+    {
+      "epoch": 0.19237883832778394,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9487146011915142e-05,
+      "loss": 1.579,
+      "mean_token_accuracy": 0.7933889277386552,
+      "mean_token_accuracy_utility": 0.7445984069700542,
+      "step": 520
+    },
+    {
+      "epoch": 0.19422863485016648,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9466529604463484e-05,
+      "loss": 1.6284,
+      "mean_token_accuracy": 0.790335880625766,
+      "mean_token_accuracy_utility": 0.7359807919074978,
+      "step": 525
+    },
+    {
+      "epoch": 0.19607843137254902,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9445518283584345e-05,
+      "loss": 1.5347,
+      "mean_token_accuracy": 0.7846728951313512,
+      "mean_token_accuracy_utility": 0.7588100954431996,
+      "step": 530
+    },
+    {
+      "epoch": 0.19792822789493156,
+      "grad_norm": 0.0,
+      "learning_rate": 1.942411292580304e-05,
+      "loss": 1.5042,
+      "mean_token_accuracy": 0.8035783937451907,
+      "mean_token_accuracy_utility": 0.7504310031909276,
+      "step": 535
+    },
+    {
+      "epoch": 0.1997780244173141,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9402314424082824e-05,
+      "loss": 1.5963,
+      "mean_token_accuracy": 0.7876869937786629,
+      "mean_token_accuracy_utility": 0.7462532308898537,
+      "step": 540
+    },
+    {
+      "epoch": 0.20162782093969664,
+      "grad_norm": 0.0,
+      "learning_rate": 1.938012368778768e-05,
+      "loss": 1.5835,
+      "mean_token_accuracy": 0.7870018873375523,
+      "mean_token_accuracy_utility": 0.7498718523651534,
+      "step": 545
+    },
+    {
+      "epoch": 0.20347761746207918,
+      "grad_norm": 0.0,
+      "learning_rate": 1.935754164264437e-05,
+      "loss": 1.5765,
+      "mean_token_accuracy": 0.7938986449862605,
+      "mean_token_accuracy_utility": 0.7474433020024746,
+      "step": 550
+    },
+    {
+      "epoch": 0.20532741398446172,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9334569230703794e-05,
+      "loss": 1.5732,
+      "mean_token_accuracy": 0.791866445306688,
+      "mean_token_accuracy_utility": 0.7478883444609311,
+      "step": 555
+    },
+    {
+      "epoch": 0.20717721050684423,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9311207410301728e-05,
+      "loss": 1.6202,
+      "mean_token_accuracy": 0.7825917307159231,
+      "mean_token_accuracy_utility": 0.7445319847777169,
+      "step": 560
+    },
+    {
+      "epoch": 0.20902700702922677,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9287457156018824e-05,
+      "loss": 1.5478,
+      "mean_token_accuracy": 0.8030853591237079,
+      "mean_token_accuracy_utility": 0.7453821098957968,
+      "step": 565
+    },
+    {
+      "epoch": 0.21087680355160932,
+      "grad_norm": 0.0,
+      "learning_rate": 1.926331945863995e-05,
+      "loss": 1.5469,
+      "mean_token_accuracy": 0.7947856528242329,
+      "mean_token_accuracy_utility": 0.7516773128885312,
+      "step": 570
+    },
+    {
+      "epoch": 0.21272660007399186,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9238795325112867e-05,
+      "loss": 1.5545,
+      "mean_token_accuracy": 0.7955434158959365,
+      "mean_token_accuracy_utility": 0.7458566136714754,
+      "step": 575
+    },
+    {
+      "epoch": 0.2145763965963744,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9213885778506228e-05,
+      "loss": 1.4586,
+      "mean_token_accuracy": 0.7973081924318149,
+      "mean_token_accuracy_utility": 0.7667047144499091,
+      "step": 580
+    },
+    {
+      "epoch": 0.21642619311875694,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9188591857966875e-05,
+      "loss": 1.5611,
+      "mean_token_accuracy": 0.7910839580494021,
+      "mean_token_accuracy_utility": 0.752717403597731,
+      "step": 585
+    },
+    {
+      "epoch": 0.21827598964113948,
+      "grad_norm": 0.0,
+      "learning_rate": 1.916291461867652e-05,
+      "loss": 1.5039,
+      "mean_token_accuracy": 0.7974834813478895,
+      "mean_token_accuracy_utility": 0.760627676760052,
+      "step": 590
+    },
+    {
+      "epoch": 0.22012578616352202,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9136855131807705e-05,
+      "loss": 1.6195,
+      "mean_token_accuracy": 0.7905506642768041,
+      "mean_token_accuracy_utility": 0.7418094377920108,
+      "step": 595
+    },
+    {
+      "epoch": 0.22197558268590456,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9110414484479116e-05,
+      "loss": 1.577,
+      "mean_token_accuracy": 0.788761194291807,
+      "mean_token_accuracy_utility": 0.7492097243372814,
+      "step": 600
+    },
+    {
+      "epoch": 0.22197558268590456,
+      "eval_mean_token_accuracy": 0.7826237115875145,
+      "eval_reasoning_loss": 0.6918679475784302,
+      "eval_reasoning_runtime": 90.1443,
+      "eval_reasoning_samples_per_second": 1.431,
+      "eval_reasoning_steps_per_second": 0.366,
+      "step": 600
+    },
+    {
+      "epoch": 0.22197558268590456,
+      "eval_mean_token_accuracy": 0.7384578556084408,
+      "eval_utility_loss": 0.9520853757858276,
+      "eval_utility_runtime": 33.9433,
+      "eval_utility_samples_per_second": 1.444,
+      "eval_utility_steps_per_second": 0.383,
+      "step": 600
+    },
+    {
+      "epoch": 0.2238253792082871,
+      "grad_norm": 0.0,
+      "learning_rate": 1.908359377971025e-05,
+      "loss": 1.6093,
+      "mean_token_accuracy": 0.7994627793099325,
+      "mean_token_accuracy_utility": 0.738760716737831,
+      "step": 605
+    },
+    {
+      "epoch": 0.2256751757306696,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9056394136375377e-05,
+      "loss": 1.5666,
+      "mean_token_accuracy": 0.8000460007767167,
+      "mean_token_accuracy_utility": 0.7431317026627413,
+      "step": 610
+    },
+    {
+      "epoch": 0.22752497225305215,
+      "grad_norm": 0.0,
+      "learning_rate": 1.902881668915688e-05,
+      "loss": 1.5537,
+      "mean_token_accuracy": 0.7886050473144798,
+      "mean_token_accuracy_utility": 0.7548217395720888,
+      "step": 615
+    },
+    {
+      "epoch": 0.2293747687754347,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9000862588497918e-05,
+      "loss": 1.4977,
+      "mean_token_accuracy": 0.7961878899829054,
+      "mean_token_accuracy_utility": 0.7570915528611445,
+      "step": 620
+    },
+    {
+      "epoch": 0.23122456529781724,
+      "grad_norm": 0.0,
+      "learning_rate": 1.897253300055443e-05,
+      "loss": 1.5307,
+      "mean_token_accuracy": 0.7927883775820486,
+      "mean_token_accuracy_utility": 0.7527269908915586,
+      "step": 625
+    },
+    {
+      "epoch": 0.23307436182019978,
+      "grad_norm": 0.0,
+      "learning_rate": 1.8943829107146485e-05,
+      "loss": 1.6508,
+      "mean_token_accuracy": 0.7882507058490691,
+      "mean_token_accuracy_utility": 0.7347466691408899,
+      "step": 630
+    },
+    {
+      "epoch": 0.23492415834258232,
+      "grad_norm": 0.0,
+      "learning_rate": 1.891475210570898e-05,
+      "loss": 1.5385,
+      "mean_token_accuracy": 0.7931458895799018,
+      "mean_token_accuracy_utility": 0.7566521015473603,
+      "step": 635
+    },
+    {
+      "epoch": 0.23677395486496486,
+      "grad_norm": 0.0,
+      "learning_rate": 1.8885303209241694e-05,
+      "loss": 1.5933,
+      "mean_token_accuracy": 0.7868157034568986,
+      "mean_token_accuracy_utility": 0.7485301365421061,
+      "step": 640
+    },
+    {
+      "epoch": 0.2386237513873474,
+      "grad_norm": 0.0,
+      "learning_rate": 1.8855483646258677e-05,
+      "loss": 1.5593,
+      "mean_token_accuracy": 0.801889314508242,
+      "mean_token_accuracy_utility": 0.737664591594324,
+      "step": 645
+    },
+    {
+      "epoch": 0.24047354790972994,
+      "grad_norm": 0.0,
+      "learning_rate": 1.882529466073701e-05,
+      "loss": 1.5271,
+      "mean_token_accuracy": 0.7983072536270053,
+      "mean_token_accuracy_utility": 0.7486188943060752,
+      "step": 650
+    },
+    {
+      "epoch": 0.24232334443211248,
+      "grad_norm": 0.0,
+      "learning_rate": 1.879473751206489e-05,
+      "loss": 1.605,
+      "mean_token_accuracy": 0.7918422770179967,
+      "mean_token_accuracy_utility": 0.740652519722483,
+      "step": 655
+    },
+    {
+      "epoch": 0.244173140954495,
+      "grad_norm": 0.0,
+      "learning_rate": 1.876381347498913e-05,
+      "loss": 1.5839,
+      "mean_token_accuracy": 0.7899708565360778,
+      "mean_token_accuracy_utility": 0.7417692789690113,
+      "step": 660
+    },
+    {
+      "epoch": 0.24602293747687753,
+      "grad_norm": 0.0,
+      "learning_rate": 1.8732523839561934e-05,
+      "loss": 1.5682,
+      "mean_token_accuracy": 0.7831179180616784,
+      "mean_token_accuracy_utility": 0.7551691191238339,
+      "step": 665
+    },
+    {
+      "epoch": 0.24787273399926008,
+      "grad_norm": 0.0,
+      "learning_rate": 1.8700869911087115e-05,
+      "loss": 1.6142,
+      "mean_token_accuracy": 0.785560742298103,
+      "mean_token_accuracy_utility": 0.7470067479444947,
+      "step": 670
+    },
+    {
+      "epoch": 0.24972253052164262,
+      "grad_norm": 0.0,
+      "learning_rate": 1.8668853010065633e-05,
+      "loss": 1.5296,
+      "mean_token_accuracy": 0.7957932970343712,
+      "mean_token_accuracy_utility": 0.7538769957768345,
+      "step": 675
+    },
+    {
+      "epoch": 0.25157232704402516,
+      "grad_norm": 0.0,
+      "learning_rate": 1.8636474472140494e-05,
+      "loss": 1.6633,
+      "mean_token_accuracy": 0.7851419074262755,
+      "mean_token_accuracy_utility": 0.7325757912613898,
+      "step": 680
+    },
+    {
+      "epoch": 0.2534221235664077,
+      "grad_norm": 0.0,
+      "learning_rate": 1.8603735648041054e-05,
+      "loss": 1.525,
+      "mean_token_accuracy": 0.7935844102074241,
+      "mean_token_accuracy_utility": 0.7529442196340099,
+      "step": 685
+    },
+    {
+      "epoch": 0.25527192008879024,
+      "grad_norm": 0.0,
+      "learning_rate": 1.857063790352665e-05,
+      "loss": 1.4989,
+      "mean_token_accuracy": 0.8140169550536964,
+      "mean_token_accuracy_utility": 0.7475299775666475,
+      "step": 690
+    },
+    {
+      "epoch": 0.2571217166111728,
+      "grad_norm": 0.0,
+      "learning_rate": 1.853718261932964e-05,
+      "loss": 1.4706,
+      "mean_token_accuracy": 0.8014806898761047,
+      "mean_token_accuracy_utility": 0.7632108700490211,
+      "step": 695
+    },
+    {
+      "epoch": 0.2589715131335553,
+      "grad_norm": 0.0,
+      "learning_rate": 1.85033711910978e-05,
+      "loss": 1.5318,
+      "mean_token_accuracy": 0.7997687608648569,
+      "mean_token_accuracy_utility": 0.7555465256521205,
+      "step": 700
+    },
+    {
+      "epoch": 0.2589715131335553,
+      "eval_mean_token_accuracy": 0.7836580775397225,
+      "eval_reasoning_loss": 0.6857641935348511,
+      "eval_reasoning_runtime": 90.1509,
+      "eval_reasoning_samples_per_second": 1.431,
+      "eval_reasoning_steps_per_second": 0.366,
+      "step": 700
+    },
+    {
+      "epoch": 0.2589715131335553,
+      "eval_mean_token_accuracy": 0.7396099323580656,
+      "eval_utility_loss": 0.9490150809288025,
+      "eval_utility_runtime": 33.9288,
+      "eval_utility_samples_per_second": 1.444,
+      "eval_utility_steps_per_second": 0.383,
+      "step": 700
+    },
+    {
+      "epoch": 0.26082130965593786,
+      "grad_norm": 0.0,
+      "learning_rate": 1.846920502933609e-05,
+      "loss": 1.4844,
+      "mean_token_accuracy": 0.8008248474488315,
+      "mean_token_accuracy_utility": 0.7579721480698782,
+      "step": 705
+    },
+    {
+      "epoch": 0.2626711061783204,
+      "grad_norm": 0.0,
+      "learning_rate": 1.843468555934783e-05,
+      "loss": 1.5649,
+      "mean_token_accuracy": 0.7900554016336423,
+      "mean_token_accuracy_utility": 0.7489296459662387,
+      "step": 710
+    },
+    {
+      "epoch": 0.26452090270070294,
+      "grad_norm": 0.0,
+      "learning_rate": 1.839981422117523e-05,
+      "loss": 1.5405,
+      "mean_token_accuracy": 0.7911753368094048,
+      "mean_token_accuracy_utility": 0.7526646736656514,
+      "step": 715
+    },
+    {
+      "epoch": 0.2663706992230855,
+      "grad_norm": 0.0,
+      "learning_rate": 1.836459246953931e-05,
+      "loss": 1.4749,
+      "mean_token_accuracy": 0.7947672770686933,
+      "mean_token_accuracy_utility": 0.7627778821056508,
+      "step": 720
+    },
+    {
+      "epoch": 0.268220495745468,
+      "grad_norm": 0.0,
+      "learning_rate": 1.8329021773779242e-05,
+      "loss": 1.5524,
+      "mean_token_accuracy": 0.7922104417426783,
+      "mean_token_accuracy_utility": 0.7525233239387653,
+      "step": 725
+    },
+    {
+      "epoch": 0.2700702922678505,
+      "grad_norm": 0.0,
+      "learning_rate": 1.8293103617791016e-05,
+      "loss": 1.6175,
+      "mean_token_accuracy": 0.7888341895385377,
+      "mean_token_accuracy_utility": 0.7377134017162835,
+      "step": 730
+    },
+    {
+      "epoch": 0.27192008879023305,
+      "grad_norm": 0.0,
+      "learning_rate": 1.825683949996556e-05,
+      "loss": 1.5974,
+      "mean_token_accuracy": 0.7863125983342814,
+      "mean_token_accuracy_utility": 0.7492471080383454,
+      "step": 735
+    },
+    {
+      "epoch": 0.2737698853126156,
+      "grad_norm": 0.0,
+      "learning_rate": 1.8220230933126234e-05,
+      "loss": 1.4785,
+      "mean_token_accuracy": 0.805230735002934,
+      "mean_token_accuracy_utility": 0.7558428141582161,
+      "step": 740
+    },
+    {
+      "epoch": 0.27561968183499813,
+      "grad_norm": 0.0,
+      "learning_rate": 1.81832794444657e-05,
+      "loss": 1.5181,
+      "mean_token_accuracy": 0.7968833952284236,
+      "mean_token_accuracy_utility": 0.7560776599282233,
+      "step": 745
+    },
+    {
+      "epoch": 0.27746947835738067,
+      "grad_norm": 0.0,
+      "learning_rate": 1.814598657548223e-05,
+      "loss": 1.5,
+      "mean_token_accuracy": 0.8011237501455408,
+      "mean_token_accuracy_utility": 0.7569460912665085,
+      "step": 750
+    },
+    {
+      "epoch": 0.2793192748797632,
+      "grad_norm": 0.0,
+      "learning_rate": 1.8108353881915403e-05,
+      "loss": 1.5446,
+      "mean_token_accuracy": 0.7963633662700839,
+      "mean_token_accuracy_utility": 0.7491219709442277,
+      "step": 755
+    },
+    {
+      "epoch": 0.28116907140214575,
+      "grad_norm": 0.0,
+      "learning_rate": 1.8070382933681185e-05,
+      "loss": 1.527,
+      "mean_token_accuracy": 0.7918860852471548,
+      "mean_token_accuracy_utility": 0.7583129923801624,
+      "step": 760
+    },
+    {
+      "epoch": 0.2830188679245283,
+      "grad_norm": 0.0,
+      "learning_rate": 1.803207531480645e-05,
+      "loss": 1.5717,
+      "mean_token_accuracy": 0.7923941542583899,
+      "mean_token_accuracy_utility": 0.7439698985388992,
+      "step": 765
+    },
+    {
+      "epoch": 0.28486866444691084,
+      "grad_norm": 0.0,
+      "learning_rate": 1.79934326233629e-05,
+      "loss": 1.4845,
+      "mean_token_accuracy": 0.801383413393107,
+      "mean_token_accuracy_utility": 0.7590267465902416,
+      "step": 770
+    },
+    {
+      "epoch": 0.2867184609692934,
+      "grad_norm": 0.0,
+      "learning_rate": 1.7954456471400393e-05,
+      "loss": 1.5702,
+      "mean_token_accuracy": 0.78829923215791,
+      "mean_token_accuracy_utility": 0.7452273802729639,
+      "step": 775
+    },
+    {
+      "epoch": 0.2885682574916759,
+      "grad_norm": 0.0,
+      "learning_rate": 1.7915148484879706e-05,
+      "loss": 1.5601,
+      "mean_token_accuracy": 0.7944471132518917,
+      "mean_token_accuracy_utility": 0.7472533747115235,
+      "step": 780
+    },
+    {
+      "epoch": 0.29041805401405846,
+      "grad_norm": 0.0,
+      "learning_rate": 1.7875510303604678e-05,
+      "loss": 1.5466,
+      "mean_token_accuracy": 0.8043097870558542,
+      "mean_token_accuracy_utility": 0.7505715820498937,
+      "step": 785
+    },
+    {
+      "epoch": 0.292267850536441,
+      "grad_norm": 0.0,
+      "learning_rate": 1.783554358115384e-05,
+      "loss": 1.5095,
+      "mean_token_accuracy": 0.7827916545947886,
+      "mean_token_accuracy_utility": 0.7651371194427845,
+      "step": 790
+    },
+    {
+      "epoch": 0.29411764705882354,
+      "grad_norm": 0.0,
+      "learning_rate": 1.7795249984811397e-05,
+      "loss": 1.5325,
+      "mean_token_accuracy": 0.8047935947895073,
+      "mean_token_accuracy_utility": 0.7483347819857946,
+      "step": 795
+    },
+    {
+      "epoch": 0.2959674435812061,
+      "grad_norm": 0.0,
+      "learning_rate": 1.77546311954977e-05,
+      "loss": 1.5218,
+      "mean_token_accuracy": 0.787624963589408,
+      "mean_token_accuracy_utility": 0.7584633633833917,
+      "step": 800
+    },
+    {
+      "epoch": 0.2959674435812061,
+      "eval_mean_token_accuracy": 0.7851614319345736,
+      "eval_reasoning_loss": 0.6805456876754761,
+      "eval_reasoning_runtime": 90.042,
+      "eval_reasoning_samples_per_second": 1.433,
+      "eval_reasoning_steps_per_second": 0.366,
+      "step": 800
+    },
+    {
+      "epoch": 0.2959674435812061,
+      "eval_mean_token_accuracy": 0.7398368534531387,
+      "eval_utility_loss": 0.9480440020561218,
+      "eval_utility_runtime": 33.9284,
+      "eval_utility_samples_per_second": 1.444,
+      "eval_utility_steps_per_second": 0.383,
+      "step": 800
+    },
+    {
+      "epoch": 0.2978172401035886,
+      "grad_norm": 0.0,
+      "learning_rate": 1.7713688907699107e-05,
+      "loss": 1.5654,
+      "mean_token_accuracy": 0.7807157293881373,
+      "mean_token_accuracy_utility": 0.7593273214487843,
+      "step": 805
+    },
+    {
+      "epoch": 0.29966703662597116,
+      "grad_norm": 0.0,
+      "learning_rate": 1.767242482939731e-05,
+      "loss": 1.5103,
+      "mean_token_accuracy": 0.7919583483571205,
+      "mean_token_accuracy_utility": 0.7604765068535908,
+      "step": 810
+    },
+    {
+      "epoch": 0.3015168331483537,
+      "grad_norm": 0.0,
+      "learning_rate": 1.7630840681998068e-05,
+      "loss": 1.4316,
+      "mean_token_accuracy": 0.8044022883003159,
+      "mean_token_accuracy_utility": 0.7689988647774719,
+      "step": 815
+    },
+    {
+      "epoch": 0.30336662967073624,
+      "grad_norm": 0.0,
+      "learning_rate": 1.75889382002594e-05,
+      "loss": 1.5025,
+      "mean_token_accuracy": 0.7904034293787084,
+      "mean_token_accuracy_utility": 0.7619356475301017,
+      "step": 820
+    },
+    {
+      "epoch": 0.3052164261931188,
+      "grad_norm": 0.0,
+      "learning_rate": 1.754671913221923e-05,
+      "loss": 1.5559,
+      "mean_token_accuracy": 0.791941284816697,
+      "mean_token_accuracy_utility": 0.7546328516605671,
+      "step": 825
+    },
+    {
+      "epoch": 0.30706622271550127,
+      "grad_norm": 0.0,
+      "learning_rate": 1.7504185239122432e-05,
+      "loss": 1.4839,
+      "mean_token_accuracy": 0.8003207343121531,
+      "mean_token_accuracy_utility": 0.7581757480943213,
+      "step": 830
+    },
+    {
+      "epoch": 0.3089160192378838,
+      "grad_norm": 0.0,
+      "learning_rate": 1.7461338295347404e-05,
+      "loss": 1.487,
+      "mean_token_accuracy": 0.8042871188065274,
+      "mean_token_accuracy_utility": 0.7527410739964393,
+      "step": 835
+    },
+    {
+      "epoch": 0.31076581576026635,
+      "grad_norm": 0.0,
+      "learning_rate": 1.7418180088332005e-05,
+      "loss": 1.4542,
+      "mean_token_accuracy": 0.7986678668737642,
+      "mean_token_accuracy_utility": 0.7671164488532893,
+      "step": 840
+    },
+    {
+      "epoch": 0.3126156122826489,
+      "grad_norm": 0.0,
+      "learning_rate": 1.7374712418498997e-05,
+      "loss": 1.5528,
+      "mean_token_accuracy": 0.798378861307587,
+      "mean_token_accuracy_utility": 0.7516931752784031,
+      "step": 845
+    },
+    {
+      "epoch": 0.31446540880503143,
+      "grad_norm": 0.0,
+      "learning_rate": 1.7330937099180956e-05,
+      "loss": 1.5784,
+      "mean_token_accuracy": 0.7876867853104713,
+      "mean_token_accuracy_utility": 0.7485566191859647,
+      "step": 850
+    },
+    {
+      "epoch": 0.316315205327414,
+      "grad_norm": 0.0,
+      "learning_rate": 1.7286855956544616e-05,
+      "loss": 1.5145,
+      "mean_token_accuracy": 0.797946962093533,
+      "mean_token_accuracy_utility": 0.7558738867730139,
+      "step": 855
+    },
+    {
+      "epoch": 0.3181650018497965,
+      "grad_norm": 0.0,
+      "learning_rate": 1.7242470829514674e-05,
+      "loss": 1.5067,
+      "mean_token_accuracy": 0.7881723727584089,
+      "mean_token_accuracy_utility": 0.7621536453662271,
+      "step": 860
+    },
+    {
+      "epoch": 0.32001479837217905,
+      "grad_norm": 0.0,
+      "learning_rate": 1.7197783569697084e-05,
+      "loss": 1.5289,
+      "mean_token_accuracy": 0.798927809203237,
+      "mean_token_accuracy_utility": 0.7538099751313516,
+      "step": 865
+    },
+    {
+      "epoch": 0.3218645948945616,
+      "grad_norm": 0.0,
+      "learning_rate": 1.715279604130183e-05,
+      "loss": 1.5506,
+      "mean_token_accuracy": 0.785820941102273,
+      "mean_token_accuracy_utility": 0.7530414768435365,
+      "step": 870
+    },
+    {
+      "epoch": 0.32371439141694414,
+      "grad_norm": 0.0,
+      "learning_rate": 1.7107510121065138e-05,
+      "loss": 1.5688,
+      "mean_token_accuracy": 0.8003720552776625,
+      "mean_token_accuracy_utility": 0.7452532283381623,
+      "step": 875
+    },
+    {
+      "epoch": 0.3255641879393267,
+      "grad_norm": 0.0,
+      "learning_rate": 1.7061927698171187e-05,
+      "loss": 1.446,
+      "mean_token_accuracy": 0.8123360330768458,
+      "mean_token_accuracy_utility": 0.7578578332033677,
+      "step": 880
+    },
+    {
+      "epoch": 0.3274139844617092,
+      "grad_norm": 0.0,
+      "learning_rate": 1.7016050674173304e-05,
+      "loss": 1.5411,
+      "mean_token_accuracy": 0.7971175228348745,
+      "mean_token_accuracy_utility": 0.7539543314099062,
+      "step": 885
+    },
+    {
+      "epoch": 0.32926378098409176,
+      "grad_norm": 0.0,
+      "learning_rate": 1.696988096291463e-05,
+      "loss": 1.5813,
+      "mean_token_accuracy": 0.7866782807751471,
+      "mean_token_accuracy_utility": 0.7574907614611288,
+      "step": 890
+    },
+    {
+      "epoch": 0.3311135775064743,
+      "grad_norm": 0.0,
+      "learning_rate": 1.6923420490448298e-05,
+      "loss": 1.4984,
+      "mean_token_accuracy": 0.7955661263576057,
+      "mean_token_accuracy_utility": 0.7586219636706723,
+      "step": 895
+    },
+    {
+      "epoch": 0.33296337402885684,
+      "grad_norm": 0.0,
+      "learning_rate": 1.687667119495705e-05,
+      "loss": 1.544,
+      "mean_token_accuracy": 0.7862944108919959,
+      "mean_token_accuracy_utility": 0.7564557318292537,
+      "step": 900
+    },
+    {
+      "epoch": 0.33296337402885684,
+      "eval_mean_token_accuracy": 0.7862377771872587,
+      "eval_reasoning_loss": 0.6764837503433228,
+      "eval_reasoning_runtime": 89.978,
+      "eval_reasoning_samples_per_second": 1.434,
+      "eval_reasoning_steps_per_second": 0.367,
+      "step": 900
+    },
+    {
+      "epoch": 0.33296337402885684,
+      "eval_mean_token_accuracy": 0.7397851720759727,
+      "eval_utility_loss": 0.9469250440597534,
+      "eval_utility_runtime": 33.8454,
+      "eval_utility_samples_per_second": 1.448,
+      "eval_utility_steps_per_second": 0.384,
+      "step": 900
+    },
+    {
+      "epoch": 0.3348131705512394,
+      "grad_norm": 0.0,
+      "learning_rate": 1.6829635026672432e-05,
+      "loss": 1.455,
+      "mean_token_accuracy": 0.7959991570517324,
+      "mean_token_accuracy_utility": 0.7680822004768213,
+      "step": 905
+    },
+    {
+      "epoch": 0.3366629670736219,
+      "grad_norm": 0.0,
+      "learning_rate": 1.6782313947793377e-05,
+      "loss": 1.4925,
+      "mean_token_accuracy": 0.794914989801385,
+      "mean_token_accuracy_utility": 0.7638315056035883,
+      "step": 910
+    },
+    {
+      "epoch": 0.33851276359600446,
+      "grad_norm": 0.0,
+      "learning_rate": 1.6734709932404404e-05,
+      "loss": 1.5189,
+      "mean_token_accuracy": 0.7944884604921236,
+      "mean_token_accuracy_utility": 0.7540919403297188,
+      "step": 915
+    },
+    {
+      "epoch": 0.340362560118387,
+      "grad_norm": 0.0,
+      "learning_rate": 1.668682496639323e-05,
+      "loss": 1.5651,
+      "mean_token_accuracy": 0.7939252998907302,
+      "mean_token_accuracy_utility": 0.7508702425035333,
+      "step": 920
+    },
+    {
+      "epoch": 0.3422123566407695,
+      "grad_norm": 0.0,
+      "learning_rate": 1.663866104736793e-05,
+      "loss": 1.5032,
+      "mean_token_accuracy": 0.8097460655650337,
+      "mean_token_accuracy_utility": 0.7506331568021628,
+      "step": 925
+    },
+    {
+      "epoch": 0.34406215316315203,
+      "grad_norm": 0.0,
+      "learning_rate": 1.6590220184573623e-05,
+      "loss": 1.5819,
+      "mean_token_accuracy": 0.7838914099504651,
+      "mean_token_accuracy_utility": 0.752055123686435,
+      "step": 930
+    },
+    {
+      "epoch": 0.34591194968553457,
+      "grad_norm": 0.0,
+      "learning_rate": 1.6541504398808633e-05,
+      "loss": 1.5354,
+      "mean_token_accuracy": 0.7959231146251906,
+      "mean_token_accuracy_utility": 0.7539707563320613,
+      "step": 935
+    },
+    {
+      "epoch": 0.3477617462079171,
+      "grad_norm": 0.0,
+      "learning_rate": 1.6492515722340193e-05,
+      "loss": 1.5019,
+      "mean_token_accuracy": 0.7807037874626199,
+      "mean_token_accuracy_utility": 0.7700537463971331,
+      "step": 940
+    },
+    {
+      "epoch": 0.34961154273029965,
+      "grad_norm": 0.0,
+      "learning_rate": 1.6443256198819665e-05,
+      "loss": 1.567,
+      "mean_token_accuracy": 0.7856720426769013,
+      "mean_token_accuracy_utility": 0.752830438601718,
+      "step": 945
+    },
+    {
+      "epoch": 0.3514613392526822,
+      "grad_norm": 0.0,
+      "learning_rate": 1.6393727883197292e-05,
+      "loss": 1.5121,
+      "mean_token_accuracy": 0.7959155614093008,
+      "mean_token_accuracy_utility": 0.7588089738880349,
+      "step": 950
+    },
+    {
+      "epoch": 0.35331113577506473,
+      "grad_norm": 0.0,
+      "learning_rate": 1.6343932841636455e-05,
+      "loss": 1.4332,
+      "mean_token_accuracy": 0.7917139692128423,
+      "mean_token_accuracy_utility": 0.7835149860114493,
+      "step": 955
+    },
+    {
+      "epoch": 0.3551609322974473,
+      "grad_norm": 0.0,
+      "learning_rate": 1.6293873151427505e-05,
+      "loss": 1.5611,
+      "mean_token_accuracy": 0.7878615093944987,
+      "mean_token_accuracy_utility": 0.7521612118532911,
+      "step": 960
+    },
+    {
+      "epoch": 0.3570107288198298,
+      "grad_norm": 0.0,
+      "learning_rate": 1.6243550900901076e-05,
+      "loss": 1.5874,
+      "mean_token_accuracy": 0.7871017346649672,
+      "mean_token_accuracy_utility": 0.7508877036734962,
+      "step": 965
+    },
+    {
+      "epoch": 0.35886052534221236,
+      "grad_norm": 0.0,
+      "learning_rate": 1.6192968189340993e-05,
+      "loss": 1.5355,
+      "mean_token_accuracy": 0.7903840430453206,
+      "mean_token_accuracy_utility": 0.752038993314494,
+      "step": 970
+    },
+    {
+      "epoch": 0.3607103218645949,
+      "grad_norm": 0.0,
+      "learning_rate": 1.6142127126896682e-05,
+      "loss": 1.546,
+      "mean_token_accuracy": 0.7877703321187377,
+      "mean_token_accuracy_utility": 0.7587006742256694,
+      "step": 975
+    },
+    {
+      "epoch": 0.36256011838697744,
+      "grad_norm": 0.0,
+      "learning_rate": 1.6091029834495133e-05,
+      "loss": 1.5623,
+      "mean_token_accuracy": 0.8012606695973318,
+      "mean_token_accuracy_utility": 0.7493255756969764,
+      "step": 980
+    },
+    {
+      "epoch": 0.36440991490936,
+      "grad_norm": 0.0,
+      "learning_rate": 1.603967844375245e-05,
+      "loss": 1.5371,
+      "mean_token_accuracy": 0.795543800919306,
+      "mean_token_accuracy_utility": 0.7529303076771718,
+      "step": 985
+    },
+    {
+      "epoch": 0.3662597114317425,
+      "grad_norm": 0.0,
+      "learning_rate": 1.59880750968849e-05,
+      "loss": 1.4815,
+      "mean_token_accuracy": 0.7955903811607893,
+      "mean_token_accuracy_utility": 0.7632127474905787,
+      "step": 990
+    },
+    {
+      "epoch": 0.36810950795412506,
+      "grad_norm": 0.0,
+      "learning_rate": 1.593622194661956e-05,
+      "loss": 1.4886,
+      "mean_token_accuracy": 0.8087892470985972,
+      "mean_token_accuracy_utility": 0.7537307519092351,
+      "step": 995
+    },
+    {
+      "epoch": 0.3699593044765076,
+      "grad_norm": 0.0,
+      "learning_rate": 1.5884121156104496e-05,
+      "loss": 1.4836,
+      "mean_token_accuracy": 0.797642302581661,
+      "mean_token_accuracy_utility": 0.7627143952389814,
+      "step": 1000
+    },
+    {
+      "epoch": 0.3699593044765076,
+      "eval_mean_token_accuracy": 0.7872027670535536,
+      "eval_reasoning_loss": 0.6731064915657043,
+      "eval_reasoning_runtime": 89.9015,
+      "eval_reasoning_samples_per_second": 1.435,
+      "eval_reasoning_steps_per_second": 0.367,
+      "step": 1000
+    },
+    {
+      "epoch": 0.3699593044765076,
+      "eval_mean_token_accuracy": 0.7401212426997975,
+      "eval_utility_loss": 0.9461431503295898,
+      "eval_utility_runtime": 33.852,
+      "eval_utility_samples_per_second": 1.447,
+      "eval_utility_steps_per_second": 0.384,
+      "step": 1000
+    },
+    {
+      "epoch": 0.37180910099889014,
+      "grad_norm": 0.0,
+      "learning_rate": 1.5831774898818558e-05,
+      "loss": 1.6078,
+      "mean_token_accuracy": 0.7873670958749359,
+      "mean_token_accuracy_utility": 0.7451432608807448,
+      "step": 1005
+    },
+    {
+      "epoch": 0.3736588975212727,
+      "grad_norm": 0.0,
+      "learning_rate": 1.5779185358480666e-05,
+      "loss": 1.5251,
+      "mean_token_accuracy": 0.7926603228278785,
+      "mean_token_accuracy_utility": 0.7580437533178841,
+      "step": 1010
+    },
+    {
+      "epoch": 0.3755086940436552,
+      "grad_norm": 0.0,
+      "learning_rate": 1.5726354728958736e-05,
+      "loss": 1.5129,
+      "mean_token_accuracy": 0.7993613364230664,
+      "mean_token_accuracy_utility": 0.7581680001146393,
+      "step": 1015
+    },
+    {
+      "epoch": 0.37735849056603776,
+      "grad_norm": 0.0,
+      "learning_rate": 1.5673285214178176e-05,
+      "loss": 1.5472,
+      "mean_token_accuracy": 0.7982543437447432,
+      "mean_token_accuracy_utility": 0.7476179511753923,
+      "step": 1020
+    },
+    {
+      "epoch": 0.37920828708842025,
+      "grad_norm": 0.0,
+      "learning_rate": 1.5619979028029898e-05,
+      "loss": 1.4757,
+      "mean_token_accuracy": 0.7992625043428105,
+      "mean_token_accuracy_utility": 0.7627592821504796,
+      "step": 1025
+    },
+    {
+      "epoch": 0.3810580836108028,
+      "grad_norm": 0.0,
+      "learning_rate": 1.5566438394278005e-05,
+      "loss": 1.5565,
+      "mean_token_accuracy": 0.8078204295821692,
+      "mean_token_accuracy_utility": 0.7379954920788065,
+      "step": 1030
+    },
+    {
+      "epoch": 0.38290788013318533,
+      "grad_norm": 0.0,
+      "learning_rate": 1.5512665546467008e-05,
+      "loss": 1.7052,
+      "mean_token_accuracy": 0.7775381295109995,
+      "mean_token_accuracy_utility": 0.738915154353019,
+      "step": 1035
+    },
+    {
+      "epoch": 0.38475767665556787,
+      "grad_norm": 0.0,
+      "learning_rate": 1.5458662727828633e-05,
+      "loss": 1.5719,
+      "mean_token_accuracy": 0.7998991869356928,
+      "mean_token_accuracy_utility": 0.7457654351769949,
+      "step": 1040
+    },
+    {
+      "epoch": 0.3866074731779504,
+      "grad_norm": 0.0,
+      "learning_rate": 1.540443219118827e-05,
+      "loss": 1.4869,
+      "mean_token_accuracy": 0.7931973553564833,
+      "mean_token_accuracy_utility": 0.7643141253434373,
+      "step": 1045
+    },
+    {
+      "epoch": 0.38845726970033295,
+      "grad_norm": 0.0,
+      "learning_rate": 1.5349976198870974e-05,
+      "loss": 1.4944,
+      "mean_token_accuracy": 0.7939729524909142,
+      "mean_token_accuracy_utility": 0.7660942203791545,
+      "step": 1050
+    },
+    {
+      "epoch": 0.3903070662227155,
+      "grad_norm": 0.0,
+      "learning_rate": 1.529529702260709e-05,
+      "loss": 1.5705,
+      "mean_token_accuracy": 0.7916943562310303,
+      "mean_token_accuracy_utility": 0.7478591112502387,
+      "step": 1055
+    },
+    {
+      "epoch": 0.39215686274509803,
+      "grad_norm": 0.0,
+      "learning_rate": 1.5240396943437483e-05,
+      "loss": 1.5109,
+      "mean_token_accuracy": 0.78538433231733,
+      "mean_token_accuracy_utility": 0.7669347686798362,
+      "step": 1060
+    },
+    {
+      "epoch": 0.3940066592674806,
+      "grad_norm": 0.0,
+      "learning_rate": 1.5185278251618391e-05,
+      "loss": 1.5281,
+      "mean_token_accuracy": 0.7911985116177733,
+      "mean_token_accuracy_utility": 0.758761562553951,
+      "step": 1065
+    },
+    {
+      "epoch": 0.3958564557898631,
+      "grad_norm": 0.0,
+      "learning_rate": 1.5129943246525871e-05,
+      "loss": 1.5266,
+      "mean_token_accuracy": 0.7908321446332665,
+      "mean_token_accuracy_utility": 0.7579443936488716,
+      "step": 1070
+    },
+    {
+      "epoch": 0.39770625231224566,
+      "grad_norm": 0.0,
+      "learning_rate": 1.5074394236559871e-05,
+      "loss": 1.4784,
+      "mean_token_accuracy": 0.809769279221028,
+      "mean_token_accuracy_utility": 0.7535115538411642,
+      "step": 1075
+    },
+    {
+      "epoch": 0.3995560488346282,
+      "grad_norm": 0.0,
+      "learning_rate": 1.5018633539047955e-05,
+      "loss": 1.6635,
+      "mean_token_accuracy": 0.7891176442399309,
+      "mean_token_accuracy_utility": 0.7419446799056593,
+      "step": 1080
+    },
+    {
+      "epoch": 0.40140584535701074,
+      "grad_norm": 0.0,
+      "learning_rate": 1.4962663480148606e-05,
+      "loss": 1.4881,
+      "mean_token_accuracy": 0.8023755464201662,
+      "mean_token_accuracy_utility": 0.7588628524428116,
+      "step": 1085
+    },
+    {
+      "epoch": 0.4032556418793933,
+      "grad_norm": 0.0,
+      "learning_rate": 1.4906486394754198e-05,
+      "loss": 1.4846,
+      "mean_token_accuracy": 0.8061914789966211,
+      "mean_token_accuracy_utility": 0.7553826289580742,
+      "step": 1090
+    },
+    {
+      "epoch": 0.4051054384017758,
+      "grad_norm": 0.0,
+      "learning_rate": 1.4850104626393598e-05,
+      "loss": 1.5268,
+      "mean_token_accuracy": 0.8024963999377321,
+      "mean_token_accuracy_utility": 0.7503925887886785,
+      "step": 1095
+    },
+    {
+      "epoch": 0.40695523492415836,
+      "grad_norm": 0.0,
+      "learning_rate": 1.4793520527134381e-05,
+      "loss": 1.5241,
+      "mean_token_accuracy": 0.8006628946009933,
+      "mean_token_accuracy_utility": 0.7544360140603183,
+      "step": 1100
+    },
+    {
+      "epoch": 0.40695523492415836,
+      "eval_mean_token_accuracy": 0.7876186858500646,
+      "eval_reasoning_loss": 0.6703318357467651,
+      "eval_reasoning_runtime": 89.8681,
+      "eval_reasoning_samples_per_second": 1.435,
+      "eval_reasoning_steps_per_second": 0.367,
+      "step": 1100
+    },
+    {
+      "epoch": 0.40695523492415836,
+      "eval_mean_token_accuracy": 0.7407484475363784,
+      "eval_utility_loss": 0.9453916549682617,
+      "eval_utility_runtime": 33.8322,
+      "eval_utility_samples_per_second": 1.448,
+      "eval_utility_steps_per_second": 0.384,
+      "step": 1100
+    },
+    {
+      "epoch": 0.4088050314465409,
+      "grad_norm": 0.0,
+      "learning_rate": 1.473673645748473e-05,
+      "loss": 1.4913,
+      "mean_token_accuracy": 0.7946363019755498,
+      "mean_token_accuracy_utility": 0.7623613771466181,
+      "step": 1105
+    },
+    {
+      "epoch": 0.41065482796892344,
+      "grad_norm": 0.0,
+      "learning_rate": 1.4679754786294954e-05,
+      "loss": 1.524,
+      "mean_token_accuracy": 0.7983827005111148,
+      "mean_token_accuracy_utility": 0.7554008470944609,
+      "step": 1110
+    },
+    {
+      "epoch": 0.412504624491306,
+      "grad_norm": 0.0,
+      "learning_rate": 1.4622577890658668e-05,
+      "loss": 1.4827,
+      "mean_token_accuracy": 0.8010155094377532,
+      "mean_token_accuracy_utility": 0.7607405379270642,
+      "step": 1115
+    },
+    {
+      "epoch": 0.41435442101368847,
+      "grad_norm": 0.0,
+      "learning_rate": 1.4565208155813626e-05,
+      "loss": 1.4834,
+      "mean_token_accuracy": 0.8022190051250799,
+      "mean_token_accuracy_utility": 0.7609878805738811,
+      "step": 1120
+    },
+    {
+      "epoch": 0.416204217536071,
+      "grad_norm": 0.0,
+      "learning_rate": 1.4507647975042221e-05,
+      "loss": 1.4894,
+      "mean_token_accuracy": 0.7945976252867061,
+      "mean_token_accuracy_utility": 0.7628875292989825,
+      "step": 1125
+    },
+    {
+      "epoch": 0.41805401405845355,
+      "grad_norm": 0.0,
+      "learning_rate": 1.4449899749571643e-05,
+      "loss": 1.3761,
+      "mean_token_accuracy": 0.7985531334973656,
+      "mean_token_accuracy_utility": 0.7903862294855586,
+      "step": 1130
+    },
+    {
+      "epoch": 0.4199038105808361,
+      "grad_norm": 0.0,
+      "learning_rate": 1.4391965888473705e-05,
+      "loss": 1.425,
+      "mean_token_accuracy": 0.8021411697728883,
+      "mean_token_accuracy_utility": 0.771525144535403,
+      "step": 1135
+    },
+    {
+      "epoch": 0.42175360710321863,
+      "grad_norm": 0.0,
+      "learning_rate": 1.4333848808564353e-05,
+      "loss": 1.4354,
+      "mean_token_accuracy": 0.7927916945399612,
+      "mean_token_accuracy_utility": 0.7726091451062768,
+      "step": 1140
+    },
+    {
+      "epoch": 0.42360340362560117,
+      "grad_norm": 0.0,
+      "learning_rate": 1.4275550934302822e-05,
+      "loss": 1.4976,
+      "mean_token_accuracy": 0.8045263184329295,
+      "mean_token_accuracy_utility": 0.7554516610430387,
+      "step": 1145
+    },
+    {
+      "epoch": 0.4254532001479837,
+      "grad_norm": 0.0,
+      "learning_rate": 1.4217074697690532e-05,
+      "loss": 1.5977,
+      "mean_token_accuracy": 0.7890858028710005,
+      "mean_token_accuracy_utility": 0.7457020175314886,
+      "step": 1150
+    },
+    {
+      "epoch": 0.42730299667036625,
+      "grad_norm": 0.0,
+      "learning_rate": 1.4158422538169596e-05,
+      "loss": 1.4735,
+      "mean_token_accuracy": 0.7922479822585428,
+      "mean_token_accuracy_utility": 0.7684627464212674,
+      "step": 1155
+    },
+    {
+      "epoch": 0.4291527931927488,
+      "grad_norm": 0.0,
+      "learning_rate": 1.4099596902521073e-05,
+      "loss": 1.4812,
+      "mean_token_accuracy": 0.7963013307874098,
+      "mean_token_accuracy_utility": 0.7610360598296635,
+      "step": 1160
+    },
+    {
+      "epoch": 0.43100258971513133,
+      "grad_norm": 0.0,
+      "learning_rate": 1.40406002447629e-05,
+      "loss": 1.4426,
+      "mean_token_accuracy": 0.7890354206189409,
+      "mean_token_accuracy_utility": 0.7719279055344669,
+      "step": 1165
+    },
+    {
+      "epoch": 0.4328523862375139,
+      "grad_norm": 0.0,
+      "learning_rate": 1.39814350260475e-05,
+      "loss": 1.4739,
+      "mean_token_accuracy": 0.7975903277665489,
+      "mean_token_accuracy_utility": 0.7599637858022397,
+      "step": 1170
+    },
+    {
+      "epoch": 0.4347021827598964,
+      "grad_norm": 0.0,
+      "learning_rate": 1.392210371455913e-05,
+      "loss": 1.4689,
+      "mean_token_accuracy": 0.7969269875040896,
+      "mean_token_accuracy_utility": 0.766528732840562,
+      "step": 1175
+    },
+    {
+      "epoch": 0.43655197928227896,
+      "grad_norm": 0.0,
+      "learning_rate": 1.3862608785410903e-05,
+      "loss": 1.5128,
+      "mean_token_accuracy": 0.8031015703573232,
+      "mean_token_accuracy_utility": 0.7525186832276506,
+      "step": 1180
+    },
+    {
+      "epoch": 0.4384017758046615,
+      "grad_norm": 0.0,
+      "learning_rate": 1.3802952720541543e-05,
+      "loss": 1.3895,
+      "mean_token_accuracy": 0.8101776636453385,
+      "mean_token_accuracy_utility": 0.7756553273278691,
+      "step": 1185
+    },
+    {
+      "epoch": 0.44025157232704404,
+      "grad_norm": 0.0,
+      "learning_rate": 1.3743138008611838e-05,
+      "loss": 1.4517,
+      "mean_token_accuracy": 0.8006501923825307,
+      "mean_token_accuracy_utility": 0.7674959411689247,
+      "step": 1190
+    },
+    {
+      "epoch": 0.4421013688494266,
+      "grad_norm": 0.0,
+      "learning_rate": 1.3683167144900833e-05,
+      "loss": 1.4696,
+      "mean_token_accuracy": 0.802009996794218,
+      "mean_token_accuracy_utility": 0.7612202954679863,
+      "step": 1195
+    },
+    {
+      "epoch": 0.4439511653718091,
+      "grad_norm": 0.0,
+      "learning_rate": 1.3623042631201713e-05,
+      "loss": 1.4639,
+      "mean_token_accuracy": 0.7866717802637719,
+      "mean_token_accuracy_utility": 0.7740715671138635,
+      "step": 1200
+    },
+    {
+      "epoch": 0.4439511653718091,
+      "eval_mean_token_accuracy": 0.7879259108126372,
+      "eval_reasoning_loss": 0.6686920523643494,
+      "eval_reasoning_runtime": 89.8699,
+      "eval_reasoning_samples_per_second": 1.435,
+      "eval_reasoning_steps_per_second": 0.367,
+      "step": 1200
+    },
+    {
+      "epoch": 0.4439511653718091,
+      "eval_mean_token_accuracy": 0.7409041422997878,
+      "eval_utility_loss": 0.9458145499229431,
+      "eval_utility_runtime": 33.8228,
+      "eval_utility_samples_per_second": 1.449,
+      "eval_utility_steps_per_second": 0.384,
+      "step": 1200
+    },
+    {
+      "epoch": 0.44580096189419166,
+      "grad_norm": 0.0,
+      "learning_rate": 1.3562766975717468e-05,
+      "loss": 1.4703,
+      "mean_token_accuracy": 0.7969164004942453,
+      "mean_token_accuracy_utility": 0.764365573881155,
+      "step": 1205
+    },
+    {
+      "epoch": 0.4476507584165742,
+      "grad_norm": 0.0,
+      "learning_rate": 1.3502342692956229e-05,
+      "loss": 1.3888,
+      "mean_token_accuracy": 0.8168788397930051,
+      "mean_token_accuracy_utility": 0.7668393046101095,
+      "step": 1210
+    },
+    {
+      "epoch": 0.44950055493895674,
+      "grad_norm": 0.0,
+      "learning_rate": 1.3441772303626387e-05,
+      "loss": 1.5214,
+      "mean_token_accuracy": 0.7903327980011445,
+      "mean_token_accuracy_utility": 0.7593918845943162,
+      "step": 1215
+    },
+    {
+      "epoch": 0.4513503514613392,
+      "grad_norm": 0.0,
+      "learning_rate": 1.3381058334531436e-05,
+      "loss": 1.4102,
+      "mean_token_accuracy": 0.8103997315711473,
+      "mean_token_accuracy_utility": 0.7660602836083255,
+      "step": 1220
+    },
+    {
+      "epoch": 0.45320014798372177,
+      "grad_norm": 0.0,
+      "learning_rate": 1.3320203318464552e-05,
+      "loss": 1.4795,
+      "mean_token_accuracy": 0.7945650797540507,
+      "mean_token_accuracy_utility": 0.7655195990766439,
+      "step": 1225
+    },
+    {
+      "epoch": 0.4550499445061043,
+      "grad_norm": 0.0,
+      "learning_rate": 1.3259209794102952e-05,
+      "loss": 1.3856,
+      "mean_token_accuracy": 0.7997550423920581,
+      "mean_token_accuracy_utility": 0.7839200613250028,
+      "step": 1230
+    },
+    {
+      "epoch": 0.45689974102848685,
+      "grad_norm": 0.0,
+      "learning_rate": 1.319808030590197e-05,
+      "loss": 1.5505,
+      "mean_token_accuracy": 0.8022905272423216,
+      "mean_token_accuracy_utility": 0.7496539437319281,
+      "step": 1235
+    },
+    {
+      "epoch": 0.4587495375508694,
+      "grad_norm": 0.0,
+      "learning_rate": 1.3136817403988918e-05,
+      "loss": 1.4135,
+      "mean_token_accuracy": 0.7915171919602922,
+      "mean_token_accuracy_utility": 0.7798013581275742,
+      "step": 1240
+    },
+    {
+      "epoch": 0.46059933407325193,
+      "grad_norm": 0.0,
+      "learning_rate": 1.3075423644056699e-05,
+      "loss": 1.5585,
+      "mean_token_accuracy": 0.7818016030467925,
+      "mean_token_accuracy_utility": 0.7565209320375926,
+      "step": 1245
+    },
+    {
+      "epoch": 0.46244913059563447,
+      "grad_norm": 0.0,
+      "learning_rate": 1.3013901587257207e-05,
+      "loss": 1.4271,
+      "mean_token_accuracy": 0.7958682501688903,
+      "mean_token_accuracy_utility": 0.7775921392523903,
+      "step": 1250
+    },
+    {
+      "epoch": 0.464298927118017,
+      "grad_norm": 0.0,
+      "learning_rate": 1.2952253800094467e-05,
+      "loss": 1.5214,
+      "mean_token_accuracy": 0.7937254228153738,
+      "mean_token_accuracy_utility": 0.7563848396476895,
+      "step": 1255
+    },
+    {
+      "epoch": 0.46614872364039955,
+      "grad_norm": 0.0,
+      "learning_rate": 1.2890482854317565e-05,
+      "loss": 1.6098,
+      "mean_token_accuracy": 0.776730263541547,
+      "mean_token_accuracy_utility": 0.7538273568975826,
+      "step": 1260
+    },
+    {
+      "epoch": 0.4679985201627821,
+      "grad_norm": 0.0,
+      "learning_rate": 1.2828591326813382e-05,
+      "loss": 1.4813,
+      "mean_token_accuracy": 0.7919601166304241,
+      "mean_token_accuracy_utility": 0.7693110577377593,
+      "step": 1265
+    },
+    {
+      "epoch": 0.46984831668516464,
+      "grad_norm": 0.0,
+      "learning_rate": 1.2766581799499078e-05,
+      "loss": 1.4664,
+      "mean_token_accuracy": 0.7964648170375954,
+      "mean_token_accuracy_utility": 0.7722150136073104,
+      "step": 1270
+    },
+    {
+      "epoch": 0.4716981132075472,
+      "grad_norm": 0.0,
+      "learning_rate": 1.2704456859214397e-05,
+      "loss": 1.6295,
+      "mean_token_accuracy": 0.7839941044333202,
+      "mean_token_accuracy_utility": 0.7434948840433725,
+      "step": 1275
+    },
+    {
+      "epoch": 0.4735479097299297,
+      "grad_norm": 0.0,
+      "learning_rate": 1.264221909761374e-05,
+      "loss": 1.4394,
+      "mean_token_accuracy": 0.8133503334809541,
+      "mean_token_accuracy_utility": 0.7614829767743203,
+      "step": 1280
+    },
+    {
+      "epoch": 0.47539770625231226,
+      "grad_norm": 0.0,
+      "learning_rate": 1.2579871111058042e-05,
+      "loss": 1.4952,
+      "mean_token_accuracy": 0.7867893788395695,
+      "mean_token_accuracy_utility": 0.7663213504070332,
+      "step": 1285
+    },
+    {
+      "epoch": 0.4772475027746948,
+      "grad_norm": 0.0,
+      "learning_rate": 1.251741550050649e-05,
+      "loss": 1.3792,
+      "mean_token_accuracy": 0.8064460224960799,
+      "mean_token_accuracy_utility": 0.7831586043619005,
+      "step": 1290
+    },
+    {
+      "epoch": 0.47909729929707734,
+      "grad_norm": 0.0,
+      "learning_rate": 1.2454854871407993e-05,
+      "loss": 1.5266,
+      "mean_token_accuracy": 0.7927932898090788,
+      "mean_token_accuracy_utility": 0.7597310875022981,
+      "step": 1295
+    },
+    {
+      "epoch": 0.4809470958194599,
+      "grad_norm": 0.0,
+      "learning_rate": 1.2392191833592495e-05,
+      "loss": 1.4731,
+      "mean_token_accuracy": 0.8034761925857122,
+      "mean_token_accuracy_utility": 0.7604243740131206,
+      "step": 1300
+    },
+    {
+      "epoch": 0.4809470958194599,
+      "eval_mean_token_accuracy": 0.7881909360767747,
+      "eval_reasoning_loss": 0.6666797995567322,
+      "eval_reasoning_runtime": 89.8841,
+      "eval_reasoning_samples_per_second": 1.435,
+      "eval_reasoning_steps_per_second": 0.367,
+      "step": 1300
+    },
+    {
+      "epoch": 0.4809470958194599,
+      "eval_mean_token_accuracy": 0.7412586607786056,
+      "eval_utility_loss": 0.9454450011253357,
+      "eval_utility_runtime": 33.8482,
+      "eval_utility_samples_per_second": 1.448,
+      "eval_utility_steps_per_second": 0.384,
+      "step": 1300
+    },
+    {
+      "epoch": 0.4827968923418424,
+      "grad_norm": 0.0,
+      "learning_rate": 1.2329429001162114e-05,
+      "loss": 1.4369,
+      "mean_token_accuracy": 0.802257655374266,
+      "mean_token_accuracy_utility": 0.7682160795329336,
+      "step": 1305
+    },
+    {
+      "epoch": 0.48464668886422496,
+      "grad_norm": 0.0,
+      "learning_rate": 1.226656899238207e-05,
+      "loss": 1.4064,
+      "mean_token_accuracy": 0.8070532787100744,
+      "mean_token_accuracy_utility": 0.7711903955333789,
+      "step": 1310
+    },
+    {
+      "epoch": 0.48649648538660745,
+      "grad_norm": 0.0,
+      "learning_rate": 1.2203614429571475e-05,
+      "loss": 1.4764,
+      "mean_token_accuracy": 0.7864109231734571,
+      "mean_token_accuracy_utility": 0.7731644510368237,
+      "step": 1315
+    },
+    {
+      "epoch": 0.48834628190899,
+      "grad_norm": 0.0,
+      "learning_rate": 1.214056793899394e-05,
+      "loss": 1.3979,
+      "mean_token_accuracy": 0.8002562210819575,
+      "mean_token_accuracy_utility": 0.7769540862341885,
+      "step": 1320
+    },
+    {
+      "epoch": 0.49019607843137253,
+      "grad_norm": 0.0,
+      "learning_rate": 1.207743215074801e-05,
+      "loss": 1.4458,
+      "mean_token_accuracy": 0.8038714839609324,
+      "mean_token_accuracy_utility": 0.7649823026023415,
+      "step": 1325
+    },
+    {
+      "epoch": 0.49204587495375507,
+      "grad_norm": 0.0,
+      "learning_rate": 1.2014209698657424e-05,
+      "loss": 1.4158,
+      "mean_token_accuracy": 0.7942955903862579,
+      "mean_token_accuracy_utility": 0.7767080171671471,
+      "step": 1330
+    },
+    {
+      "epoch": 0.4938956714761376,
+      "grad_norm": 0.0,
+      "learning_rate": 1.1950903220161286e-05,
+      "loss": 1.533,
+      "mean_token_accuracy": 0.7940201241289818,
+      "mean_token_accuracy_utility": 0.7504826966786995,
+      "step": 1335
+    },
+    {
+      "epoch": 0.49574546799852015,
+      "grad_norm": 0.0,
+      "learning_rate": 1.1887515356203994e-05,
+      "loss": 1.5212,
+      "mean_token_accuracy": 0.7924755760032018,
+      "mean_token_accuracy_utility": 0.7604314409752628,
+      "step": 1340
+    },
+    {
+      "epoch": 0.4975952645209027,
+      "grad_norm": 0.0,
+      "learning_rate": 1.1824048751125101e-05,
+      "loss": 1.4734,
+      "mean_token_accuracy": 0.7902913681531248,
+      "mean_token_accuracy_utility": 0.7660344346345421,
+      "step": 1345
+    },
+    {
+      "epoch": 0.49944506104328523,
+      "grad_norm": 0.0,
+      "learning_rate": 1.1760506052548984e-05,
+      "loss": 1.4983,
+      "mean_token_accuracy": 0.7910196872189412,
+      "mean_token_accuracy_utility": 0.7675096874668521,
+      "step": 1350
+    },
+    {
+      "epoch": 0.5012948575656678,
+      "grad_norm": 0.0,
+      "learning_rate": 1.1696889911274394e-05,
+      "loss": 1.4569,
+      "mean_token_accuracy": 0.8005747903165548,
+      "mean_token_accuracy_utility": 0.7648615554488213,
+      "step": 1355
+    },
+    {
+      "epoch": 0.5031446540880503,
+      "grad_norm": 0.0,
+      "learning_rate": 1.1633202981163882e-05,
+      "loss": 1.4921,
+      "mean_token_accuracy": 0.7900144110423492,
+      "mean_token_accuracy_utility": 0.7700804303801341,
+      "step": 1360
+    },
+    {
+      "epoch": 0.5049944506104328,
+      "grad_norm": 0.0,
+      "learning_rate": 1.1569447919033086e-05,
+      "loss": 1.4117,
+      "mean_token_accuracy": 0.7905631288690286,
+      "mean_token_accuracy_utility": 0.7813303091959046,
+      "step": 1365
+    },
+    {
+      "epoch": 0.5068442471328154,
+      "grad_norm": 0.0,
+      "learning_rate": 1.1505627384539888e-05,
+      "loss": 1.5166,
+      "mean_token_accuracy": 0.7967665728364299,
+      "mean_token_accuracy_utility": 0.7552322974287076,
+      "step": 1370
+    },
+    {
+      "epoch": 0.5086940436551979,
+      "grad_norm": 0.0,
+      "learning_rate": 1.1441744040073469e-05,
+      "loss": 1.4736,
+      "mean_token_accuracy": 0.7903410505497324,
+      "mean_token_accuracy_utility": 0.7678365562801917,
+      "step": 1375
+    },
+    {
+      "epoch": 0.5105438401775805,
+      "grad_norm": 0.0,
+      "learning_rate": 1.1377800550643246e-05,
+      "loss": 1.5149,
+      "mean_token_accuracy": 0.7969502210740911,
+      "mean_token_accuracy_utility": 0.7581058159745326,
+      "step": 1380
+    },
+    {
+      "epoch": 0.512393636699963,
+      "grad_norm": 0.0,
+      "learning_rate": 1.1313799583767693e-05,
+      "loss": 1.3884,
+      "mean_token_accuracy": 0.7985772693688294,
+      "mean_token_accuracy_utility": 0.7806748182087777,
+      "step": 1385
+    },
+    {
+      "epoch": 0.5142434332223456,
+      "grad_norm": 0.0,
+      "learning_rate": 1.124974380936305e-05,
+      "loss": 1.4761,
+      "mean_token_accuracy": 0.793807908957953,
+      "mean_token_accuracy_utility": 0.7642041118402763,
+      "step": 1390
+    },
+    {
+      "epoch": 0.516093229744728,
+      "grad_norm": 0.0,
+      "learning_rate": 1.1185635899631963e-05,
+      "loss": 1.5259,
+      "mean_token_accuracy": 0.7948374519255568,
+      "mean_token_accuracy_utility": 0.760306502270094,
+      "step": 1395
+    },
+    {
+      "epoch": 0.5179430262671106,
+      "grad_norm": 0.0,
+      "learning_rate": 1.1121478528951994e-05,
+      "loss": 1.4255,
+      "mean_token_accuracy": 0.8019869715758359,
+      "mean_token_accuracy_utility": 0.7757139743985751,
+      "step": 1400
+    },
+    {
+      "epoch": 0.5179430262671106,
+      "eval_mean_token_accuracy": 0.7884291443610675,
+      "eval_reasoning_loss": 0.6651748418807983,
+      "eval_reasoning_runtime": 89.921,
+      "eval_reasoning_samples_per_second": 1.435,
+      "eval_reasoning_steps_per_second": 0.367,
+      "step": 1400
+    },
+    {
+      "epoch": 0.5179430262671106,
+      "eval_mean_token_accuracy": 0.7407914043674722,
+      "eval_utility_loss": 0.9454284310340881,
+      "eval_utility_runtime": 33.8406,
+      "eval_utility_samples_per_second": 1.448,
+      "eval_utility_steps_per_second": 0.384,
+      "step": 1400
+    },
+    {
+      "epoch": 0.5197928227894931,
+      "grad_norm": 0.0,
+      "learning_rate": 1.1057274373764056e-05,
+      "loss": 1.4517,
+      "mean_token_accuracy": 0.7981587706441997,
+      "mean_token_accuracy_utility": 0.7697954707371983,
+      "step": 1405
+    },
+    {
+      "epoch": 0.5216426193118757,
+      "grad_norm": 0.0,
+      "learning_rate": 1.0993026112460765e-05,
+      "loss": 1.4725,
+      "mean_token_accuracy": 0.7989297031752693,
+      "mean_token_accuracy_utility": 0.7632571901739025,
+      "step": 1410
+    },
+    {
+      "epoch": 0.5234924158342582,
+      "grad_norm": 0.0,
+      "learning_rate": 1.0928736425274702e-05,
+      "loss": 1.4666,
+      "mean_token_accuracy": 0.7983556265833286,
+      "mean_token_accuracy_utility": 0.7656110742491901,
+      "step": 1415
+    },
+    {
+      "epoch": 0.5253422123566408,
+      "grad_norm": 0.0,
+      "learning_rate": 1.0864407994166609e-05,
+      "loss": 1.4867,
+      "mean_token_accuracy": 0.7976140625490723,
+      "mean_token_accuracy_utility": 0.7658263029860184,
+      "step": 1420
+    },
+    {
+      "epoch": 0.5271920088790233,
+      "grad_norm": 0.0,
+      "learning_rate": 1.0800043502713486e-05,
+      "loss": 1.4221,
+      "mean_token_accuracy": 0.7997529305228654,
+      "mean_token_accuracy_utility": 0.7701824882124282,
+      "step": 1425
+    },
+    {
+      "epoch": 0.5290418054014059,
+      "grad_norm": 0.0,
+      "learning_rate": 1.0735645635996676e-05,
+      "loss": 1.5483,
+      "mean_token_accuracy": 0.7920482348480815,
+      "mean_token_accuracy_utility": 0.7501540566472471,
+      "step": 1430
+    },
+    {
+      "epoch": 0.5308916019237884,
+      "grad_norm": 0.0,
+      "learning_rate": 1.0671217080489816e-05,
+      "loss": 1.4296,
+      "mean_token_accuracy": 0.8033824277775177,
+      "mean_token_accuracy_utility": 0.768735575489728,
+      "step": 1435
+    },
+    {
+      "epoch": 0.532741398446171,
+      "grad_norm": 0.0,
+      "learning_rate": 1.0606760523946789e-05,
+      "loss": 1.4097,
+      "mean_token_accuracy": 0.8113958244622406,
+      "mean_token_accuracy_utility": 0.7679295988359269,
+      "step": 1440
+    },
+    {
+      "epoch": 0.5345911949685535,
+      "grad_norm": 0.0,
+      "learning_rate": 1.0542278655289588e-05,
+      "loss": 1.4599,
+      "mean_token_accuracy": 0.7997273599445293,
+      "mean_token_accuracy_utility": 0.7641978307367348,
+      "step": 1445
+    },
+    {
+      "epoch": 0.536440991490936,
+      "grad_norm": 0.0,
+      "learning_rate": 1.047777416449615e-05,
+      "loss": 1.5324,
+      "mean_token_accuracy": 0.8006738452567326,
+      "mean_token_accuracy_utility": 0.7520984634772152,
+      "step": 1450
+    },
+    {
+      "epoch": 0.5382907880133185,
+      "grad_norm": 0.0,
+      "learning_rate": 1.0413249742488132e-05,
+      "loss": 1.4799,
+      "mean_token_accuracy": 0.7983709869861861,
+      "mean_token_accuracy_utility": 0.7603470377735433,
+      "step": 1455
+    },
+    {
+      "epoch": 0.540140584535701,
+      "grad_norm": 0.0,
+      "learning_rate": 1.034870808101866e-05,
+      "loss": 1.4373,
+      "mean_token_accuracy": 0.7986303268061958,
+      "mean_token_accuracy_utility": 0.7713622746962407,
+      "step": 1460
+    },
+    {
+      "epoch": 0.5419903810580836,
+      "grad_norm": 0.0,
+      "learning_rate": 1.0284151872560042e-05,
+      "loss": 1.5574,
+      "mean_token_accuracy": 0.7975773166964892,
+      "mean_token_accuracy_utility": 0.7566143640053309,
+      "step": 1465
+    },
+    {
+      "epoch": 0.5438401775804661,
+      "grad_norm": 0.0,
+      "learning_rate": 1.0219583810191422e-05,
+      "loss": 1.4725,
+      "mean_token_accuracy": 0.8011400064681833,
+      "mean_token_accuracy_utility": 0.7651626654847309,
+      "step": 1470
+    },
+    {
+      "epoch": 0.5456899741028487,
+      "grad_norm": 0.0,
+      "learning_rate": 1.0155006587486468e-05,
+      "loss": 1.5269,
+      "mean_token_accuracy": 0.7863179243876416,
+      "mean_token_accuracy_utility": 0.7605489845475167,
+      "step": 1475
+    },
+    {
+      "epoch": 0.5475397706252312,
+      "grad_norm": 0.0,
+      "learning_rate": 1.0090422898400985e-05,
+      "loss": 1.5378,
+      "mean_token_accuracy": 0.7990884330917826,
+      "mean_token_accuracy_utility": 0.7522383259719694,
+      "step": 1480
+    },
+    {
+      "epoch": 0.5493895671476138,
+      "grad_norm": 0.0,
+      "learning_rate": 1.0025835437160523e-05,
+      "loss": 1.671,
+      "mean_token_accuracy": 0.790815640034279,
+      "mean_token_accuracy_utility": 0.7362262745265992,
+      "step": 1485
+    },
+    {
+      "epoch": 0.5512393636699963,
+      "grad_norm": 0.0,
+      "learning_rate": 9.96124689814801e-06,
+      "loss": 1.4237,
+      "mean_token_accuracy": 0.7886717390988448,
+      "mean_token_accuracy_utility": 0.782383685828936,
+      "step": 1490
+    },
+    {
+      "epoch": 0.5530891601923789,
+      "grad_norm": 0.0,
+      "learning_rate": 9.89665997579133e-06,
+      "loss": 1.4505,
+      "mean_token_accuracy": 0.7936965138257361,
+      "mean_token_accuracy_utility": 0.7725070051324321,
+      "step": 1495
+    },
+    {
+      "epoch": 0.5549389567147613,
+      "grad_norm": 0.0,
+      "learning_rate": 9.832077364450915e-06,
+      "loss": 1.4188,
+      "mean_token_accuracy": 0.8137404036326373,
+      "mean_token_accuracy_utility": 0.7639404592889536,
+      "step": 1500
+    },
+    {
+      "epoch": 0.5549389567147613,
+      "eval_mean_token_accuracy": 0.7888067444502695,
+      "eval_reasoning_loss": 0.6640897989273071,
+      "eval_reasoning_runtime": 89.9362,
+      "eval_reasoning_samples_per_second": 1.434,
+      "eval_reasoning_steps_per_second": 0.367,
+      "step": 1500
+    },
+    {
+      "epoch": 0.5549389567147613,
+      "eval_mean_token_accuracy": 0.7408938582385841,
+      "eval_utility_loss": 0.9461740255355835,
+      "eval_utility_runtime": 33.8455,
+      "eval_utility_samples_per_second": 1.448,
+      "eval_utility_steps_per_second": 0.384,
+      "step": 1500
+    },
+    {
+      "epoch": 0.5567887532371439,
+      "grad_norm": 0.0,
+      "learning_rate": 9.767501758307376e-06,
+      "loss": 1.4185,
+      "mean_token_accuracy": 0.8035864588653734,
+      "mean_token_accuracy_utility": 0.7719513652206769,
+      "step": 1505
+    },
+    {
+      "epoch": 0.5586385497595264,
+      "grad_norm": 0.0,
+      "learning_rate": 9.70293585124907e-06,
+      "loss": 1.4136,
+      "mean_token_accuracy": 0.8089429184621121,
+      "mean_token_accuracy_utility": 0.7691278482402658,
+      "step": 1510
+    },
+    {
+      "epoch": 0.560488346281909,
+      "grad_norm": 0.0,
+      "learning_rate": 9.638382336759749e-06,
+      "loss": 1.4065,
+      "mean_token_accuracy": 0.7946615794501248,
+      "mean_token_accuracy_utility": 0.7822854973587642,
+      "step": 1515
+    },
+    {
+      "epoch": 0.5623381428042915,
+      "grad_norm": 0.0,
+      "learning_rate": 9.57384390780618e-06,
+      "loss": 1.4485,
+      "mean_token_accuracy": 0.7994958711366259,
+      "mean_token_accuracy_utility": 0.7667057263360888,
+      "step": 1520
+    },
+    {
+      "epoch": 0.5641879393266741,
+      "grad_norm": 0.0,
+      "learning_rate": 9.50932325672582e-06,
+      "loss": 1.4892,
+      "mean_token_accuracy": 0.7882357584963915,
+      "mean_token_accuracy_utility": 0.7648433392091636,
+      "step": 1525
+    },
+    {
+      "epoch": 0.5660377358490566,
+      "grad_norm": 0.0,
+      "learning_rate": 9.444823075114481e-06,
+      "loss": 1.446,
+      "mean_token_accuracy": 0.7998468103714088,
+      "mean_token_accuracy_utility": 0.7681386505359604,
+      "step": 1530
+    },
+    {
+      "epoch": 0.5678875323714392,
+      "grad_norm": 0.0,
+      "learning_rate": 9.380346053714055e-06,
+      "loss": 1.3889,
+      "mean_token_accuracy": 0.8089379739055185,
+      "mean_token_accuracy_utility": 0.7716691077571857,
+      "step": 1535
+    },
+    {
+      "epoch": 0.5697373288938217,
+      "grad_norm": 0.0,
+      "learning_rate": 9.315894882300269e-06,
+      "loss": 1.373,
+      "mean_token_accuracy": 0.8177979752847383,
+      "mean_token_accuracy_utility": 0.7727258943649378,
+      "step": 1540
+    },
+    {
+      "epoch": 0.5715871254162043,
+      "grad_norm": 0.0,
+      "learning_rate": 9.251472249570458e-06,
+      "loss": 1.4044,
+      "mean_token_accuracy": 0.7964566521895206,
+      "mean_token_accuracy_utility": 0.7795725154261552,
+      "step": 1545
+    },
+    {
+      "epoch": 0.5734369219385868,
+      "grad_norm": 0.0,
+      "learning_rate": 9.187080843031436e-06,
+      "loss": 1.426,
+      "mean_token_accuracy": 0.8044486027610906,
+      "mean_token_accuracy_utility": 0.7703032217633237,
+      "step": 1550
+    },
+    {
+      "epoch": 0.5752867184609693,
+      "grad_norm": 0.0,
+      "learning_rate": 9.12272334888734e-06,
+      "loss": 1.4886,
+      "mean_token_accuracy": 0.7905742744755002,
+      "mean_token_accuracy_utility": 0.7656434613501935,
+      "step": 1555
+    },
+    {
+      "epoch": 0.5771365149833518,
+      "grad_norm": 0.0,
+      "learning_rate": 9.058402451927599e-06,
+      "loss": 1.4292,
+      "mean_token_accuracy": 0.8073844017751608,
+      "mean_token_accuracy_utility": 0.7673141176867363,
+      "step": 1560
+    },
+    {
+      "epoch": 0.5789863115057343,
+      "grad_norm": 0.0,
+      "learning_rate": 8.99412083541492e-06,
+      "loss": 1.3778,
+      "mean_token_accuracy": 0.8023886939555986,
+      "mean_token_accuracy_utility": 0.7821687180004189,
+      "step": 1565
+    },
+    {
+      "epoch": 0.5808361080281169,
+      "grad_norm": 0.0,
+      "learning_rate": 8.929881180973365e-06,
+      "loss": 1.435,
+      "mean_token_accuracy": 0.792943780808643,
+      "mean_token_accuracy_utility": 0.7763741810826348,
+      "step": 1570
+    },
+    {
+      "epoch": 0.5826859045504994,
+      "grad_norm": 0.0,
+      "learning_rate": 8.865686168476458e-06,
+      "loss": 1.5048,
+      "mean_token_accuracy": 0.7973622799293358,
+      "mean_token_accuracy_utility": 0.75906594310153,
+      "step": 1575
+    },
+    {
+      "epoch": 0.584535701072882,
+      "grad_norm": 0.0,
+      "learning_rate": 8.801538475935413e-06,
+      "loss": 1.4225,
+      "mean_token_accuracy": 0.8051300362993757,
+      "mean_token_accuracy_utility": 0.7689461921368711,
+      "step": 1580
+    },
+    {
+      "epoch": 0.5863854975952645,
+      "grad_norm": 0.0,
+      "learning_rate": 8.737440779387398e-06,
+      "loss": 1.4652,
+      "mean_token_accuracy": 0.797523109303803,
+      "mean_token_accuracy_utility": 0.7672406849856225,
+      "step": 1585
+    },
+    {
+      "epoch": 0.5882352941176471,
+      "grad_norm": 0.0,
+      "learning_rate": 8.673395752783924e-06,
+      "loss": 1.5243,
+      "mean_token_accuracy": 0.7953130733762459,
+      "mean_token_accuracy_utility": 0.7548385503777952,
+      "step": 1590
+    },
+    {
+      "epoch": 0.5900850906400296,
+      "grad_norm": 0.0,
+      "learning_rate": 8.609406067879258e-06,
+      "loss": 1.432,
+      "mean_token_accuracy": 0.8023134025857994,
+      "mean_token_accuracy_utility": 0.7690530497321236,
+      "step": 1595
+    },
+    {
+      "epoch": 0.5919348871624122,
+      "grad_norm": 0.0,
+      "learning_rate": 8.54547439411901e-06,
+      "loss": 1.4801,
+      "mean_token_accuracy": 0.7920564007951546,
+      "mean_token_accuracy_utility": 0.7705305890654016,
+      "step": 1600
+    },
+    {
+      "epoch": 0.5919348871624122,
+      "eval_mean_token_accuracy": 0.7889511234937255,
+      "eval_reasoning_loss": 0.6631338596343994,
+      "eval_reasoning_runtime": 89.9889,
+      "eval_reasoning_samples_per_second": 1.434,
+      "eval_reasoning_steps_per_second": 0.367,
+      "step": 1600
+    },
+    {
+      "epoch": 0.5919348871624122,
+      "eval_mean_token_accuracy": 0.7412115492044549,
+      "eval_utility_loss": 0.9460353255271912,
+      "eval_utility_runtime": 33.8746,
+      "eval_utility_samples_per_second": 1.447,
+      "eval_utility_steps_per_second": 0.384,
+      "step": 1600
+    },
+    {
+      "epoch": 0.5937846836847946,
+      "grad_norm": 0.0,
+      "learning_rate": 8.481603398528727e-06,
+      "loss": 1.4539,
+      "mean_token_accuracy": 0.8114288563954117,
+      "mean_token_accuracy_utility": 0.7601788696255147,
+      "step": 1605
+    },
+    {
+      "epoch": 0.5956344802071772,
+      "grad_norm": 0.0,
+      "learning_rate": 8.417795745602679e-06,
+      "loss": 1.4108,
+      "mean_token_accuracy": 0.7952787650747931,
+      "mean_token_accuracy_utility": 0.7792657532331443,
+      "step": 1610
+    },
+    {
+      "epoch": 0.5974842767295597,
+      "grad_norm": 0.0,
+      "learning_rate": 8.35405409719266e-06,
+      "loss": 1.4751,
+      "mean_token_accuracy": 0.7918483028140966,
+      "mean_token_accuracy_utility": 0.7641662416926822,
+      "step": 1615
+    },
+    {
+      "epoch": 0.5993340732519423,
+      "grad_norm": 0.0,
+      "learning_rate": 8.290381112396989e-06,
+      "loss": 1.4118,
+      "mean_token_accuracy": 0.804444056462352,
+      "mean_token_accuracy_utility": 0.7727990065224815,
+      "step": 1620
+    },
+    {
+      "epoch": 0.6011838697743248,
+      "grad_norm": 0.0,
+      "learning_rate": 8.226779447449538e-06,
+      "loss": 1.3957,
+      "mean_token_accuracy": 0.8064870454619321,
+      "mean_token_accuracy_utility": 0.7754858635437462,
+      "step": 1625
+    },
+    {
+      "epoch": 0.6030336662967074,
+      "grad_norm": 0.0,
+      "learning_rate": 8.163251755608954e-06,
+      "loss": 1.4934,
+      "mean_token_accuracy": 0.7995332592188626,
+      "mean_token_accuracy_utility": 0.7593364533160478,
+      "step": 1630
+    },
+    {
+      "epoch": 0.6048834628190899,
+      "grad_norm": 0.0,
+      "learning_rate": 8.099800687047958e-06,
+      "loss": 1.4381,
+      "mean_token_accuracy": 0.8017799165992889,
+      "mean_token_accuracy_utility": 0.7696555476604915,
+      "step": 1635
+    },
+    {
+      "epoch": 0.6067332593414725,
+      "grad_norm": 0.0,
+      "learning_rate": 8.036428888742789e-06,
+      "loss": 1.4115,
+      "mean_token_accuracy": 0.8001610125615075,
+      "mean_token_accuracy_utility": 0.7750077905489927,
+      "step": 1640
+    },
+    {
+      "epoch": 0.608583055863855,
+      "grad_norm": 0.0,
+      "learning_rate": 7.973139004362794e-06,
+      "loss": 1.4568,
+      "mean_token_accuracy": 0.7974479688553353,
+      "mean_token_accuracy_utility": 0.7737382720000607,
+      "step": 1645
+    },
+    {
+      "epoch": 0.6104328523862376,
+      "grad_norm": 0.0,
+      "learning_rate": 7.909933674160118e-06,
+      "loss": 1.4594,
+      "mean_token_accuracy": 0.7918142430263156,
+      "mean_token_accuracy_utility": 0.7692910859322422,
+      "step": 1650
+    },
+    {
+      "epoch": 0.61228264890862,
+      "grad_norm": 0.0,
+      "learning_rate": 7.846815534859592e-06,
+      "loss": 1.5114,
+      "mean_token_accuracy": 0.7957898869343035,
+      "mean_token_accuracy_utility": 0.7598237487866906,
+      "step": 1655
+    },
+    {
+      "epoch": 0.6141324454310025,
+      "grad_norm": 0.0,
+      "learning_rate": 7.7837872195487e-06,
+      "loss": 1.3833,
+      "mean_token_accuracy": 0.7984821797155183,
+      "mean_token_accuracy_utility": 0.7824561652033958,
+      "step": 1660
+    },
+    {
+      "epoch": 0.6159822419533851,
+      "grad_norm": 0.0,
+      "learning_rate": 7.720851357567778e-06,
+      "loss": 1.4411,
+      "mean_token_accuracy": 0.8057899380105551,
+      "mean_token_accuracy_utility": 0.7648707878880501,
+      "step": 1665
+    },
+    {
+      "epoch": 0.6178320384757676,
+      "grad_norm": 0.0,
+      "learning_rate": 7.658010574400285e-06,
+      "loss": 1.4696,
+      "mean_token_accuracy": 0.8023791318126087,
+      "mean_token_accuracy_utility": 0.7652019704353817,
+      "step": 1670
+    },
+    {
+      "epoch": 0.6196818349981502,
+      "grad_norm": 0.0,
+      "learning_rate": 7.595267491563311e-06,
+      "loss": 1.4252,
+      "mean_token_accuracy": 0.8029975907941749,
+      "mean_token_accuracy_utility": 0.7692018098699458,
+      "step": 1675
+    },
+    {
+      "epoch": 0.6215316315205327,
+      "grad_norm": 0.0,
+      "learning_rate": 7.532624726498181e-06,
+      "loss": 1.4745,
+      "mean_token_accuracy": 0.7986012765363674,
+      "mean_token_accuracy_utility": 0.7639380589323593,
+      "step": 1680
+    },
+    {
+      "epoch": 0.6233814280429153,
+      "grad_norm": 0.0,
+      "learning_rate": 7.470084892461305e-06,
+      "loss": 1.4693,
+      "mean_token_accuracy": 0.7914230182713636,
+      "mean_token_accuracy_utility": 0.770391557051372,
+      "step": 1685
+    },
+    {
+      "epoch": 0.6252312245652978,
+      "grad_norm": 0.0,
+      "learning_rate": 7.407650598415119e-06,
+      "loss": 1.4309,
+      "mean_token_accuracy": 0.7979528804996306,
+      "mean_token_accuracy_utility": 0.7729017966208745,
+      "step": 1690
+    },
+    {
+      "epoch": 0.6270810210876804,
+      "grad_norm": 0.0,
+      "learning_rate": 7.34532444891928e-06,
+      "loss": 1.4533,
+      "mean_token_accuracy": 0.8010745377616116,
+      "mean_token_accuracy_utility": 0.7693516292115888,
+      "step": 1695
+    },
+    {
+      "epoch": 0.6289308176100629,
+      "grad_norm": 0.0,
+      "learning_rate": 7.283109044021987e-06,
+      "loss": 1.5813,
+      "mean_token_accuracy": 0.7794394533988436,
+      "mean_token_accuracy_utility": 0.7545082561454449,
+      "step": 1700
+    },
+    {
+      "epoch": 0.6289308176100629,
+      "eval_mean_token_accuracy": 0.7891158557218554,
+      "eval_reasoning_loss": 0.6626559495925903,
+      "eval_reasoning_runtime": 89.9264,
+      "eval_reasoning_samples_per_second": 1.435,
+      "eval_reasoning_steps_per_second": 0.367,
+      "step": 1700
+    },
+    {
+      "epoch": 0.6289308176100629,
+      "eval_mean_token_accuracy": 0.7414511072545265,
+      "eval_utility_loss": 0.9454062581062317,
+      "eval_utility_runtime": 33.8793,
+      "eval_utility_samples_per_second": 1.446,
+      "eval_utility_steps_per_second": 0.384,
+      "step": 1700
+    },
+    {
+      "epoch": 0.6307806141324455,
+      "grad_norm": 0.0,
+      "learning_rate": 7.221006979151546e-06,
+      "loss": 1.4771,
+      "mean_token_accuracy": 0.7966882586264398,
+      "mean_token_accuracy_utility": 0.76997032207052,
+      "step": 1705
+    },
+    {
+      "epoch": 0.632630410654828,
+      "grad_norm": 0.0,
+      "learning_rate": 7.159020845008056e-06,
+      "loss": 1.5535,
+      "mean_token_accuracy": 0.7779027705156573,
+      "mean_token_accuracy_utility": 0.7593526488651257,
+      "step": 1710
+    },
+    {
+      "epoch": 0.6344802071772105,
+      "grad_norm": 0.0,
+      "learning_rate": 7.097153227455379e-06,
+      "loss": 1.4544,
+      "mean_token_accuracy": 0.7998120172132179,
+      "mean_token_accuracy_utility": 0.764023821138527,
+      "step": 1715
+    },
+    {
+      "epoch": 0.636330003699593,
+      "grad_norm": 0.0,
+      "learning_rate": 7.035406707413221e-06,
+      "loss": 1.4901,
+      "mean_token_accuracy": 0.7888191263011988,
+      "mean_token_accuracy_utility": 0.7675702408682823,
+      "step": 1720
+    },
+    {
+      "epoch": 0.6381798002219756,
+      "grad_norm": 0.0,
+      "learning_rate": 6.973783860749499e-06,
+      "loss": 1.4433,
+      "mean_token_accuracy": 0.79687914887677,
+      "mean_token_accuracy_utility": 0.7712060219037722,
+      "step": 1725
+    },
+    {
+      "epoch": 0.6400295967443581,
+      "grad_norm": 0.0,
+      "learning_rate": 6.912287258172869e-06,
+      "loss": 1.3608,
+      "mean_token_accuracy": 0.8183243311457472,
+      "mean_token_accuracy_utility": 0.7712664960549414,
+      "step": 1730
+    },
+    {
+      "epoch": 0.6418793932667407,
+      "grad_norm": 0.0,
+      "learning_rate": 6.8509194651254825e-06,
+      "loss": 1.5488,
+      "mean_token_accuracy": 0.7903365441543045,
+      "mean_token_accuracy_utility": 0.756691494648907,
+      "step": 1735
+    },
+    {
+      "epoch": 0.6437291897891232,
+      "grad_norm": 0.0,
+      "learning_rate": 6.789683041675977e-06,
+      "loss": 1.4451,
+      "mean_token_accuracy": 0.7930503038767649,
+      "mean_token_accuracy_utility": 0.7735197678756534,
+      "step": 1740
+    },
+    {
+      "epoch": 0.6455789863115058,
+      "grad_norm": 0.0,
+      "learning_rate": 6.7285805424126585e-06,
+      "loss": 1.4251,
+      "mean_token_accuracy": 0.7911912176256257,
+      "mean_token_accuracy_utility": 0.7790633731307564,
+      "step": 1745
+    },
+    {
+      "epoch": 0.6474287828338883,
+      "grad_norm": 0.0,
+      "learning_rate": 6.667614516336949e-06,
+      "loss": 1.4391,
+      "mean_token_accuracy": 0.7993747966151794,
+      "mean_token_accuracy_utility": 0.7745588579700644,
+      "step": 1750
+    },
+    {
+      "epoch": 0.6492785793562708,
+      "grad_norm": 0.0,
+      "learning_rate": 6.60678750675704e-06,
+      "loss": 1.4756,
+      "mean_token_accuracy": 0.7897338465823748,
+      "mean_token_accuracy_utility": 0.7698821895469827,
+      "step": 1755
+    },
+    {
+      "epoch": 0.6511283758786534,
+      "grad_norm": 0.0,
+      "learning_rate": 6.546102051181808e-06,
+      "loss": 1.4847,
+      "mean_token_accuracy": 0.8110885712933588,
+      "mean_token_accuracy_utility": 0.7580209894711196,
+      "step": 1760
+    },
+    {
+      "epoch": 0.6529781724010358,
+      "grad_norm": 0.0,
+      "learning_rate": 6.485560681214933e-06,
+      "loss": 1.4729,
+      "mean_token_accuracy": 0.7950825103303787,
+      "mean_token_accuracy_utility": 0.7687819602384801,
+      "step": 1765
+    },
+    {
+      "epoch": 0.6548279689234184,
+      "grad_norm": 0.0,
+      "learning_rate": 6.425165922449319e-06,
+      "loss": 1.4696,
+      "mean_token_accuracy": 0.7894440957022719,
+      "mean_token_accuracy_utility": 0.7712943735552685,
+      "step": 1770
+    },
+    {
+      "epoch": 0.6566777654458009,
+      "grad_norm": 0.0,
+      "learning_rate": 6.364920294361701e-06,
+      "loss": 1.4141,
+      "mean_token_accuracy": 0.7978398757598896,
+      "mean_token_accuracy_utility": 0.7799618477281332,
+      "step": 1775
+    },
+    {
+      "epoch": 0.6585275619681835,
+      "grad_norm": 0.0,
+      "learning_rate": 6.3048263102075766e-06,
+      "loss": 1.4306,
+      "mean_token_accuracy": 0.794975950564929,
+      "mean_token_accuracy_utility": 0.7744765171442859,
+      "step": 1780
+    },
+    {
+      "epoch": 0.660377358490566,
+      "grad_norm": 0.0,
+      "learning_rate": 6.244886476916325e-06,
+      "loss": 1.473,
+      "mean_token_accuracy": 0.7940074577170748,
+      "mean_token_accuracy_utility": 0.7660302456626216,
+      "step": 1785
+    },
+    {
+      "epoch": 0.6622271550129486,
+      "grad_norm": 0.0,
+      "learning_rate": 6.185103294986654e-06,
+      "loss": 1.5502,
+      "mean_token_accuracy": 0.7970587599046872,
+      "mean_token_accuracy_utility": 0.755714367339824,
+      "step": 1790
+    },
+    {
+      "epoch": 0.6640769515353311,
+      "grad_norm": 0.0,
+      "learning_rate": 6.125479258382268e-06,
+      "loss": 1.488,
+      "mean_token_accuracy": 0.7966358071148147,
+      "mean_token_accuracy_utility": 0.7569478380018032,
+      "step": 1795
+    },
+    {
+      "epoch": 0.6659267480577137,
+      "grad_norm": 0.0,
+      "learning_rate": 6.066016854427846e-06,
+      "loss": 1.3917,
+      "mean_token_accuracy": 0.8009998724811462,
+      "mean_token_accuracy_utility": 0.7796975279860796,
+      "step": 1800
+    },
+    {
+      "epoch": 0.6659267480577137,
+      "eval_mean_token_accuracy": 0.7893308508198826,
+      "eval_reasoning_loss": 0.6621161699295044,
+      "eval_reasoning_runtime": 89.9715,
+      "eval_reasoning_samples_per_second": 1.434,
+      "eval_reasoning_steps_per_second": 0.367,
+      "step": 1800
+    },
+    {
+      "epoch": 0.6659267480577137,
+      "eval_mean_token_accuracy": 0.7412605987126059,
+      "eval_utility_loss": 0.9453049302101135,
+      "eval_utility_runtime": 33.8634,
+      "eval_utility_samples_per_second": 1.447,
+      "eval_utility_steps_per_second": 0.384,
+      "step": 1800
+    },
+    {
+      "epoch": 0.6677765445800962,
+      "grad_norm": 0.0,
+      "learning_rate": 6.006718563705258e-06,
+      "loss": 1.4572,
+      "mean_token_accuracy": 0.7888068658454819,
+      "mean_token_accuracy_utility": 0.7734194608652001,
+      "step": 1805
+    },
+    {
+      "epoch": 0.6696263411024788,
+      "grad_norm": 0.0,
+      "learning_rate": 5.947586859950103e-06,
+      "loss": 1.3835,
+      "mean_token_accuracy": 0.8051813675259318,
+      "mean_token_accuracy_utility": 0.7808083174270846,
+      "step": 1810
+    },
+    {
+      "epoch": 0.6714761376248612,
+      "grad_norm": 0.0,
+      "learning_rate": 5.888624209948495e-06,
+      "loss": 1.5353,
+      "mean_token_accuracy": 0.7984623257132955,
+      "mean_token_accuracy_utility": 0.7556040222289508,
+      "step": 1815
+    },
+    {
+      "epoch": 0.6733259341472438,
+      "grad_norm": 0.0,
+      "learning_rate": 5.82983307343416e-06,
+      "loss": 1.3978,
+      "mean_token_accuracy": 0.807197575831931,
+      "mean_token_accuracy_utility": 0.7733396915091537,
+      "step": 1820
+    },
+    {
+      "epoch": 0.6751757306696263,
+      "grad_norm": 0.0,
+      "learning_rate": 5.771215902985848e-06,
+      "loss": 1.4245,
+      "mean_token_accuracy": 0.8027392652276883,
+      "mean_token_accuracy_utility": 0.7695503262503385,
+      "step": 1825
+    },
+    {
+      "epoch": 0.6770255271920089,
+      "grad_norm": 0.0,
+      "learning_rate": 5.712775143924978e-06,
+      "loss": 1.4598,
+      "mean_token_accuracy": 0.8067033838171345,
+      "mean_token_accuracy_utility": 0.7606681961453423,
+      "step": 1830
+    },
+    {
+      "epoch": 0.6788753237143914,
+      "grad_norm": 0.0,
+      "learning_rate": 5.654513234213663e-06,
+      "loss": 1.4207,
+      "mean_token_accuracy": 0.7998780666709633,
+      "mean_token_accuracy_utility": 0.7768345361742576,
+      "step": 1835
+    },
+    {
+      "epoch": 0.680725120236774,
+      "grad_norm": 0.0,
+      "learning_rate": 5.596432604352982e-06,
+      "loss": 1.4976,
+      "mean_token_accuracy": 0.8055079971457427,
+      "mean_token_accuracy_utility": 0.7524714544329426,
+      "step": 1840
+    },
+    {
+      "epoch": 0.6825749167591565,
+      "grad_norm": 0.0,
+      "learning_rate": 5.538535677281608e-06,
+      "loss": 1.4921,
+      "mean_token_accuracy": 0.7959607050990434,
+      "mean_token_accuracy_utility": 0.7631530055040496,
+      "step": 1845
+    },
+    {
+      "epoch": 0.684424713281539,
+      "grad_norm": 0.0,
+      "learning_rate": 5.480824868274699e-06,
+      "loss": 1.4764,
+      "mean_token_accuracy": 0.7958508977988242,
+      "mean_token_accuracy_utility": 0.7668248576228198,
+      "step": 1850
+    },
+    {
+      "epoch": 0.6862745098039216,
+      "grad_norm": 0.0,
+      "learning_rate": 5.423302584843186e-06,
+      "loss": 1.5358,
+      "mean_token_accuracy": 0.7908297738286488,
+      "mean_token_accuracy_utility": 0.7580432720723081,
+      "step": 1855
+    },
+    {
+      "epoch": 0.6881243063263041,
+      "grad_norm": 0.0,
+      "learning_rate": 5.365971226633294e-06,
+      "loss": 1.4324,
+      "mean_token_accuracy": 0.8066440331820157,
+      "mean_token_accuracy_utility": 0.7659994870024947,
+      "step": 1860
+    },
+    {
+      "epoch": 0.6899741028486867,
+      "grad_norm": 0.0,
+      "learning_rate": 5.308833185326472e-06,
+      "loss": 1.5183,
+      "mean_token_accuracy": 0.7963535148376292,
+      "mean_token_accuracy_utility": 0.7601163025087555,
+      "step": 1865
+    },
+    {
+      "epoch": 0.6918238993710691,
+      "grad_norm": 0.0,
+      "learning_rate": 5.25189084453959e-06,
+      "loss": 1.4778,
+      "mean_token_accuracy": 0.7969017881309979,
+      "mean_token_accuracy_utility": 0.7600198527482539,
+      "step": 1870
+    },
+    {
+      "epoch": 0.6936736958934517,
+      "grad_norm": 0.0,
+      "learning_rate": 5.19514657972553e-06,
+      "loss": 1.4025,
+      "mean_token_accuracy": 0.8135830914582888,
+      "mean_token_accuracy_utility": 0.7728499185655984,
+      "step": 1875
+    },
+    {
+      "epoch": 0.6955234924158342,
+      "grad_norm": 0.0,
+      "learning_rate": 5.138602758074062e-06,
+      "loss": 1.3872,
+      "mean_token_accuracy": 0.8135303989158436,
+      "mean_token_accuracy_utility": 0.7707382965986183,
+      "step": 1880
+    },
+    {
+      "epoch": 0.6973732889382168,
+      "grad_norm": 0.0,
+      "learning_rate": 5.082261738413124e-06,
+      "loss": 1.4893,
+      "mean_token_accuracy": 0.785450924420247,
+      "mean_token_accuracy_utility": 0.7692337614284864,
+      "step": 1885
+    },
+    {
+      "epoch": 0.6992230854605993,
+      "grad_norm": 0.0,
+      "learning_rate": 5.026125871110383e-06,
+      "loss": 1.5117,
+      "mean_token_accuracy": 0.7965815195118735,
+      "mean_token_accuracy_utility": 0.7587083367287034,
+      "step": 1890
+    },
+    {
+      "epoch": 0.7010728819829819,
+      "grad_norm": 0.0,
+      "learning_rate": 4.970197497975216e-06,
+      "loss": 1.4662,
+      "mean_token_accuracy": 0.8075670217491482,
+      "mean_token_accuracy_utility": 0.7586032400187501,
+      "step": 1895
+    },
+    {
+      "epoch": 0.7029226785053644,
+      "grad_norm": 0.0,
+      "learning_rate": 4.914478952160997e-06,
+      "loss": 1.4766,
+      "mean_token_accuracy": 0.7936536226054172,
+      "mean_token_accuracy_utility": 0.7730669472361533,
+      "step": 1900
+    },
+    {
+      "epoch": 0.7029226785053644,
+      "eval_mean_token_accuracy": 0.7893582876453399,
+      "eval_reasoning_loss": 0.6619249582290649,
+      "eval_reasoning_runtime": 89.9542,
+      "eval_reasoning_samples_per_second": 1.434,
+      "eval_reasoning_steps_per_second": 0.367,
+      "step": 1900
+    },
+    {
+      "epoch": 0.7029226785053644,
+      "eval_mean_token_accuracy": 0.7408278323100401,
+      "eval_utility_loss": 0.9458861351013184,
+      "eval_utility_runtime": 33.8657,
+      "eval_utility_samples_per_second": 1.447,
+      "eval_utility_steps_per_second": 0.384,
+      "step": 1900
+    },
+    {
+      "epoch": 0.704772475027747,
+      "grad_norm": 0.0,
+      "learning_rate": 4.858972558067784e-06,
+      "loss": 1.398,
+      "mean_token_accuracy": 0.8139111662626869,
+      "mean_token_accuracy_utility": 0.7690568507206584,
+      "step": 1905
+    },
+    {
+      "epoch": 0.7066222715501295,
+      "grad_norm": 0.0,
+      "learning_rate": 4.803680631245333e-06,
+      "loss": 1.3648,
+      "mean_token_accuracy": 0.80932881245564,
+      "mean_token_accuracy_utility": 0.7819303708230234,
+      "step": 1910
+    },
+    {
+      "epoch": 0.7084720680725121,
+      "grad_norm": 0.0,
+      "learning_rate": 4.748605478296508e-06,
+      "loss": 1.4107,
+      "mean_token_accuracy": 0.8090043828097793,
+      "mean_token_accuracy_utility": 0.7692716354237314,
+      "step": 1915
+    },
+    {
+      "epoch": 0.7103218645948945,
+      "grad_norm": 0.0,
+      "learning_rate": 4.6937493967810665e-06,
+      "loss": 1.504,
+      "mean_token_accuracy": 0.8026019559352772,
+      "mean_token_accuracy_utility": 0.756783641568626,
+      "step": 1920
+    },
+    {
+      "epoch": 0.7121716611172771,
+      "grad_norm": 0.0,
+      "learning_rate": 4.639114675119797e-06,
+      "loss": 1.5259,
+      "mean_token_accuracy": 0.7891990050109243,
+      "mean_token_accuracy_utility": 0.7590334171067801,
+      "step": 1925
+    },
+    {
+      "epoch": 0.7140214576396596,
+      "grad_norm": 0.0,
+      "learning_rate": 4.584703592499068e-06,
+      "loss": 1.4604,
+      "mean_token_accuracy": 0.7992271055553481,
+      "mean_token_accuracy_utility": 0.7672508497368753,
+      "step": 1930
+    },
+    {
+      "epoch": 0.7158712541620422,
+      "grad_norm": 0.0,
+      "learning_rate": 4.530518418775734e-06,
+      "loss": 1.4364,
+      "mean_token_accuracy": 0.7996222325915878,
+      "mean_token_accuracy_utility": 0.7713781801088937,
+      "step": 1935
+    },
+    {
+      "epoch": 0.7177210506844247,
+      "grad_norm": 0.0,
+      "learning_rate": 4.476561414382451e-06,
+      "loss": 1.3144,
+      "mean_token_accuracy": 0.8161018460919596,
+      "mean_token_accuracy_utility": 0.7881745634186711,
+      "step": 1940
+    },
+    {
+      "epoch": 0.7195708472068072,
+      "grad_norm": 0.0,
+      "learning_rate": 4.422834830233378e-06,
+      "loss": 1.4128,
+      "mean_token_accuracy": 0.8017413293435744,
+      "mean_token_accuracy_utility": 0.7765822899655034,
+      "step": 1945
+    },
+    {
+      "epoch": 0.7214206437291898,
+      "grad_norm": 0.0,
+      "learning_rate": 4.369340907630284e-06,
+      "loss": 1.4823,
+      "mean_token_accuracy": 0.7898830788475875,
+      "mean_token_accuracy_utility": 0.7689727769873496,
+      "step": 1950
+    },
+    {
+      "epoch": 0.7232704402515723,
+      "grad_norm": 0.0,
+      "learning_rate": 4.3160818781690286e-06,
+      "loss": 1.4379,
+      "mean_token_accuracy": 0.7947224049226878,
+      "mean_token_accuracy_utility": 0.776082118273089,
+      "step": 1955
+    },
+    {
+      "epoch": 0.7251202367739549,
+      "grad_norm": 0.0,
+      "learning_rate": 4.26305996364649e-06,
+      "loss": 1.4663,
+      "mean_token_accuracy": 0.8023473789285578,
+      "mean_token_accuracy_utility": 0.764713921392923,
+      "step": 1960
+    },
+    {
+      "epoch": 0.7269700332963374,
+      "grad_norm": 0.0,
+      "learning_rate": 4.210277375967855e-06,
+      "loss": 1.4059,
+      "mean_token_accuracy": 0.7963506097398114,
+      "mean_token_accuracy_utility": 0.7775025384803577,
+      "step": 1965
+    },
+    {
+      "epoch": 0.72881982981872,
+      "grad_norm": 0.0,
+      "learning_rate": 4.157736317054373e-06,
+      "loss": 1.5219,
+      "mean_token_accuracy": 0.7973201265987285,
+      "mean_token_accuracy_utility": 0.7557925051067267,
+      "step": 1970
+    },
+    {
+      "epoch": 0.7306696263411024,
+      "grad_norm": 0.0,
+      "learning_rate": 4.105438978751465e-06,
+      "loss": 1.4753,
+      "mean_token_accuracy": 0.8036728312797463,
+      "mean_token_accuracy_utility": 0.7567998322350222,
+      "step": 1975
+    },
+    {
+      "epoch": 0.732519422863485,
+      "grad_norm": 0.0,
+      "learning_rate": 4.053387542737323e-06,
+      "loss": 1.4149,
+      "mean_token_accuracy": 0.8031151982080837,
+      "mean_token_accuracy_utility": 0.7767337082232579,
+      "step": 1980
+    },
+    {
+      "epoch": 0.7343692193858675,
+      "grad_norm": 0.0,
+      "learning_rate": 4.001584180431869e-06,
+      "loss": 1.4603,
+      "mean_token_accuracy": 0.8076524415465958,
+      "mean_token_accuracy_utility": 0.761882758103068,
+      "step": 1985
+    },
+    {
+      "epoch": 0.7362190159082501,
+      "grad_norm": 0.0,
+      "learning_rate": 3.950031052906183e-06,
+      "loss": 1.4615,
+      "mean_token_accuracy": 0.8005713405992845,
+      "mean_token_accuracy_utility": 0.7653999541425996,
+      "step": 1990
+    },
+    {
+      "epoch": 0.7380688124306326,
+      "grad_norm": 0.0,
+      "learning_rate": 3.898730310792346e-06,
+      "loss": 1.4491,
+      "mean_token_accuracy": 0.7953040888491817,
+      "mean_token_accuracy_utility": 0.7741969939339663,
+      "step": 1995
+    },
+    {
+      "epoch": 0.7399186089530152,
+      "grad_norm": 0.0,
+      "learning_rate": 3.847684094193733e-06,
+      "loss": 1.4411,
+      "mean_token_accuracy": 0.7911381448487306,
+      "mean_token_accuracy_utility": 0.7743672491667726,
+      "step": 2000
+    },
+    {
+      "epoch": 0.7399186089530152,
+      "eval_mean_token_accuracy": 0.7893956198837546,
+      "eval_reasoning_loss": 0.6617336869239807,
+      "eval_reasoning_runtime": 90.0327,
+      "eval_reasoning_samples_per_second": 1.433,
+      "eval_reasoning_steps_per_second": 0.367,
+      "step": 2000
+    },
+    {
+      "epoch": 0.7399186089530152,
+      "eval_mean_token_accuracy": 0.7414982881904864,
+      "eval_utility_loss": 0.9460301399230957,
+      "eval_utility_runtime": 33.9035,
+      "eval_utility_samples_per_second": 1.445,
+      "eval_utility_steps_per_second": 0.383,
+      "step": 2000
+    },
+    {
+      "epoch": 0.7417684054753977,
+      "grad_norm": 0.0,
+      "learning_rate": 3.7968945325957175e-06,
+      "loss": 1.4952,
+      "mean_token_accuracy": 0.7936326551955625,
+      "mean_token_accuracy_utility": 0.7643957765243694,
+      "step": 2005
+    },
+    {
+      "epoch": 0.7436182019977803,
+      "grad_norm": 0.0,
+      "learning_rate": 3.746363744776844e-06,
+      "loss": 1.4843,
+      "mean_token_accuracy": 0.7932937154051534,
+      "mean_token_accuracy_utility": 0.7641589961473966,
+      "step": 2010
+    },
+    {
+      "epoch": 0.7454679985201628,
+      "grad_norm": 0.0,
+      "learning_rate": 3.69609383872045e-06,
+      "loss": 1.4573,
+      "mean_token_accuracy": 0.7996641829771651,
+      "mean_token_accuracy_utility": 0.7651040199200598,
+      "step": 2015
+    },
+    {
+      "epoch": 0.7473177950425454,
+      "grad_norm": 0.0,
+      "learning_rate": 3.646086911526706e-06,
+      "loss": 1.3727,
+      "mean_token_accuracy": 0.8058435656693395,
+      "mean_token_accuracy_utility": 0.7787953549124685,
+      "step": 2020
+    },
+    {
+      "epoch": 0.7491675915649278,
+      "grad_norm": 0.0,
+      "learning_rate": 3.5963450493251552e-06,
+      "loss": 1.3685,
+      "mean_token_accuracy": 0.7933896059827604,
+      "mean_token_accuracy_utility": 0.7914208192352726,
+      "step": 2025
+    },
+    {
+      "epoch": 0.7510173880873104,
+      "grad_norm": 0.0,
+      "learning_rate": 3.5468703271876626e-06,
+      "loss": 1.4147,
+      "mean_token_accuracy": 0.8078674722131662,
+      "mean_token_accuracy_utility": 0.7714893133522255,
+      "step": 2030
+    },
+    {
+      "epoch": 0.7528671846096929,
+      "grad_norm": 0.0,
+      "learning_rate": 3.4976648090418685e-06,
+      "loss": 1.4755,
+      "mean_token_accuracy": 0.8010973331000215,
+      "mean_token_accuracy_utility": 0.7616263750759638,
+      "step": 2035
+    },
+    {
+      "epoch": 0.7547169811320755,
+      "grad_norm": 0.0,
+      "learning_rate": 3.4487305475850754e-06,
+      "loss": 1.5541,
+      "mean_token_accuracy": 0.7953744509562961,
+      "mean_token_accuracy_utility": 0.7522818860870958,
+      "step": 2040
+    },
+    {
+      "epoch": 0.756566777654458,
+      "grad_norm": 0.0,
+      "learning_rate": 3.400069584198633e-06,
+      "loss": 1.51,
+      "mean_token_accuracy": 0.7990586499406823,
+      "mean_token_accuracy_utility": 0.7575872345053186,
+      "step": 2045
+    },
+    {
+      "epoch": 0.7584165741768405,
+      "grad_norm": 0.0,
+      "learning_rate": 3.351683948862753e-06,
+      "loss": 1.4646,
+      "mean_token_accuracy": 0.7913004047306783,
+      "mean_token_accuracy_utility": 0.7649798012488134,
+      "step": 2050
+    },
+    {
+      "epoch": 0.7602663706992231,
+      "grad_norm": 0.0,
+      "learning_rate": 3.3035756600718515e-06,
+      "loss": 1.4488,
+      "mean_token_accuracy": 0.802978849262735,
+      "mean_token_accuracy_utility": 0.7643597115094828,
+      "step": 2055
+    },
+    {
+      "epoch": 0.7621161672216056,
+      "grad_norm": 0.0,
+      "learning_rate": 3.2557467247503205e-06,
+      "loss": 1.4448,
+      "mean_token_accuracy": 0.7974946504824356,
+      "mean_token_accuracy_utility": 0.7744927322917436,
+      "step": 2060
+    },
+    {
+      "epoch": 0.7639659637439882,
+      "grad_norm": 0.0,
+      "learning_rate": 3.208199138168826e-06,
+      "loss": 1.4992,
+      "mean_token_accuracy": 0.7854159286290064,
+      "mean_token_accuracy_utility": 0.7708999619166182,
+      "step": 2065
+    },
+    {
+      "epoch": 0.7658157602663707,
+      "grad_norm": 0.0,
+      "learning_rate": 3.160934883861053e-06,
+      "loss": 1.4338,
+      "mean_token_accuracy": 0.7924753722183275,
+      "mean_token_accuracy_utility": 0.7775050108427186,
+      "step": 2070
+    },
+    {
+      "epoch": 0.7676655567887533,
+      "grad_norm": 0.0,
+      "learning_rate": 3.113955933540973e-06,
+      "loss": 1.4087,
+      "mean_token_accuracy": 0.8180705648346256,
+      "mean_token_accuracy_utility": 0.7659782447260743,
+      "step": 2075
+    },
+    {
+      "epoch": 0.7695153533111357,
+      "grad_norm": 0.0,
+      "learning_rate": 3.0672642470205794e-06,
+      "loss": 1.5673,
+      "mean_token_accuracy": 0.800698591683294,
+      "mean_token_accuracy_utility": 0.7489991199953019,
+      "step": 2080
+    },
+    {
+      "epoch": 0.7713651498335183,
+      "grad_norm": 0.0,
+      "learning_rate": 3.020861772128145e-06,
+      "loss": 1.4892,
+      "mean_token_accuracy": 0.7838394155977515,
+      "mean_token_accuracy_utility": 0.7672138276124784,
+      "step": 2085
+    },
+    {
+      "epoch": 0.7732149463559008,
+      "grad_norm": 0.0,
+      "learning_rate": 2.9747504446269447e-06,
+      "loss": 1.4263,
+      "mean_token_accuracy": 0.8040732745357649,
+      "mean_token_accuracy_utility": 0.7703933241175338,
+      "step": 2090
+    },
+    {
+      "epoch": 0.7750647428782834,
+      "grad_norm": 0.0,
+      "learning_rate": 2.9289321881345257e-06,
+      "loss": 1.4797,
+      "mean_token_accuracy": 0.793325840100947,
+      "mean_token_accuracy_utility": 0.7691592978318226,
+      "step": 2095
+    },
+    {
+      "epoch": 0.7769145394006659,
+      "grad_norm": 0.0,
+      "learning_rate": 2.883408914042435e-06,
+      "loss": 1.4718,
+      "mean_token_accuracy": 0.7855140286192999,
+      "mean_token_accuracy_utility": 0.7743906160592808,
+      "step": 2100
+    },
+    {
+      "epoch": 0.7769145394006659,
+      "eval_mean_token_accuracy": 0.7895270533059304,
+      "eval_reasoning_loss": 0.6615400314331055,
+      "eval_reasoning_runtime": 90.0936,
+      "eval_reasoning_samples_per_second": 1.432,
+      "eval_reasoning_steps_per_second": 0.366,
+      "step": 2100
+    },
+    {
+      "epoch": 0.7769145394006659,
+      "eval_mean_token_accuracy": 0.7411485253160033,
+      "eval_utility_loss": 0.9461448788642883,
+      "eval_utility_runtime": 33.9306,
+      "eval_utility_samples_per_second": 1.444,
+      "eval_utility_steps_per_second": 0.383,
+      "step": 2100
+    },
+    {
+      "epoch": 0.7787643359230485,
+      "grad_norm": 0.0,
+      "learning_rate": 2.838182521436498e-06,
+      "loss": 1.4643,
+      "mean_token_accuracy": 0.7989379912972882,
+      "mean_token_accuracy_utility": 0.7641540680369201,
+      "step": 2105
+    },
+    {
+      "epoch": 0.780614132445431,
+      "grad_norm": 0.0,
+      "learning_rate": 2.7932548970175977e-06,
+      "loss": 1.4249,
+      "mean_token_accuracy": 0.7902313896122444,
+      "mean_token_accuracy_utility": 0.7836814897273753,
+      "step": 2110
+    },
+    {
+      "epoch": 0.7824639289678136,
+      "grad_norm": 0.0,
+      "learning_rate": 2.74862791502295e-06,
+      "loss": 1.4155,
+      "mean_token_accuracy": 0.801362316548647,
+      "mean_token_accuracy_utility": 0.7750410252768204,
+      "step": 2115
+    },
+    {
+      "epoch": 0.7843137254901961,
+      "grad_norm": 0.0,
+      "learning_rate": 2.7043034371479404e-06,
+      "loss": 1.4088,
+      "mean_token_accuracy": 0.7918788533306401,
+      "mean_token_accuracy_utility": 0.7834347771619454,
+      "step": 2120
+    },
+    {
+      "epoch": 0.7861635220125787,
+      "grad_norm": 0.0,
+      "learning_rate": 2.660283312468438e-06,
+      "loss": 1.4242,
+      "mean_token_accuracy": 0.8146513534893358,
+      "mean_token_accuracy_utility": 0.7649577044948932,
+      "step": 2125
+    },
+    {
+      "epoch": 0.7880133185349611,
+      "grad_norm": 0.0,
+      "learning_rate": 2.616569377363668e-06,
+      "loss": 1.539,
+      "mean_token_accuracy": 0.7871744759003871,
+      "mean_token_accuracy_utility": 0.7600869755539919,
+      "step": 2130
+    },
+    {
+      "epoch": 0.7898631150573437,
+      "grad_norm": 0.0,
+      "learning_rate": 2.573163455439601e-06,
+      "loss": 1.4764,
+      "mean_token_accuracy": 0.7953277008294214,
+      "mean_token_accuracy_utility": 0.7714632803052093,
+      "step": 2135
+    },
+    {
+      "epoch": 0.7917129115797262,
+      "grad_norm": 0.0,
+      "learning_rate": 2.530067357452892e-06,
+      "loss": 1.5559,
+      "mean_token_accuracy": 0.7970430096950489,
+      "mean_token_accuracy_utility": 0.7499168518497921,
+      "step": 2140
+    },
+    {
+      "epoch": 0.7935627081021087,
+      "grad_norm": 0.0,
+      "learning_rate": 2.4872828812353146e-06,
+      "loss": 1.3166,
+      "mean_token_accuracy": 0.8050843057362155,
+      "mean_token_accuracy_utility": 0.7939503914555368,
+      "step": 2145
+    },
+    {
+      "epoch": 0.7954125046244913,
+      "grad_norm": 0.0,
+      "learning_rate": 2.4448118116187847e-06,
+      "loss": 1.4518,
+      "mean_token_accuracy": 0.7887234380960955,
+      "mean_token_accuracy_utility": 0.7735189523518317,
+      "step": 2150
+    },
+    {
+      "epoch": 0.7972623011468738,
+      "grad_norm": 0.0,
+      "learning_rate": 2.402655920360889e-06,
+      "loss": 1.4261,
+      "mean_token_accuracy": 0.8009981444736324,
+      "mean_token_accuracy_utility": 0.7738952564686956,
+      "step": 2155
+    },
+    {
+      "epoch": 0.7991120976692564,
+      "grad_norm": 0.0,
+      "learning_rate": 2.360816966070986e-06,
+      "loss": 1.4062,
+      "mean_token_accuracy": 0.8015563120077382,
+      "mean_token_accuracy_utility": 0.7762990598997237,
+      "step": 2160
+    },
+    {
+      "epoch": 0.8009618941916389,
+      "grad_norm": 0.0,
+      "learning_rate": 2.3192966941368247e-06,
+      "loss": 1.466,
+      "mean_token_accuracy": 0.7945381462505681,
+      "mean_token_accuracy_utility": 0.7689171334813447,
+      "step": 2165
+    },
+    {
+      "epoch": 0.8028116907140215,
+      "grad_norm": 0.0,
+      "learning_rate": 2.2780968366517476e-06,
+      "loss": 1.4585,
+      "mean_token_accuracy": 0.811144306414362,
+      "mean_token_accuracy_utility": 0.7574754239308172,
+      "step": 2170
+    },
+    {
+      "epoch": 0.804661487236404,
+      "grad_norm": 0.0,
+      "learning_rate": 2.237219112342426e-06,
+      "loss": 1.4613,
+      "mean_token_accuracy": 0.8013560979327986,
+      "mean_token_accuracy_utility": 0.7640644875816933,
+      "step": 2175
+    },
+    {
+      "epoch": 0.8065112837587866,
+      "grad_norm": 0.0,
+      "learning_rate": 2.196665226497169e-06,
+      "loss": 1.4914,
+      "mean_token_accuracy": 0.7974859898064274,
+      "mean_token_accuracy_utility": 0.7614511140883214,
+      "step": 2180
+    },
+    {
+      "epoch": 0.808361080281169,
+      "grad_norm": 0.0,
+      "learning_rate": 2.156436870894767e-06,
+      "loss": 1.5309,
+      "mean_token_accuracy": 0.7897351307715961,
+      "mean_token_accuracy_utility": 0.7578031225686629,
+      "step": 2185
+    },
+    {
+      "epoch": 0.8102108768035516,
+      "grad_norm": 0.0,
+      "learning_rate": 2.116535723733938e-06,
+      "loss": 1.3694,
+      "mean_token_accuracy": 0.8082425231424363,
+      "mean_token_accuracy_utility": 0.7749078038317758,
+      "step": 2190
+    },
+    {
+      "epoch": 0.8120606733259341,
+      "grad_norm": 0.0,
+      "learning_rate": 2.0769634495632986e-06,
+      "loss": 1.5083,
+      "mean_token_accuracy": 0.7968229872868917,
+      "mean_token_accuracy_utility": 0.7602161454950332,
+      "step": 2195
+    },
+    {
+      "epoch": 0.8139104698483167,
+      "grad_norm": 0.0,
+      "learning_rate": 2.037721699211933e-06,
+      "loss": 1.5002,
+      "mean_token_accuracy": 0.7907592646211535,
+      "mean_token_accuracy_utility": 0.7691067727966869,
+      "step": 2200
+    },
+    {
+      "epoch": 0.8139104698483167,
+      "eval_mean_token_accuracy": 0.7895529925720965,
+      "eval_reasoning_loss": 0.6614713668823242,
+      "eval_reasoning_runtime": 90.0716,
+      "eval_reasoning_samples_per_second": 1.432,
+      "eval_reasoning_steps_per_second": 0.366,
+      "step": 2200
+    },
+    {
+      "epoch": 0.8139104698483167,
+      "eval_mean_token_accuracy": 0.7412540367655184,
+      "eval_utility_loss": 0.9463587999343872,
+      "eval_utility_runtime": 33.9155,
+      "eval_utility_samples_per_second": 1.445,
+      "eval_utility_steps_per_second": 0.383,
+      "step": 2200
+    },
+    {
+      "epoch": 0.8157602663706992,
+      "grad_norm": 0.0,
+      "learning_rate": 1.998812109720535e-06,
+      "loss": 1.4185,
+      "mean_token_accuracy": 0.7944796413819558,
+      "mean_token_accuracy_utility": 0.7782242638761245,
+      "step": 2205
+    },
+    {
+      "epoch": 0.8176100628930818,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9602363042730964e-06,
+      "loss": 1.4699,
+      "mean_token_accuracy": 0.7876595054510245,
+      "mean_token_accuracy_utility": 0.7731822012064906,
+      "step": 2210
+    },
+    {
+      "epoch": 0.8194598594154643,
+      "grad_norm": 0.0,
+      "learning_rate": 1.921995892129208e-06,
+      "loss": 1.3657,
+      "mean_token_accuracy": 0.7996072587980493,
+      "mean_token_accuracy_utility": 0.7882187526553273,
+      "step": 2215
+    },
+    {
+      "epoch": 0.8213096559378469,
+      "grad_norm": 0.0,
+      "learning_rate": 1.8840924685569162e-06,
+      "loss": 1.417,
+      "mean_token_accuracy": 0.8048879945611697,
+      "mean_token_accuracy_utility": 0.772087444022632,
+      "step": 2220
+    },
+    {
+      "epoch": 0.8231594524602294,
+      "grad_norm": 0.0,
+      "learning_rate": 1.8465276147661905e-06,
+      "loss": 1.4079,
+      "mean_token_accuracy": 0.8108912607044283,
+      "mean_token_accuracy_utility": 0.767960452374641,
+      "step": 2225
+    },
+    {
+      "epoch": 0.825009248982612,
+      "grad_norm": 0.0,
+      "learning_rate": 1.809302897842934e-06,
+      "loss": 1.4577,
+      "mean_token_accuracy": 0.7955433343325028,
+      "mean_token_accuracy_utility": 0.7733865833334741,
+      "step": 2230
+    },
+    {
+      "epoch": 0.8268590455049944,
+      "grad_norm": 0.0,
+      "learning_rate": 1.7724198706836372e-06,
+      "loss": 1.3886,
+      "mean_token_accuracy": 0.8077197433541998,
+      "mean_token_accuracy_utility": 0.7796317936864602,
+      "step": 2235
+    },
+    {
+      "epoch": 0.8287088420273769,
+      "grad_norm": 0.0,
+      "learning_rate": 1.7358800719305735e-06,
+      "loss": 1.4444,
+      "mean_token_accuracy": 0.7920487463007458,
+      "mean_token_accuracy_utility": 0.7727177801110853,
+      "step": 2240
+    },
+    {
+      "epoch": 0.8305586385497595,
+      "grad_norm": 0.0,
+      "learning_rate": 1.6996850259076303e-06,
+      "loss": 1.4969,
+      "mean_token_accuracy": 0.8034080330648447,
+      "mean_token_accuracy_utility": 0.757701880712035,
+      "step": 2245
+    },
+    {
+      "epoch": 0.832408435072142,
+      "grad_norm": 0.0,
+      "learning_rate": 1.6638362425567012e-06,
+      "loss": 1.5348,
+      "mean_token_accuracy": 0.7886081733882132,
+      "mean_token_accuracy_utility": 0.7545162194856292,
+      "step": 2250
+    },
+    {
+      "epoch": 0.8342582315945246,
+      "grad_norm": 0.0,
+      "learning_rate": 1.6283352173747148e-06,
+      "loss": 1.4202,
+      "mean_token_accuracy": 0.7990345385617618,
+      "mean_token_accuracy_utility": 0.7786083325860162,
+      "step": 2255
+    },
+    {
+      "epoch": 0.8361080281169071,
+      "grad_norm": 0.0,
+      "learning_rate": 1.5931834313512307e-06,
+      "loss": 1.495,
+      "mean_token_accuracy": 0.7916907780427946,
+      "mean_token_accuracy_utility": 0.7652359127048942,
+      "step": 2260
+    },
+    {
+      "epoch": 0.8379578246392897,
+      "grad_norm": 0.0,
+      "learning_rate": 1.5583823509066665e-06,
+      "loss": 1.4146,
+      "mean_token_accuracy": 0.7978792241617884,
+      "mean_token_accuracy_utility": 0.7786343102658775,
+      "step": 2265
+    },
+    {
+      "epoch": 0.8398076211616722,
+      "grad_norm": 0.0,
+      "learning_rate": 1.5239334278311214e-06,
+      "loss": 1.4018,
+      "mean_token_accuracy": 0.8008069895530172,
+      "mean_token_accuracy_utility": 0.7806878857646108,
+      "step": 2270
+    },
+    {
+      "epoch": 0.8416574176840548,
+      "grad_norm": 0.0,
+      "learning_rate": 1.489838099223816e-06,
+      "loss": 1.4255,
+      "mean_token_accuracy": 0.8004267651468242,
+      "mean_token_accuracy_utility": 0.7742721284141806,
+      "step": 2275
+    },
+    {
+      "epoch": 0.8435072142064373,
+      "grad_norm": 0.0,
+      "learning_rate": 1.4560977874331284e-06,
+      "loss": 1.3967,
+      "mean_token_accuracy": 0.8042174465966051,
+      "mean_token_accuracy_utility": 0.7791510443005892,
+      "step": 2280
+    },
+    {
+      "epoch": 0.8453570107288199,
+      "grad_norm": 0.0,
+      "learning_rate": 1.4227138999972801e-06,
+      "loss": 1.4072,
+      "mean_token_accuracy": 0.8028825528414106,
+      "mean_token_accuracy_utility": 0.7792153801118895,
+      "step": 2285
+    },
+    {
+      "epoch": 0.8472068072512023,
+      "grad_norm": 0.0,
+      "learning_rate": 1.389687829585592e-06,
+      "loss": 1.4942,
+      "mean_token_accuracy": 0.7899770141658031,
+      "mean_token_accuracy_utility": 0.7700222056319248,
+      "step": 2290
+    },
+    {
+      "epoch": 0.8490566037735849,
+      "grad_norm": 0.0,
+      "learning_rate": 1.3570209539404067e-06,
+      "loss": 1.4556,
+      "mean_token_accuracy": 0.8079959294908653,
+      "mean_token_accuracy_utility": 0.7603729123358955,
+      "step": 2295
+    },
+    {
+      "epoch": 0.8509064002959674,
+      "grad_norm": 0.0,
+      "learning_rate": 1.3247146358196073e-06,
+      "loss": 1.4841,
+      "mean_token_accuracy": 0.7956275663056676,
+      "mean_token_accuracy_utility": 0.7626902466805247,
+      "step": 2300
+    },
+    {
+      "epoch": 0.8509064002959674,
+      "eval_mean_token_accuracy": 0.7895751661432567,
+      "eval_reasoning_loss": 0.6614620685577393,
+      "eval_reasoning_runtime": 90.0461,
+      "eval_reasoning_samples_per_second": 1.433,
+      "eval_reasoning_steps_per_second": 0.366,
+      "step": 2300
+    },
+    {
+      "epoch": 0.8509064002959674,
+      "eval_mean_token_accuracy": 0.7411330598093776,
+      "eval_utility_loss": 0.9463009238243103,
+      "eval_utility_runtime": 33.8929,
+      "eval_utility_samples_per_second": 1.446,
+      "eval_utility_steps_per_second": 0.384,
+      "step": 2300
+    },
+    {
+      "epoch": 0.85275619681835,
+      "grad_norm": 0.0,
+      "learning_rate": 1.2927702229397633e-06,
+      "loss": 1.4242,
+      "mean_token_accuracy": 0.7952264248764017,
+      "mean_token_accuracy_utility": 0.7763534711030321,
+      "step": 2305
+    },
+    {
+      "epoch": 0.8546059933407325,
+      "grad_norm": 0.0,
+      "learning_rate": 1.261189047919913e-06,
+      "loss": 1.4325,
+      "mean_token_accuracy": 0.793119335474529,
+      "mean_token_accuracy_utility": 0.7772030312102974,
+      "step": 2310
+    },
+    {
+      "epoch": 0.8564557898631151,
+      "grad_norm": 0.0,
+      "learning_rate": 1.2299724282259685e-06,
+      "loss": 1.466,
+      "mean_token_accuracy": 0.7931401419600121,
+      "mean_token_accuracy_utility": 0.7655450095258358,
+      "step": 2315
+    },
+    {
+      "epoch": 0.8583055863854976,
+      "grad_norm": 0.0,
+      "learning_rate": 1.1991216661157611e-06,
+      "loss": 1.4494,
+      "mean_token_accuracy": 0.809854879237534,
+      "mean_token_accuracy_utility": 0.7635330833199278,
+      "step": 2320
+    },
+    {
+      "epoch": 0.8601553829078802,
+      "grad_norm": 0.0,
+      "learning_rate": 1.1686380485847027e-06,
+      "loss": 1.4261,
+      "mean_token_accuracy": 0.7928946524958799,
+      "mean_token_accuracy_utility": 0.7784719724482723,
+      "step": 2325
+    },
+    {
+      "epoch": 0.8620051794302627,
+      "grad_norm": 0.0,
+      "learning_rate": 1.1385228473121113e-06,
+      "loss": 1.3874,
+      "mean_token_accuracy": 0.8120659377424045,
+      "mean_token_accuracy_utility": 0.7748399863714485,
+      "step": 2330
+    },
+    {
+      "epoch": 0.8638549759526452,
+      "grad_norm": 0.0,
+      "learning_rate": 1.1087773186081474e-06,
+      "loss": 1.4668,
+      "mean_token_accuracy": 0.7912231579288357,
+      "mean_token_accuracy_utility": 0.7729300020810411,
+      "step": 2335
+    },
+    {
+      "epoch": 0.8657047724750278,
+      "grad_norm": 0.0,
+      "learning_rate": 1.0794027033614175e-06,
+      "loss": 1.4142,
+      "mean_token_accuracy": 0.8014873793533702,
+      "mean_token_accuracy_utility": 0.7766108508438645,
+      "step": 2340
+    },
+    {
+      "epoch": 0.8675545689974102,
+      "grad_norm": 0.0,
+      "learning_rate": 1.0504002269871927e-06,
+      "loss": 1.4282,
+      "mean_token_accuracy": 0.8044117140931201,
+      "mean_token_accuracy_utility": 0.770941425727283,
+      "step": 2345
+    },
+    {
+      "epoch": 0.8694043655197928,
+      "grad_norm": 0.0,
+      "learning_rate": 1.0217710993763086e-06,
+      "loss": 1.368,
+      "mean_token_accuracy": 0.8122492255661286,
+      "mean_token_accuracy_utility": 0.7809756086077309,
+      "step": 2350
+    },
+    {
+      "epoch": 0.8712541620421753,
+      "grad_norm": 0.0,
+      "learning_rate": 9.935165148446658e-07,
+      "loss": 1.4386,
+      "mean_token_accuracy": 0.7867734376734361,
+      "mean_token_accuracy_utility": 0.7784626771260942,
+      "step": 2355
+    },
+    {
+      "epoch": 0.8731039585645579,
+      "grad_norm": 0.0,
+      "learning_rate": 9.656376520834365e-07,
+      "loss": 1.4864,
+      "mean_token_accuracy": 0.7983358819270492,
+      "mean_token_accuracy_utility": 0.7630155826298158,
+      "step": 2360
+    },
+    {
+      "epoch": 0.8749537550869404,
+      "grad_norm": 0.0,
+      "learning_rate": 9.381356741098702e-07,
+      "loss": 1.3765,
+      "mean_token_accuracy": 0.8106915380119133,
+      "mean_token_accuracy_utility": 0.7774656732354063,
+      "step": 2365
+    },
+    {
+      "epoch": 0.876803551609323,
+      "grad_norm": 0.0,
+      "learning_rate": 9.110117282187892e-07,
+      "loss": 1.3945,
+      "mean_token_accuracy": 0.8171934641231038,
+      "mean_token_accuracy_utility": 0.7677326040917677,
+      "step": 2370
+    },
+    {
+      "epoch": 0.8786533481317055,
+      "grad_norm": 0.0,
+      "learning_rate": 8.842669459347186e-07,
+      "loss": 1.3747,
+      "mean_token_accuracy": 0.7946265283096288,
+      "mean_token_accuracy_utility": 0.787389812491943,
+      "step": 2375
+    },
+    {
+      "epoch": 0.8805031446540881,
+      "grad_norm": 0.0,
+      "learning_rate": 8.579024429646932e-07,
+      "loss": 1.4635,
+      "mean_token_accuracy": 0.791452782402297,
+      "mean_token_accuracy_utility": 0.7694642034714203,
+      "step": 2380
+    },
+    {
+      "epoch": 0.8823529411764706,
+      "grad_norm": 0.0,
+      "learning_rate": 8.319193191517016e-07,
+      "loss": 1.406,
+      "mean_token_accuracy": 0.8022488645589239,
+      "mean_token_accuracy_utility": 0.7765649848238517,
+      "step": 2385
+    },
+    {
+      "epoch": 0.8842027376988532,
+      "grad_norm": 0.0,
+      "learning_rate": 8.063186584288108e-07,
+      "loss": 1.5018,
+      "mean_token_accuracy": 0.8017531909559363,
+      "mean_token_accuracy_utility": 0.7576114922778928,
+      "step": 2390
+    },
+    {
+      "epoch": 0.8860525342212356,
+      "grad_norm": 0.0,
+      "learning_rate": 7.81101528773951e-07,
+      "loss": 1.4494,
+      "mean_token_accuracy": 0.7914263391025488,
+      "mean_token_accuracy_utility": 0.7736085324780383,
+      "step": 2395
+    },
+    {
+      "epoch": 0.8879023307436182,
+      "grad_norm": 0.0,
+      "learning_rate": 7.562689821653568e-07,
+      "loss": 1.4247,
+      "mean_token_accuracy": 0.8059374522425206,
+      "mean_token_accuracy_utility": 0.769224343207265,
+      "step": 2400
+    },
+    {
+      "epoch": 0.8879023307436182,
+      "eval_mean_token_accuracy": 0.7895029749066712,
+      "eval_reasoning_loss": 0.6614764332771301,
+      "eval_reasoning_runtime": 90.0526,
+      "eval_reasoning_samples_per_second": 1.432,
+      "eval_reasoning_steps_per_second": 0.366,
+      "step": 2400
+    },
+    {
+      "epoch": 0.8879023307436182,
+      "eval_mean_token_accuracy": 0.7412620123266438,
+      "eval_utility_loss": 0.9462777972221375,
+      "eval_utility_runtime": 33.9148,
+      "eval_utility_samples_per_second": 1.445,
+      "eval_utility_steps_per_second": 0.383,
+      "step": 2400
+    },
+    {
+      "epoch": 0.8897521272660007,
+      "grad_norm": 0.0,
+      "learning_rate": 7.318220545376842e-07,
+      "loss": 1.4415,
+      "mean_token_accuracy": 0.8040485986240762,
+      "mean_token_accuracy_utility": 0.7683348486087104,
+      "step": 2405
+    },
+    {
+      "epoch": 0.8916019237883833,
+      "grad_norm": 0.0,
+      "learning_rate": 7.077617657387936e-07,
+      "loss": 1.4191,
+      "mean_token_accuracy": 0.799078914922803,
+      "mean_token_accuracy_utility": 0.7771965516507447,
+      "step": 2410
+    },
+    {
+      "epoch": 0.8934517203107658,
+      "grad_norm": 0.0,
+      "learning_rate": 6.840891194872112e-07,
+      "loss": 1.5289,
+      "mean_token_accuracy": 0.7928659930593456,
+      "mean_token_accuracy_utility": 0.7597423881843971,
+      "step": 2415
+    },
+    {
+      "epoch": 0.8953015168331484,
+      "grad_norm": 0.0,
+      "learning_rate": 6.608051033302476e-07,
+      "loss": 1.4214,
+      "mean_token_accuracy": 0.7944005657586646,
+      "mean_token_accuracy_utility": 0.7760974579100629,
+      "step": 2420
+    },
+    {
+      "epoch": 0.8971513133555309,
+      "grad_norm": 0.0,
+      "learning_rate": 6.379106886028086e-07,
+      "loss": 1.3813,
+      "mean_token_accuracy": 0.8118200678932374,
+      "mean_token_accuracy_utility": 0.7759665302515888,
+      "step": 2425
+    },
+    {
+      "epoch": 0.8990011098779135,
+      "grad_norm": 0.0,
+      "learning_rate": 6.154068303868666e-07,
+      "loss": 1.4263,
+      "mean_token_accuracy": 0.8010710072822761,
+      "mean_token_accuracy_utility": 0.7723346737859476,
+      "step": 2430
+    },
+    {
+      "epoch": 0.900850906400296,
+      "grad_norm": 0.0,
+      "learning_rate": 5.932944674716279e-07,
+      "loss": 1.4334,
+      "mean_token_accuracy": 0.8004711418145416,
+      "mean_token_accuracy_utility": 0.7692365927599699,
+      "step": 2435
+    },
+    {
+      "epoch": 0.9027007029226785,
+      "grad_norm": 0.0,
+      "learning_rate": 5.715745223143566e-07,
+      "loss": 1.4949,
+      "mean_token_accuracy": 0.7962964668437011,
+      "mean_token_accuracy_utility": 0.7592583939313767,
+      "step": 2440
+    },
+    {
+      "epoch": 0.904550499445061,
+      "grad_norm": 0.0,
+      "learning_rate": 5.502479010019046e-07,
+      "loss": 1.3367,
+      "mean_token_accuracy": 0.808686246304279,
+      "mean_token_accuracy_utility": 0.7805374741464927,
+      "step": 2445
+    },
+    {
+      "epoch": 0.9064002959674435,
+      "grad_norm": 0.0,
+      "learning_rate": 5.293154932129019e-07,
+      "loss": 1.5471,
+      "mean_token_accuracy": 0.7969484595377728,
+      "mean_token_accuracy_utility": 0.7528586772125215,
+      "step": 2450
+    },
+    {
+      "epoch": 0.9082500924898261,
+      "grad_norm": 0.0,
+      "learning_rate": 5.087781721806539e-07,
+      "loss": 1.4248,
+      "mean_token_accuracy": 0.8104537993295192,
+      "mean_token_accuracy_utility": 0.7679100877865175,
+      "step": 2455
+    },
+    {
+      "epoch": 0.9100998890122086,
+      "grad_norm": 0.0,
+      "learning_rate": 4.886367946567005e-07,
+      "loss": 1.4893,
+      "mean_token_accuracy": 0.7940417511839355,
+      "mean_token_accuracy_utility": 0.7659992447640074,
+      "step": 2460
+    },
+    {
+      "epoch": 0.9119496855345912,
+      "grad_norm": 0.0,
+      "learning_rate": 4.6889220087508514e-07,
+      "loss": 1.4297,
+      "mean_token_accuracy": 0.809281221252329,
+      "mean_token_accuracy_utility": 0.7677439929623963,
+      "step": 2465
+    },
+    {
+      "epoch": 0.9137994820569737,
+      "grad_norm": 0.0,
+      "learning_rate": 4.4954521451729626e-07,
+      "loss": 1.3521,
+      "mean_token_accuracy": 0.8239091021510194,
+      "mean_token_accuracy_utility": 0.7717407232485263,
+      "step": 2470
+    },
+    {
+      "epoch": 0.9156492785793563,
+      "grad_norm": 0.0,
+      "learning_rate": 4.305966426779118e-07,
+      "loss": 1.394,
+      "mean_token_accuracy": 0.7975280785486076,
+      "mean_token_accuracy_utility": 0.7868530155369994,
+      "step": 2475
+    },
+    {
+      "epoch": 0.9174990751017388,
+      "grad_norm": 0.0,
+      "learning_rate": 4.120472758309213e-07,
+      "loss": 1.4491,
+      "mean_token_accuracy": 0.8126811324790821,
+      "mean_token_accuracy_utility": 0.7683515135936111,
+      "step": 2480
+    },
+    {
+      "epoch": 0.9193488716241214,
+      "grad_norm": 0.0,
+      "learning_rate": 3.9389788779675806e-07,
+      "loss": 1.4964,
+      "mean_token_accuracy": 0.790327059856317,
+      "mean_token_accuracy_utility": 0.7654504474489825,
+      "step": 2485
+    },
+    {
+      "epoch": 0.9211986681465039,
+      "grad_norm": 0.0,
+      "learning_rate": 3.761492357100172e-07,
+      "loss": 1.4798,
+      "mean_token_accuracy": 0.7952300339233833,
+      "mean_token_accuracy_utility": 0.765146255027257,
+      "step": 2490
+    },
+    {
+      "epoch": 0.9230484646688865,
+      "grad_norm": 0.0,
+      "learning_rate": 3.588020599878639e-07,
+      "loss": 1.5226,
+      "mean_token_accuracy": 0.7978150700911035,
+      "mean_token_accuracy_utility": 0.756286465819616,
+      "step": 2495
+    },
+    {
+      "epoch": 0.9248982611912689,
+      "grad_norm": 0.0,
+      "learning_rate": 3.41857084299152e-07,
+      "loss": 1.3921,
+      "mean_token_accuracy": 0.8072262584882541,
+      "mean_token_accuracy_utility": 0.7801340084279913,
+      "step": 2500
+    },
+    {
+      "epoch": 0.9248982611912689,
+      "eval_mean_token_accuracy": 0.7893823660325413,
+      "eval_reasoning_loss": 0.6614469885826111,
+      "eval_reasoning_runtime": 90.0652,
+      "eval_reasoning_samples_per_second": 1.432,
+      "eval_reasoning_steps_per_second": 0.366,
+      "step": 2500
+    },
+    {
+      "epoch": 0.9248982611912689,
+      "eval_mean_token_accuracy": 0.7411567941874316,
+      "eval_utility_loss": 0.9462554454803467,
+      "eval_utility_runtime": 33.9102,
+      "eval_utility_samples_per_second": 1.445,
+      "eval_utility_steps_per_second": 0.383,
+      "step": 2500
+    },
+    {
+      "epoch": 0.9267480577136515,
+      "grad_norm": 0.0,
+      "learning_rate": 3.2531501553422884e-07,
+      "loss": 1.4411,
+      "mean_token_accuracy": 0.8082072529532983,
+      "mean_token_accuracy_utility": 0.761644401497436,
+      "step": 2505
+    },
+    {
+      "epoch": 0.928597854236034,
+      "grad_norm": 0.0,
+      "learning_rate": 3.0917654377545567e-07,
+      "loss": 1.4642,
+      "mean_token_accuracy": 0.8061030064875346,
+      "mean_token_accuracy_utility": 0.7617885364872369,
+      "step": 2510
+    },
+    {
+      "epoch": 0.9304476507584166,
+      "grad_norm": 0.0,
+      "learning_rate": 2.9344234226840964e-07,
+      "loss": 1.4748,
+      "mean_token_accuracy": 0.7975434858176673,
+      "mean_token_accuracy_utility": 0.7662827733701569,
+      "step": 2515
+    },
+    {
+      "epoch": 0.9322974472807991,
+      "grad_norm": 0.0,
+      "learning_rate": 2.781130673938037e-07,
+      "loss": 1.4873,
+      "mean_token_accuracy": 0.7997350924839894,
+      "mean_token_accuracy_utility": 0.7601820731966786,
+      "step": 2520
+    },
+    {
+      "epoch": 0.9341472438031817,
+      "grad_norm": 0.0,
+      "learning_rate": 2.6318935864010133e-07,
+      "loss": 1.4349,
+      "mean_token_accuracy": 0.7972682727204423,
+      "mean_token_accuracy_utility": 0.7724672682806689,
+      "step": 2525
+    },
+    {
+      "epoch": 0.9359970403255642,
+      "grad_norm": 0.0,
+      "learning_rate": 2.4867183857684385e-07,
+      "loss": 1.3681,
+      "mean_token_accuracy": 0.8073082820498328,
+      "mean_token_accuracy_utility": 0.7776397534922881,
+      "step": 2530
+    },
+    {
+      "epoch": 0.9378468368479467,
+      "grad_norm": 0.0,
+      "learning_rate": 2.3456111282867178e-07,
+      "loss": 1.4856,
+      "mean_token_accuracy": 0.8022881230128192,
+      "mean_token_accuracy_utility": 0.7578028423606742,
+      "step": 2535
+    },
+    {
+      "epoch": 0.9396966333703293,
+      "grad_norm": 0.0,
+      "learning_rate": 2.2085777005006692e-07,
+      "loss": 1.4316,
+      "mean_token_accuracy": 0.8027777079919736,
+      "mean_token_accuracy_utility": 0.7739825876666385,
+      "step": 2540
+    },
+    {
+      "epoch": 0.9415464298927118,
+      "grad_norm": 0.0,
+      "learning_rate": 2.0756238190078991e-07,
+      "loss": 1.3941,
+      "mean_token_accuracy": 0.8051378984716665,
+      "mean_token_accuracy_utility": 0.7762998827492094,
+      "step": 2545
+    },
+    {
+      "epoch": 0.9433962264150944,
+      "grad_norm": 0.0,
+      "learning_rate": 1.946755030220393e-07,
+      "loss": 1.4994,
+      "mean_token_accuracy": 0.786562207783354,
+      "mean_token_accuracy_utility": 0.768073294348075,
+      "step": 2550
+    },
+    {
+      "epoch": 0.9452460229374768,
+      "grad_norm": 0.0,
+      "learning_rate": 1.8219767101330442e-07,
+      "loss": 1.4662,
+      "mean_token_accuracy": 0.8038949260606376,
+      "mean_token_accuracy_utility": 0.7612681316227898,
+      "step": 2555
+    },
+    {
+      "epoch": 0.9470958194598594,
+      "grad_norm": 0.0,
+      "learning_rate": 1.7012940640995012e-07,
+      "loss": 1.4218,
+      "mean_token_accuracy": 0.8002563695046311,
+      "mean_token_accuracy_utility": 0.7737219821776355,
+      "step": 2560
+    },
+    {
+      "epoch": 0.9489456159822419,
+      "grad_norm": 0.0,
+      "learning_rate": 1.5847121266148847e-07,
+      "loss": 1.401,
+      "mean_token_accuracy": 0.7980697668097291,
+      "mean_token_accuracy_utility": 0.7789439333914348,
+      "step": 2565
+    },
+    {
+      "epoch": 0.9507954125046245,
+      "grad_norm": 0.0,
+      "learning_rate": 1.472235761105878e-07,
+      "loss": 1.4313,
+      "mean_token_accuracy": 0.8166805484403357,
+      "mean_token_accuracy_utility": 0.7593327658386856,
+      "step": 2570
+    },
+    {
+      "epoch": 0.952645209027007,
+      "grad_norm": 0.0,
+      "learning_rate": 1.3638696597277678e-07,
+      "loss": 1.4705,
+      "mean_token_accuracy": 0.7967654267292792,
+      "mean_token_accuracy_utility": 0.7663863362068684,
+      "step": 2575
+    },
+    {
+      "epoch": 0.9544950055493896,
+      "grad_norm": 0.0,
+      "learning_rate": 1.2596183431687336e-07,
+      "loss": 1.4908,
+      "mean_token_accuracy": 0.8012718460080869,
+      "mean_token_accuracy_utility": 0.7613880863078661,
+      "step": 2580
+    },
+    {
+      "epoch": 0.9563448020717721,
+      "grad_norm": 0.0,
+      "learning_rate": 1.159486160461265e-07,
+      "loss": 1.4154,
+      "mean_token_accuracy": 0.7973637050759855,
+      "mean_token_accuracy_utility": 0.7785129116526914,
+      "step": 2585
+    },
+    {
+      "epoch": 0.9581945985941547,
+      "grad_norm": 0.0,
+      "learning_rate": 1.0634772888007073e-07,
+      "loss": 1.4296,
+      "mean_token_accuracy": 0.8043098261502681,
+      "mean_token_accuracy_utility": 0.7693451545089101,
+      "step": 2590
+    },
+    {
+      "epoch": 0.9600443951165372,
+      "grad_norm": 0.0,
+      "learning_rate": 9.715957333710447e-08,
+      "loss": 1.3414,
+      "mean_token_accuracy": 0.8073507813066053,
+      "mean_token_accuracy_utility": 0.7843060859791887,
+      "step": 2595
+    },
+    {
+      "epoch": 0.9618941916389198,
+      "grad_norm": 0.0,
+      "learning_rate": 8.838453271777347e-08,
+      "loss": 1.4408,
+      "mean_token_accuracy": 0.8093079518121475,
+      "mean_token_accuracy_utility": 0.7607055189483243,
+      "step": 2600
+    },
+    {
+      "epoch": 0.9618941916389198,
+      "eval_mean_token_accuracy": 0.789397195128951,
+      "eval_reasoning_loss": 0.6614738702774048,
+      "eval_reasoning_runtime": 90.1162,
+      "eval_reasoning_samples_per_second": 1.431,
+      "eval_reasoning_steps_per_second": 0.366,
+      "step": 2600
+    },
+    {
+      "epoch": 0.9618941916389198,
+      "eval_mean_token_accuracy": 0.740960111532442,
+      "eval_utility_loss": 0.9462173581123352,
+      "eval_utility_runtime": 33.9511,
+      "eval_utility_samples_per_second": 1.443,
+      "eval_utility_steps_per_second": 0.383,
+      "step": 2600
+    },
+    {
+      "epoch": 0.9637439881613022,
+      "grad_norm": 0.0,
+      "learning_rate": 8.002297308879359e-08,
+      "loss": 1.4346,
+      "mean_token_accuracy": 0.7950495397047203,
+      "mean_token_accuracy_utility": 0.7753533455607997,
+      "step": 2605
+    },
+    {
+      "epoch": 0.9655937846836848,
+      "grad_norm": 0.0,
+      "learning_rate": 7.207524326776628e-08,
+      "loss": 1.4369,
+      "mean_token_accuracy": 0.8057911402502536,
+      "mean_token_accuracy_utility": 0.7653143731379081,
+      "step": 2610
+    },
+    {
+      "epoch": 0.9674435812060673,
+      "grad_norm": 0.0,
+      "learning_rate": 6.454167480863694e-08,
+      "loss": 1.5096,
+      "mean_token_accuracy": 0.7885679017577502,
+      "mean_token_accuracy_utility": 0.7639327434205142,
+      "step": 2615
+    },
+    {
+      "epoch": 0.9692933777284499,
+      "grad_norm": 0.0,
+      "learning_rate": 5.7422581987856e-08,
+      "loss": 1.4533,
+      "mean_token_accuracy": 0.79659056923555,
+      "mean_token_accuracy_utility": 0.7685288185365129,
+      "step": 2620
+    },
+    {
+      "epoch": 0.9711431742508324,
+      "grad_norm": 0.0,
+      "learning_rate": 5.0718261791274924e-08,
+      "loss": 1.4926,
+      "mean_token_accuracy": 0.7860157926512651,
+      "mean_token_accuracy_utility": 0.769098329798789,
+      "step": 2625
+    },
+    {
+      "epoch": 0.9729929707732149,
+      "grad_norm": 0.0,
+      "learning_rate": 4.4428993901753924e-08,
+      "loss": 1.3574,
+      "mean_token_accuracy": 0.8058947839179945,
+      "mean_token_accuracy_utility": 0.7851936394727521,
+      "step": 2630
+    },
+    {
+      "epoch": 0.9748427672955975,
+      "grad_norm": 0.0,
+      "learning_rate": 3.8555040687493494e-08,
+      "loss": 1.4011,
+      "mean_token_accuracy": 0.8074352300054171,
+      "mean_token_accuracy_utility": 0.7748815076401374,
+      "step": 2635
+    },
+    {
+      "epoch": 0.97669256381798,
+      "grad_norm": 0.0,
+      "learning_rate": 3.309664719108874e-08,
+      "loss": 1.4439,
+      "mean_token_accuracy": 0.8136909193156885,
+      "mean_token_accuracy_utility": 0.7614267360088858,
+      "step": 2640
+    },
+    {
+      "epoch": 0.9785423603403626,
+      "grad_norm": 0.0,
+      "learning_rate": 2.805404111931198e-08,
+      "loss": 1.4444,
+      "mean_token_accuracy": 0.800891875313995,
+      "mean_token_accuracy_utility": 0.7681447655296878,
+      "step": 2645
+    },
+    {
+      "epoch": 0.9803921568627451,
+      "grad_norm": 0.0,
+      "learning_rate": 2.3427432833607046e-08,
+      "loss": 1.4013,
+      "mean_token_accuracy": 0.801268748082045,
+      "mean_token_accuracy_utility": 0.7768968324976477,
+      "step": 2650
+    },
+    {
+      "epoch": 0.9822419533851277,
+      "grad_norm": 0.0,
+      "learning_rate": 1.9217015341318478e-08,
+      "loss": 1.4843,
+      "mean_token_accuracy": 0.787426328879073,
+      "mean_token_accuracy_utility": 0.7710254014736793,
+      "step": 2655
+    },
+    {
+      "epoch": 0.9840917499075101,
+      "grad_norm": 0.0,
+      "learning_rate": 1.542296428763801e-08,
+      "loss": 1.4251,
+      "mean_token_accuracy": 0.8083423239431686,
+      "mean_token_accuracy_utility": 0.7621392707194995,
+      "step": 2660
+    },
+    {
+      "epoch": 0.9859415464298927,
+      "grad_norm": 0.0,
+      "learning_rate": 1.2045437948275952e-08,
+      "loss": 1.5416,
+      "mean_token_accuracy": 0.7942349465947605,
+      "mean_token_accuracy_utility": 0.7522429829508346,
+      "step": 2665
+    },
+    {
+      "epoch": 0.9877913429522752,
+      "grad_norm": 0.0,
+      "learning_rate": 9.084577222860935e-09,
+      "loss": 1.4033,
+      "mean_token_accuracy": 0.8088550342116608,
+      "mean_token_accuracy_utility": 0.7717437822429438,
+      "step": 2670
+    },
+    {
+      "epoch": 0.9896411394746578,
+      "grad_norm": 0.0,
+      "learning_rate": 6.540505629061278e-09,
+      "loss": 1.4164,
+      "mean_token_accuracy": 0.7991348502522018,
+      "mean_token_accuracy_utility": 0.7763252465178438,
+      "step": 2675
+    },
+    {
+      "epoch": 0.9914909359970403,
+      "grad_norm": 0.0,
+      "learning_rate": 4.413329297430213e-09,
+      "loss": 1.4364,
+      "mean_token_accuracy": 0.8018239075254682,
+      "mean_token_accuracy_utility": 0.7691227825323005,
+      "step": 2680
+    },
+    {
+      "epoch": 0.9933407325194229,
+      "grad_norm": 0.0,
+      "learning_rate": 2.7031369669816566e-09,
+      "loss": 1.4808,
+      "mean_token_accuracy": 0.7831432182568145,
+      "mean_token_accuracy_utility": 0.774530497096497,
+      "step": 2685
+    },
+    {
+      "epoch": 0.9951905290418054,
+      "grad_norm": 0.0,
+      "learning_rate": 1.4099999814853883e-09,
+      "loss": 1.5085,
+      "mean_token_accuracy": 0.7978510432882366,
+      "mean_token_accuracy_utility": 0.7588601879743533,
+      "step": 2690
+    },
+    {
+      "epoch": 0.997040325564188,
+      "grad_norm": 0.0,
+      "learning_rate": 5.339722864927677e-10,
+      "loss": 1.4677,
+      "mean_token_accuracy": 0.7947258221332009,
+      "mean_token_accuracy_utility": 0.7671935837797715,
+      "step": 2695
+    },
+    {
+      "epoch": 0.9988901220865705,
+      "grad_norm": 0.0,
+      "learning_rate": 7.509042708631065e-11,
+      "loss": 1.501,
+      "mean_token_accuracy": 0.7974223863256141,
+      "mean_token_accuracy_utility": 0.759458093160426,
+      "step": 2700
+    },
+    {
+      "epoch": 0.9988901220865705,
+      "eval_mean_token_accuracy": 0.7896325177377955,
+      "eval_reasoning_loss": 0.6613715291023254,
+      "eval_reasoning_runtime": 91.1119,
+      "eval_reasoning_samples_per_second": 1.416,
+      "eval_reasoning_steps_per_second": 0.362,
+      "step": 2700
+    },
+    {
+      "epoch": 0.9988901220865705,
+      "eval_mean_token_accuracy": 0.7410466212629234,
+      "eval_utility_loss": 0.9463363885879517,
+      "eval_utility_runtime": 34.3188,
+      "eval_utility_samples_per_second": 1.428,
+      "eval_utility_steps_per_second": 0.379,
+      "step": 2700
+    },
+    {
+      "epoch": 1.0,
+      "mean_token_accuracy": 0.7960500389995716,
+      "mean_token_accuracy_utility": 0.786398114254285,
+      "step": 2703,
+      "total_flos": 0.0,
+      "train_loss": 1.5309620150538934,
+      "train_runtime": 111932.1245,
+      "train_samples_per_second": 0.193,
+      "train_steps_per_second": 0.024
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 2703,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": false,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 0.0,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}